Hacker News new | past | comments | ask | show | jobs | submit login
Customers don't want chat bots (creativegood.com)
366 points by blueridge on July 7, 2023 | hide | past | favorite | 265 comments



A big part of the problem isn’t actually chat bots even though it may well seem to be.

It’s that you can’t pop some system atop a teetering ziggurat of shitty data from mutually incompatible systems and think the fundamental problems are going to just disappear.

Example: I am currently executor of my dad’s estate and trying to get various policies and pensions and whatnot transferred into my mother’s name. One company seemed especially clueless when I emailed them so I called. Part of the voice prompt on the phone system was something like “if the statement we send you comes from city A, press 1, if it comes from city B press 2”. Well that’s a weird red flag right there also a teeny problem: my statement comes from city C. I eventually speak to a human but try as they might they can’t find anything about my dad.

Eventually I write to the company at city C and they’re super helpful. Basically this company is formed from the corpses of various pension funds which just get welded together. There is no underlying integration at all between the systems. They are separate companies in all but name.

You could have an agi chatbot and its still going to be lobotomised by this level of disfunction. In my experience the data landscape at many enterprises is like this.


You have my condolences. My brother and I ended up hiring a lawyer to handle our dad's estate, and I'd recommend it to anyone. She did things like send the hospital a letter that said in essence "give me a single bill with a total for all services provided at your hospital. If it's not on this bill I'm not going to pay it."

(Although legal fees for estates are capped here in Ohio, not sure if that's true where you are.)


> She did things like send the hospital a letter that said in essence "give me a single bill with a total for all services provided at your hospital. If it's not on this bill I'm not going to pay it."

That is a great thing to say, but does it actually hold any weight? Or is it like the construction trucks that have signs saying they are not responsible for damage from items blown off them?


Well, it worked in the case of my dad's estate.

I think the fact that it was from a lawyer made a difference. I had a different situation with Wells Fargo where I quoted their own document back to them and they still denied me. When I got a lawyer to send essentially the same letter, but with his letterhead, they suddenly bent over backwards to help me out, going above and beyond their obligations.


That's because most companies operate their customer support at extremely basic level and hope that those with complex needs will simply give up in frustration.

A lawyer letterhead short-circuits that because companies fear that someone who is actually willing to pay upfront for a lawyer is also going to follow through with a lawsuit or, worse, a complaint to the regulatory agency so they'll kiss asses beyond belief just to prevent a case from ever reaching public record and a resulting bad image in the eyes of the regulator who can actually impose pretty expensive fines.


Yeah, absolutely. In the Wells Fargo case, thousands of dollars were on the line, so it was absolutely worth the couple hundred dollars for the lawyer. But if I had been disputing a $50 overdraft fee or something like that, I'm not sure what I would have done.

(I might contact the CFPB these days, but I think this incident was before they existed.)


The CFPB is completely useless. I've never received a worthwhile response to any complaint I submitted through them, out of the ten or so complaints I've submitted to them in the last ~decade.

They simply forward your complaint to the same barely-literate customer "service" department that created the problem in the first place, so that someone who doesn't understand the problem can cut-and-paste their original response back to you again.

You're given an opportunity to respond after that. You get no response if you explain why the bank's reply wasn't responsive to your complaint. I also once tried cutting-and-pasting several paragraphs of hard-core gay erotic Star Trek fan-fiction (searching-and-replacing one of the main characters' names for that of the bank's CEO, with a note that said, "this is to test if anyone actually reads this"), hoping to get the bank or CFPB to even acknowledge me in a non-automated fashion. I got no indication that they did; not even a "please don't do that again." (I wasn't bold enough to try death threats or child porn, but I'd wager even those would be ignored.)

I wrote my senators and representative in Congress, asking them why the CFPB seemingly ignored complaints. One of them CC'd me on a letter to the CFPB requesting someone contact me. Noone ever did. As far as I can tell, the entire agency is just a black hole for wasting tax dollars, likely a graft to help ex-bank employees collect Federal paychecks and pensions while doing nothing and claiming to be working on "reform."


> I also once tried cutting-and-pasting several paragraphs of hard-core gay erotic Star Trek fan-fiction

Ballsy move


“Do you always fly at the ~~ladies~~ guys at Warp 10, Ensign?”


> I also once tried cutting-and-pasting several paragraphs of hard-core gay erotic Star Trek fan-fiction

holy shit lmao


The "weight" in the situation is good ol' fashioned incentive. The patient in question is dead, and the estate attorney is wrapping up business. After a certain date, there won't be an entity to collect from. Relatives of the deceased are not liable for medical debts (that doesn't mean the medical provider won't try to collect it from relatives, though).

So the medical provider faces two distinct choices:

A.) Get their act together and get paid a guaranteed sum from an estate that has been established by a professional entity

B.) Hem and haw, miss the boat, and try to collect from an estate that's run dry because the money has been dispersed to relatives and/or try to goad relatives into paying. Sometimes, those relatives don't exist.


Not really, same idea. Hospitals want to recover as much as they can with as little headache. Coming from a lawyer vs trying to negotiate yourself just means "I can actually make this difficult for you, play ball"


In the case of a death or bankruptcy, yes it does work. There is a hard cutoff date and if they miss it the debt is uncollectible.

It does not work for the living.


I should have been more specific, or understood the question more generally as "can you actually negotiate with the hospital on your bills?"


You can do this, even as a living person. I once got a $10k bill reduced to $700. (It took a lot of back-and-forth, but it was worth it.)


We're about to find out too. Dad passed 8 months ago; we got a $1500 hospital bill last week (just after the probate window closed) w/o explanation or description of services rendered.


That sounds to me like the hospital screwed up and is asking for charity. If the probate window is closed before they sent the bill (check the postmark date), then I don't think you owe them anything.


The billing department can call your bluff, then get in line in probate court to get paid. Possibly years down the road depending on how long it takes. Keep in mind hospital bills are just a fantasy, I've received bills >$50,000 US for less than 1 hour in the hospital.


> Keep in mind hospital bills are just a fantasy, I've received bills >$50,000 US for less than 1 hour in the hospital.

Then it makes even more sense to do that.


Seems plausible to me, during bankruptcy or winding down of a company the same happens. If you miss your chance to join the list of creditors then the money has already been disbursed and there's nobody responsible for making you whole. Of course there are a lot of rules and details. A death is like a bankruptcy in that assets are used to pay for debts and remaining debts don't get inherited.


In the US that sounds roughly consistent with part of the probate process, which is the way that estates are closed out.

More or less you publish a letter in a journal of record saying legal boilerplate to the effect of 'so and so has died, if they owed you money now is the time to bring it up'. Any claims on the estate not brought up timely will more or less be extinguished, the estate is eventually closed, and life goes on.


It’s nice to know bankruptcy at least works when I die


I've done similar to this and it has worked. You dump the onus onto them. If they don't explicitly reject it then they've tacitly accepted it, and you've got it in writing too.


I wish I could pass on some wisdom or tips for how to deal with estate crap better, but I don't know that there is. It's just a fucking slog. The best cases are where you say the magic word, it gets you to a dedicated department staffed by real humans who actually know what they're talking about, sends you the right paperwork, and that paperwork finishes out what you need from them. The bad cases are basically an exercise in reverse engineering their org chart, to figure out what department is actually responsible for a given corner case, and how their policies and the legal regulations actually fit together. I remember being on the phone with Vanguard to sort something out, I think I got pinballed between 4-5 different ostensibly-human reps in changing departments. Each one profusely apologizing that they had to transfer me again, and I'm just laughing it off because it's the only way to survive.

Get a durable power of attorney for your mother sooner rather than later. It won't solve everything, but can make some inquiries easier to just do on your own rather than needing a group project. Also health care proxy and a new will if those are necessary. IME old people can be very open to such changes soon after a death, but then become much more hesitant and procrastinatey months later. I think it's the difference between ascribing the reason to their partner's death, or to their own looming death. Strike while the iron is hot.


>> and think the fundamental problems are going to just disappear.

Au contraire. If you implement things poorly enough then all those people on the phone asking for tech support really do just go away. They learn to solve their own problems. Make your problem-solving/help-line interface sufficiently horrible that it develops a reputation (Google) and they won't even make an attempt to contact you.


Sometimes I think this is the point.

Like lately out cell service has been horrible at the house and my wife keeps asking me to call them but why? It's just going to be screaming into the void of some automated system and nothing will be done.


Call to cancel your account, mention moving to another service, etc. Often times companies will transfer you to more empowered "retention specialists".


Recently, I noticed my broadband fees were double what the company would charge a new customer at the same address. I reached out to the telecom and stated with the “I’d like to close my service” track. That got me a human but when it was clear that my issue was price they had a script for that. It was full of delays and circular “offers”. After a hour it finally end with their final offer - split the difference between what I am paying and the new customer price. What is amazing to me is that another competitor offers the same new customer price they do and it all uses the same infrastructure. It is literally a terminal entry switching process. Apparently they have calculated they can keep more a actually lose very few customers.


> teetering ziggurat

My next internal project codename


Are you working with web frameworks?


s/with/against!


> It’s that you can’t pop some system atop a teetering ziggurat of shitty data from mutually incompatible systems and think the fundamental problems are going to just disappear.

I mean, works for CSS.


Where AGI is still the realm of science fiction, I think an AGI could recognize that <CompanyName, City> is a unique entity, and that CompanyName isn't the same from city to city.


That’s not the problem here unfortunately. The problem here is that there is no interconnection of any kind between any system that knows about city A and B and the systems which know about city C. So there’s literally no way to help a customer who has a problem described in those systems. An AGI chatbot couldn’t be any more helpful than the human was because it has no access to anything about city C.

Now could that be fixed? From a tech perspective of course it could. However there’s probably a deep organizational reason this hasn’t happened. In this case, it’s because city C is a bunch of customers they bought during the windup of a bankrupt life insurance company and are running them down over time. So there probably isn’t enough profit in the portfolio they are running down to justify the IT investment.

As I say, these are the kind of things a chatbot (however smart) isn’t going to be able to fix.


Right, after mashing the companies together one would need to integrate systems - but that costs money! Money that could be used for executive pay - which they "earned" because of the "successful" merger.


> there is no interconnection of any kind between any system that knows about city A and B and the systems which know about city C

The AGI, via DuplexAI, could just call up the office in city C and ask. It's been calling restaurants to get their hours for years. Having it call a listed phone number to ask a person about an account doesn't seem far fetched at all to me.


The person at the original call center literally had no idea of the existence of the operation of the company at city C and it didn't have any sort of online presence. The online presence only knew about the operations at city A and B. Remember that C was the zombie part of a defunct company they bought so even the original contact details which might once have existed wouldn't work or be listed any more. The call center person I originally spoke to literally tried to google it while I was on the phone and got an incorrect contact number which I tried so I know it didn't work. The only correct contact details I could find were on the paper statement which was how I knew to write them an old fashioned letter (which was the thing which eventually seems to have worked). Noone at city A or B had this statement - I had it.

In any event we're in danger of fixating on my original example and missing the broader point which is that any system or intelligence, human or otherwise, is limited by the data it is able to access. If a company has a very messy and/or disfunctional data landscape[1] then it is difficult bordering on impossible to paper over the cracks with systems or operational processes.

[1] which I gave an intentionally extreme example of, but lots of companies are formed from multiple acquisitions or have systems built by rival IT groups in different parts of the org or by outside vendors or contractors etc so a milder form of this is very common in my experience.


fair point. or otherwise is as much as I was getting at. mail scanning services and text to speech gateways extend the reach of a hypothetical AGI. we don't need to wait for ARA (artificial robot actuation) for the skynetpocalypse to come. plenty of building blocks are already here.


I think humans already struggle with this, without sufficient background knowledge and context. Reality is messy.


But GPT-4 100% could have looked that up and given you that information. It's clearly not a unique situation.


Typically chatbots that I interact with are just attempting to steer me back towards FAQs. I don’t need a different interface to search your documentation in an attempt to keep me on the cheap customer service path and away from the expensive customer service path.

Which isn’t to say they can’t be helpful. I’d much rather chat with a bot than call a person, if the bot is capable of doing the things I need. The thing is nobody seems willing to let the system actually do anything except escalate to a person, so why bother?


Yeah this is my take too. Chat bots are usually transparently obviously something the CTO bought, then Support dumped some basic configuration into it, and now it's a glorified redirect-provider to poorly map english questions to help articles. There's 0 vertical integration and it's purely to try to shunt traffic down cheaper support channels.

I've literally had 1 good experience with bot-based support. Exactly one. It was when I forgot I had used my Lowes' store card that month and they charged me a late fee for missing a payment. I went through the chatbot UI because I couldn't figure out how to even begin to contact human support.

And can you imagine my utter shock and disbelief when the bot offered to waive the late fee if I set up autopay? Holy shit! I'm still in awe. The chatbot waived my fee. It actually DID something. It genuinely saved them and me a support call that day instead of just feeling like they were trying to persuade me to give up by throwing friction into the process of getting to someone who could actually waive my fee.


I agree. In my head I group chatbots into: 1. bot that can regurgitate the faqs 2. bot that can perform actions

I've had good experiences with type 2 bots with Amazon, Uber, and Doordash, refunding orders that weren't delivered, etc. I assume they have a budget for refunds based on your purchase and refund volume, very similar to the agency a normal customer service agent would have.


> very similar to the agency a normal customer service agent would have.

Exactly this is the differentiating factor. Even if you have a human on the other side, if they don't have the agency accorded to them to perform a particular action within given bounds, they are no good compared to a Chatbot. A chatbot will help such companies cut costs, but they'll also reduce the CX - but I doubt those companies give a damn about that.


> And can you imagine my utter shock and disbelief when the bot offered to waive the late fee if I set up autopay? Holy shit! I'm still in awe. The chatbot waived my fee. It actually DID something. It genuinely saved them and me a support call that day instead of just feeling like they were trying to persuade me to give up by throwing friction into the process of getting to someone who could actually waive my fee.

I think I had a similar experience with the Amazon customer support bot just yesterday. It authorized a refund on a $20 item that I said arrived damaged.


That is incredibly impressive. To people who are not technical, I wonder if this is impressive to them.


> The thing is nobody seems willing to let the system actually do anything except escalate to a person, so why bother?

Because you are not the target user for the bot. For every user that proactively reads the FAQs, there's 10 more that don't and would happily open a support ticket to have someone copy/paste the FAQ answer to them. The bot is to there deflect those cases.


I don't think this is true, because if it were true, people would want chatbots. The last thing most people want to do is to call or open a chat with customer service, and they've usually looked through everything they can find to avoid it.

One of the reasons some chatbots, like at Amazon, could be useful is because their FAQs are impossible to find.


"People" aren't the ones deploying chatbots, companies are, because they successfully deflect cases.

If you think most people have done everything to avoid reaching out to support, I can only imagine you haven't worked in a "Tier 1" support role.


Support Chat: "Thank you for asking about [question], here is a [link to FAQ Page!"

The FAQ Page: "Was this article helpful? If not, contact our support chat!"

Endlessly. It's so stupid.


That the key point for me. If the chatbot can’t act on the system for me beside « get me human », it’s useless.

I like them when : - it can schedule a call back with a human - it can open ticket, ask for document on my side and make it faster for my query to process overall.


I had a savings account where the interface was basically a chat bot.

It was basically guess the verb. And also not obvious when you're talking to a human through the exact same interface.

It's a menu system but harder to use.


Wow! “Guess the verb.” That is exactly how text based adventure games worked in 80s. The interface of the future is Zork.


Omg the number of times I went into a mental meltdown with text adventures over that.

I finally have the thing I need

I am in the right place to use the thing such that it will fix whatever problem so I can move onwards

How hard can it be to figure out how to ask the game to use the thing to fulfil its purpose?

Quite fucking hard, it turns out. Quite fucking hard.


Infocom had a 80s text adventure game called Bureaucracy[1] that captures this modern scenario exactly.

[1] https://en.wikipedia.org/wiki/Bureaucracy_(video_game)


In those years my English was quite basic. I abandoned an adventure in a mine in front of an elevator door because open and every other verb I could think about didn't work. Very frustrating but I lived on and forgot about it. A few years ago I found a walkthrough on some retrogames site. I checked that location and... slide!


EverQuest did the same thing, except it was "guess the noun".

> What broach?

> What sword?

The keywords could be so obscure that allegedly, 25 years after release, some quests remain uncompleted by even a single player.


I recently had a chat with the Amazon bot for a refund of something that had not been delivered. That went very well!


I had the same, it refunded me, then 3 weeks later I got a notification from Amazon that they would be taking the money off me.

I complained and was ignored. It was only $10 but it teaches a lesson.


I'm very surprised. My experience with Amazon refunds and in general customer support is that they care more about upsetting a customer than fighting for pesky 10 bucks.

Could it be that their behaviour is different depending on who is the customer, i.e. what is the customers yearly spend?

I shop frequently on Amazon. Last year I sent a gift to my nephew how was living in a dorm at the University. The gift was a comouter monitor about $500 worth. The package disappeared soon after it got signed by someone that was at the front desk and never made it to my nephew.

I was already getting ready for a long discussion but Amazon said they will handle it and my nephew got another monitor two days after.

I also routinely return stuff I buy by mistake and I honestly report I just bought it by mistake


So was I, the first few times this happened.

My experience is that Amazon customer service did a complete 180 in 2020. Prior to that, every issue was resolved to my satisfaction, and they felt too generous, if anything.

Today, every Amazon rep I speak to isn't even empowered to resolve the most basic issues, and sounds like they hold their employer in complete contempt. Which is to say, they follow the scripts and /try/ to sound friendly, but it's clear they hate their employer, and want to satisfice their KPIs to keep their job, but all else being equal, would prefer for Amazon to die.

This also corresponds to a huge increase in problem rates. Prior to 2020, problems with Amazon were rare. Today, it seems like screwups happen all the time. I don't think they show up on any KPIs, though, as from Amazon's perspective, if a customer service rep e.g. satisficied and pretended to issue a refund without actually issuing one, or a customer gave up on resolving an issue, they have no visibility into the issue.

A basic, obvious issue like a missing package or a return will get handled, but anything more complex is now a black hole.


I can't say I had a bad experience with my Amazon interactions.

I have 3 free UPS's, that were wrong and Amazon just refunded me without requesting a return.

And I even had a hard drive go back, because of degraded performance - but it did require some extra troubleshooting on the phone.

Granted, I live in a rural area and I spend a lot on Amazon with Prime - from cleaning supplies and food to furniture and tech.


> UPS's, that were wrong and Amazon just refunded me without requesting a return.

That's because they contain batteries that Amazon will not accept.


Magnets are the same. I ended up with an extra subwoofer that way once.


In 2018 I ordered something from Amazon and it got marked as delivered but didn't make its way to me. Amazon sent a replacement AND refunded me for the item.

In 2022 I ordered an item of similar size and cost and it was marked as delivered along with a photo of a house I could not recognize at all. When I asked Amazon to look into it I got the corporate flow-chart equivalent of "Oh haha, that sucks". I asked for it to be escalated and they more or less said that the order already has the "oh haha, that sucks" flag on it and can't be changed.


Yeah, their delivery accuracy has dropped. Since the great yuck I have had at least one delivery a year next door (easily retrieved) and one I completely didn't recognize.


> I'm very surprised. My experience with Amazon refunds and in general customer support is that they care more about upsetting a customer than fighting for pesky 10 bucks.

I'm not surprised at all. There have only been two times when I needed to return an item I got from Amazon, and both times Amazon stiffed me. One of those items was for around $10.

Amazon seems to care about petty dollar amounts more than avoiding upset customers.


Two weeks before leaving home for six months, I bought two suitcases. They were delayed/lost and Amazon suggested I reorder.

I reordered and they still were going to be delayed until the day after we left.

I went to a local store and bought one.

I started a chat and told them that even if the suitcases did arrive that evening, I wouldn’t be able to send them back for a refund.

They said keep all of the suitcases and I would still get my refund.

They did come later that evening.


The stuff Amazon gets away with is incredible. Imagine if you returned a product to WalMart and they decided to charge you for it anyways 3 weeks later.


It’s always been possible to get refunds from Amazon without contacting a human, through a very simple form. I haven’t used their chatbot, but I’m curious how it’s better than the old form.


Some forms were replaced with bots. For me is the "Request invoice from vendor". I used to have a canned email I would copy paste into their contact form but now I have to answer a few questions the bot does so it is three or four copy/pastes instead of one. Just an annoyance but still, not progress.


This is my experience as well. Almost anything that the chatbots do competently shouldn't require more than a form on their site with a few questions and often did in the past.


I like the vision our chatbot team at work has there. They want their bot to have 2, maybe three tasks. First, it's supposed to be a neater way of presenting that initial phone choice of "Press 1 if you have an internet issue. Press 2 if you have a telephone issue". It's easier for some people to say "My phone is dead and has no free tone when picked up" and have the system recognize this as a phone issue. And second, it's supposed to make the collection of necessary data about the call smoother - since when does the issue occur, which phone number is affected, what's your account number. Once it has those (or it gives up after 1-2 tries without progress) it hands over to an actual agent.

The value proposition there is that you save a few human employee minutes per support case.


Unfortunately, the majority of times I call customer service, none of this collected information gets transferred or they make me confirm everything again. It's extremely aggravating. For every company who gets this right, there are ten who don't.

Even if you do everything perfectly for every customer, I lump you in with the other ten because I think I got lucky rather than thinking you having a working system. :(


> Once it has those (or it gives up after 1-2 tries without progress) it hands over to an actual agent.

And that human will immediately ask you to repeat all of the information you just gave to the screening bot.


Save a few minutes of paying someone, piss off your paying customers


It's actually more than a few minutes. I can't speak for other places, but generally there will be an agent that is monitoring many inbound chats, and in a lot of cases the resolution can likewise be automated.

In the past this was managed by a fairly deep call queue, and has saved fairly significant amount of $ vs receiving and resolving everything manually.

Indeed we have internal clients (vs paying customers) but overall this has been quite successful for us.


Its not a bad idea, but its like phone trees: abused by so many places the only thing I can do is try to get a human.


Yup. The chatbots are basically a friendly version of search, useful for the non-tech but anyone here will almost certainly not benefit.

You want to cut down your customer service costs, actually provide answers online! I would say at least 80% of what I call companies about are things that perfectly well could have been done online if their website actually allowed it.


I don't think they're meant to make end user lives easier, but to frustrate them into giving up.


Exactly, yes. I never use the chat support (whether robotic or manned) because if I'm at the point where that's necessary, I'm at the point where I need to talk to a human being personally.


Let me walk you through the steps of restarting your modem....


I dont want useless customer support. Wether that's a human or a bot.

It's extremely frustrating to have a real issue, take for instance a plane cancelled or delayed when travelling with kids, and not being able to talk to a real person who can actually help. Let alone having to wait 30-60 minutes to get a hold of someone.

Nowadays most large organisations are structured now that it's impossible to have a conversation with a human who actually has the power to make a decision. All they repeat is info that's found online and tell you to fill in a form or email.

In that case a chatbot is probably better for both sides, mainly for the person behind the phone who doesn't get verbally abused on a regular basis.

Sadly best way to get helped by corps nowadays is to tweet to them.


> Nowadays most large organisations are structured now that it's impossible to have a conversation with a human who actually has the power to make a decision.

In my experience it's equally impossible to have a conversation with a human that has a 1st grade level set of communication skills. You write out a detailed account of your problem and then get some copy/paste response back that has nothing to do with your complaint, and suggests that the person did not even bother to skim your complaint, let alone try and comprehend it.


It's probably not communication skills but workload. Their daily pay may depend on how many people they knock off, and an impulsive, quick response might work 90% of the time. If you're one of the 10%, they'll finally read what you wrote instead of skimming it quickly, hitting F6, and selecting the third option.


That has a lot more to do with the requirements laid out by their management than their communication skills. They stick to the script because they take performance hits if they don't, not because they don't know how to talk to a human.


Yeah talking to stripe customer support is like talking to gpt. Have to remind them every now and then to go back and reread the conversation.


When you outsource your customer support to a call centre where the people there have no power whatsoever to actually help you, who are just trained to get you off the call with a "if there's anything else we can help you with...", why bother? Support needs to be in-house, and the people doing it need to have the power to actually do something when you call them. Otherwise, it's just for show, it's ticking a box in some compliance or PR checklist


I agree. I've run a support team for a mid-sized SaaS company for many years and have so far resisted any outsourcing. Unsurprisingly, Support is one of the commonly cited high points of our product. The members of my team are well trained and knowledgeable and can escalate things easily so that the vast majority of customer problems can be resolved satisfactorily.

The challenge with resisting outsourcing is justifying the cost of it. I try and frame it as: what is the likelihood that a customer churns if their problem goes unresolved?

Of course in our case, each customer represents thousands of dollars in ARR, so that's an easier argument to make.


Same.

Few years ago, I was stuck at Dulles due to a snow storm and had to resort to online customer chat support via the United app. I only used it because United was pushing people really hard to use the app for support. I was connected with a support rep who obviously did not live in the states because he tried to give me an impossible itinerary. Anyone living in DMV area knows taking an AA flight from DCA that's about to take off in 40 minutes when you're currently located in terminal C/D in IAD is impossible.

My current gig requires me to travel more and now I am able to enjoy the privileges of dedicated support line for 1K. The difference is night and day; you not only get connected to reps who are obviously American, but also reps who deal with frequent travelers so they have contextual information to get what you need.


> Anyone living in DMV area knows taking an AA flight from DCA that's about to take off in 40 minutes when you're currently located in terminal C/D in IAD is impossible.

I don’t think you need to live in the DMV area or even need to be a human to figure this out…


You'd be surprised!

My thought process is that people who get contracted/hired to do cheap labor won't appreciate how big the US is (IAD to DCA is 40 minutes alone without inclement weather + bad traffic) and how TSA affects one's transit time.


But there are easy resources available to them to determine that. Google Maps alone would indicate that it’s pretty much impossible. I am actually surprised that their system even would suggest something like that.

For fun, I asked ChatGPT (gpt-4)…

Prompt: Can I get from Terminal C in IAD to DCA in 40 minutes?

Response: Washington Dulles International Airport (IAD) and Ronald Reagan Washington National Airport (DCA) are two separate airports located in the Northern Virginia area, near Washington, D.C.

The travel time between these two airports is approximately 45 minutes to 1 hour by car under normal conditions. However, this can vary depending on the time of day, traffic, and other factors. Furthermore, this timing does not include time for potential issues such as returning a rental car, waiting for a taxi or rideshare, airport security, and other possible delays.

Therefore, a 40-minute window to travel between these two airports and catch a flight is not realistic and highly risky. It's recommended to allow a much greater time buffer when scheduling flights with a transfer between these airports.


I'm sure they could if they really wanted to.

But for farm support center, I'd wager that they're just given SOPs to follow mindlessly or don't really care that much.


I figured out most of the airport acronyms... except DMV... Delaware-Maryland-Virginia? (Search focuses on motor vehicles, even with "-motor")


I always thought it stood for "DC Maryland Virginia".. basically the greater Washington DC metro area.


Yep. Poor Delaware.


Yes, also known as Delmarva.


Typically I call customer service because I'm trying to tackle a problem that cannot be solved by using their normal interface. There is usually some subtlety or exception that requires an agent capable of understanding them at the other end, otherwise I wouldn't have bothered calling.

Chatbots almost by definition cannot deal with that, at least I've never experienced one ever. And I imagine that if they can understand what I want it's just too easy to code them to toe the company line and tell me to pound sand.

Empathy and understanding are not parameters, yet.


Right. You contact a human because you want a guarded administrative action to be taken.

For all the impressiveness of these chat bots, I think people are still skeptical and hesitant to give it levers that are actually connected to things.

It's hard to hold a computer program accountable in any nuanced way when it starts misbehaving other than kicking it to the curb


> For all the impressiveness of these chat bots, I think people are still skeptical and hesitant to give it levers that are actually connected to things.

They don't want a person to escalate you either, but you can pressure a person in ways that a chatbot cannot be pressured. At least a person (and their employer) wants to get you off the phone; the person because they're sick of being yelled at and coming up with reasons not to escalate, and their employer because they're paying for the time spent.

A chatbot is happy to talk you in circles all day, and costs near nothing. Empowering chatbots is not the point, one of the advantages of chatbots is that they can be disempowered more easily than humans.


“At least a person (and their employer) wants to get you off the phone; the person because they're sick of being yelled at and coming up with reasons not to escalate, and their employer because they're paying for the time spent.”

Click. Disconnect. Easy. They just disconnect the call when a customer is frustrating them and oftentimes disconnects happen when first reaching a person after sitting on hold for a long time. The other tactic is to transfer the customer around the call center into oblivion.


For your average customer support case, there is nobody to hold accountable either when they screw up and cost you time or money. It's just another faceless agent of the company.


So I work in customer service for tech company, I see the tickets and about 60% of them can be solved with a chatbot. Most of the time the instructions were already written on the customers screen and they just didn't bother to read them.

I agree, calling up and getting a bot is irritating, but personally fielding every call without screening out the silly PEBCAC tickets gets expensive quick and hold-times pay the price.


> and they just didn't bother to read them

I would encourage you to reconsider this sentiment, and to read "The Design of Everyday Things".

Sometimes you really do have a situation where the user needs to read and they just don't. But that's less common than them "not reading" because:

* The relevant information is buried in what appears to be bumph that we all sensibly skip over.

* The instructions are actually not clear.

* The user reasonably didn't expect that instructions would even be required so they didn't look for them.

* There was a more obvious action to take on the page than reading text, e.g. "Continue".

Those people making tickets can read. The fact that they aren't suggests there's something wrong with the instructions or how you are presenting them - or more likely, the fact that they exist at all.

If you have any specific examples I'd be interested.


One example is when a user tries to upload a file to the system that is not allowed (permitted files are plainly listed), a large red error message appears that says "This file type is not allowed. Please upload one of the following:"

You would not believe the amount of tickets we get daily where people are confused as to why their file wasn't uploaded.

Automated bots clear a lot of this stuff up.


I think the issue there is that you're assuming all users are familiar with what "file types" are. It could also be that the format they are uploading is entirely reasonable so it should work. A better solution is surely to convert the file for them. That's what you're making them do themselves right?

I've ran into a similar situation lots of times when uploading photos on mobile. Often you get "this file is too large". Well no shit, we have the technology to automatically resize it now! It's not difficult! (Also have you noticed how Android lacks that basic functionality?)

Yes it is more work than not doing it, but it's definitely less work than implementing an AI chat bot.


In my experience, customer service are not that much more helpful than a chatbot (heavily undertrained or no privileges to do anything at all).


They’re untrained and untrusted by their employer, which means they can’t have privileges to do anything helpful. It’s super common and I don’t even understand why they have support at that point.


Sorry, who do you deal with these days who has a customer service line? The few that do, you have to go through like 3 minutes of robot "press 3 for this department," with the 0 shortcut to talk to somebody leading to an instant hangup ("goodbye") except on menu 27B where it works


I think that every business I deal with has a customer service line or a physical place I can go to and speak to a human. I try my best to avoid businesses that don't.


> Typically I call customer service because I'm trying to tackle a problem that cannot be solved by using their normal interface. There is usually some subtlety or exception that requires an agent capable of understanding them at the other end, otherwise I wouldn't have bothered calling.

I never get anyone that understands the basics let alone the subtlety. Do you have some secret method to get people that are knowledgeable?

I find that calling support is worse than doing nothing in a lot of cases. They simply waste time that you could be spending finding a workaround or migrating. Migrating is becoming a problem though because everything is consolidated into a few shitty companies with terrible support (ex: Microsoft).


What percentage of chatbots have you used that were running on GPT-4?


People don't want a useless chat bots, more often than not they seem deliberately programmed to be obstructive.

If you told it you wanted a refund for order xyz, and it immediately gave you a refund, I'm sure people would like it.


That's the thing.

At work, I hated an hr bot for years. But in the last year it became good and useful and I now like it.

If a bot gives me what I need I'll like it. If it only provides value for YOU though and makes MY life more difficult then I will not enjoy it very much.

Big problem with most bots is they don't have authority and they refuse to turn you over to someone who does. If I'm engaging with your customer service, it means I have an issue I need resolved. The worst bits are the equivalent of incessantly yet cheerfully asking me to restart my computer in an endless loop with no path out.


You should try talking to people who work in customer service call centers, most of their calls are arrogant opportunists that aren't calling to fix any issues... but to get something extra.

"This item is scratched, so I want a 50% discount refunded" - comes to my mind. Most of customer service is dealing with Karens


If you wouldn't buy a scratched item in store, why should you have to accept it from an online retailer?

I wouldn't call a person wanting their new item to be in new condition a scammer.


I think the idea is that the item isn't scratched. They're just claiming it is to try and get a refund/credit.


Oh that's fair enough. Both sides suck! I never worked in call centre, but I worked 6 months in Future Shop as a teenager (it was a Canadian version of Best Buy in the 90s), and there's a reason why I stopped as soon as I could - both the sellers and the buyers had higher percentage of dishonest bums than I was comfortable with.

The story that made me quit:

We had an extremely clear policy of no returns or exchange on opened printer cartridges.

A customer came in to obtain a refund. He claimed it was the wrong cartridge.

Did anybody help him or instruct him to take that cartridge? Nope.

Did he notice that it's explicitly not for his printer? Before he opened it? Nope.

I pick up the cartridge and... it's empty. Dry. Nothing left in it.

Before I could even say anything he looks at me, and start yelling and slamming his hands on the counter and making a scene. Not because he was upset, there was nothing to be upset about yet. It was just a rational, calculated move to obtain a refund on a cartridge he fully used. 30 seconds later the store manager comes over and gives him the refund.

Store manager coached me and indicated, this is basically retail life and where he comes in to provides judgment - the disruption and bad image this person created FAR outweighed the $50 he gave away. It was nothing personal and I should not feel bad or betrayed that I was overruled after following clear rules.

I absolutely 100% agree with manager's perspective. He did the right thing. I could not blame him or wish that he did otherwise. The only way I could find for that scenario to have a different outcome was for me to not be there anymore, so I tendered my resignation the next day.


You can provide that with a simple HTML form though.


That honestly never crossed my mind, but I think your right. Anything a chat bot could do or approve could simply be accomplished by an HTML form or some type of self service.

Chat bots will take over more and more, but only to save money, not to improve the customer experience. I do get that some people call in with the simplest of question or problems, but for the people who call in with actual issues, I doubt that companies will even grant the bots the freedom they'd need. Customer service people already aren't given the freedom and autonomy needed to help customers, so why should bot get it?

Say you call in to an ISP to have your internet connection fixed, is the ISP going to give a bot free-range to make adjustments in backends or dispatching technicians? I doubt it. All this is going to do for the companies that implement it is that they'll save a bit of money and have the bots hide any underlying issues or errors on policies.


Many of them are forms in disguise. I furiously hate Virgin Media's "chatbot", because every conversation with it is "what's your name? Postcode? Account number? Select from these 3 options". This is just a contact form with a slower UI.


If you issue everyone a refund, you won't be very successful in business. I think a GPT-powered chatbot (not an old-school one) could screen between:

- Genuine issues

- Scammers

Probably more effectively than a minimum-wage apathetic human being.


Isn't that the point of "This could be a form". The chatbot would have certain rules that it would need to follow and I don't see why those rules couldn't be translated into a form.

I wouldn't underestimate the mental capacity of scammers, it would only be a matter of time before they would know exactly how to prompt the bot to issue a refund.


That's orthogonal to whether you use a simple form or not. Filtering out false leads is a hard problem though.


For super simple stuff, sure, but many issues are easier to explain in natural language and a fluent UI (something with a lot of conditionals, well beyond a simple HTML form.


Yes, - but there are no bragging rights, promotions, or other rewards for the team/manager who put clear, useful HTML forms into production.


Depends on the management above them. (And maybe depends on the spin: "We introduced a clear HTML form that reduced customer loathing by 30%...")


Point. (And LOL on your "loathing" quip!)


NC DMV replaced all of their HTML forms with chatbots and it drives me nuts because I used to be able to pay my taxes and fees in under a minute, but now I have a stupid chat bot interaction that takes closer to 5, because you can no longer just complete the form and submit it but have to answer queries from the bot one at a time with some BS fake typing delay.


Yes. This article is criticizing the primitive chat bots deployed by companies today, not good ones that ought to be possible with the latest and upcoming LLMs.


I have never met a chat bot that worked. I try the same question worded 6 different ways, I try asking it to connect me to an actual fucking human being. If it even does I still get stuck in a choice menu at first, which half of the time tells me that I can get help with this via their chat bots and disconnects.

Fuck chat bots and the money-saving horse they rode in on.


Amazon. My personal best time for a full refund was slightly under 2 minutes.

It may be that having a long standing account, not claiming every delivery has been "lost" and rarely returning products makes me appear unlikely to be the usual customer asking for a refund.


My experience with that is the opposite. Bought inks, one bottle arrived broken. Took me a few hours to get through to a place where I could put my complaint.

Side note: after that they became even more useless. "Please mail the item back to us." Yeah no sure I'll mail you the shards of the broken bottle and a puddle of drying ink.


Had the same thing with some lotion.

Played dumb and contacted them again the next day and told them I put it back in the box and took it to the post office but they refused to take the package because they said you can’t ship leaking things. Asked them what they wanted me to do now.

They told me not to ship it and refunded it.


Meanwhile buying from the sketchiest possible "this official Arduino is 12 cents" sellers, I have lost $4 I spent on a bluetooth dongle that the "company" selling it disappeared a day after, but every other interaction, the sellers will jump over themselves to convince you to rate them higher than their competition. I had what I thought was a tiny drop shipper send me a double order because they were concerned I wouldn't get the product in time (like a day late on a 4 day shipping window, for free shipping) without me even complaining. Hell, they do this even though ebay probably lists me as a terrible buyer because I never add seller feedback.


> It may be that having a long standing account, not claiming every delivery has been "lost" and rarely returning products makes me appear unlikely to be the usual customer asking for a refund.

You had a request that was simple enough and common enough the bot's design took it fully into account.


If I’m not mistaken Amazon basically has a reputation for customers that is intended to penalize bad actors (and conversely reward good actors).


They have reputation metrics for the speed which someone walks around the warehouse picking stuff off shelves and the speed with which they use a bathroom (I assume using the toilet fast is prefered but washing the hands fast would be penalised).

I also suspect that the majority of refund requests are fraudlent, so if they dont have a reputation scoring system for their customers they are stupid.


> It may be that having a long standing account, not claiming every delivery has been "lost" and rarely returning products makes me appear unlikely to be the usual customer asking for a refund.

Being that sort of customer certainly didn't get Amazon to treat me decently.


Does having "Prime" give you better support, I wonder? (Do you have it?)


This exactly. I want a solution to my problem. If an AI can help me, great.

But it's not even close to a human level of service yet.

So I'll always try to get to a human ASAP knowing the AI will just waste my time.


They combine the charm and brevity of telephone voice recognition with the text parsing power of a knock-off early 90s adventure game.


Now you're just making it sound cooler than it actually is.


As a customer you want someone to resolve your issue, which is mostly a matter of someone being responsible to file a ticket and ensure it gets solved.

A chatbot has no responsibility, it is just a useless parrot of the documentation. It doesn't serve any useful purpose except worsening the experience for the customer.


The worst are the bots that can't figure out what you're trying to ask about and _refuse_ to turn over to a human because you the end user clearly aren't in what they are good at.


Had this experience with Vodaphone when I tried to set up my account and add my fucking phone number so I had insight in my bills. Didn't work. Got directed to a bot, which continually misunderstood me and no amount of prompting from me made them direct me to a human person.


The WORST is when calling a specific department you in the generic queue and the bot wants a piece of information that you don't have (eg. demanding an ATM card number when you only have an investment account) and hangs up when you don't give to them.


The thing is, I wouldn't be asking for support if I was able to find a fix for my issue in the first place. ChatBots are unable to fix the initial problem because the system used to fix said issue doesn't already exist in the account tools. Human intervention is required to fix the problem, so therefore chatbots are completely useless in this context.


> The thing is, I wouldn't be asking for support if I was able to find a fix for my issue in the first place.

You are the exception though, the vast majority of people don't do that. Those are the users the chatbot is designed to help (because they represent a large portion of a typical support team's interactions).


> Instead, the reason people go to customer service is because of a question that’s so specific, or complicated, or gnarly in some respect, that there’s no way the app will have the answer: you need a human.

I can see the author probably hasn't done much customer service? I'd say maybe 1% of customer queries go in that direction.

I agree dealing with a bot in that situation sucks, but a big chunk of customer questions can probably be answered without human intervation by LLMs, assuming they have enough data about the organization and its products.


You've nailed it. I've been a customer support specialist in tech startups for the last many years, and the HUGE and overwhelming majority of questions are about things that have been covered in detail in the support documentation by any even half-competent team. People do not normally search for their own answers. More than 90% of tickets are well-answered (according to the user's satisfaction score!) by a pre-written response explaining how an article in the support center addresses their specific question, and then linking to it.

This is why large companies end up hiring thousands of low-wage workers who are prohibited by their software from sending anything but a pre-written response. The first-level agents can't even edit it. They can only send you something someone else wrote for the common situation.

Chat bots could easily replace these workers, and the customer experience would be better for it.

You will (probably) always still need the higher level agents who are really tech support, the ones who can identify when someone's report really is a bug and get it reported to engineers for fixing.

Those aren't the low-level agents in these outsourced call centers, though. Even if those people do recognize an issue, they can't do anything until they've sent you some number of generic responses that haven't satisfied you.

At least the bot could be trained to customize responses within a set of parameters and match the tone of the asker.


But they're not replacing those agents with chatbots. They're replacing all agents with chatbots.


When I use these chatbots or contact customer support, it's probably because I need a very basic and simple thing done.

Not because I don't want to do it myself. Because the website I'm using is throwing error messages, the form to do it myself has disappeared five years ago, and the only way to accomplish what I want is to bother someone over the phone.

When I worked customer service, I'm sure customers could've done most of the things they wanted to do themselves, if the website wasn't unclear, the help articles weren't incomplete and outdated, the manuals were up to date and the necessary buttons were accessible to the end user.

Companies sabotage themselves with shitty business practices so their cheap support lines get overwhelmed. Almost everything I've wanted to get done through a chatbot should've been an HTML form in my account panel, but businesses don't want to make it easy to return things or ask for refunds. They want you to jump through as many hoops and redirections as possible, because that makes them money.

With the way ChatGPT just lies and deceives out of the box, I've started taking screenshots of chatbot conversations. These businesses are making these chatbots their official point of contact so I take everything these bots promise to do or claim to have done as an official statement from their support department.


> I agree dealing with a bot in that situation sucks, but a big chunk of customer questions can probably be answered without human intervation by LLMs, assuming they have enough data about the organization and its products.

Hoenstly? I don't care about the company's problems giving support. If I have a problem, and the company won't resolve it, or if they waste too much of my time forcing me to go through a bot first, then I'm not their customer anymore. There are plenty of other places who actually appreciate my business.


Yeah, the author is assuming that the general populace is like themselves (and like most HN commenters, I suspect): fairly high degree of technical competence, able to find workarounds for most issues in an app or website or forum somewhere.

But most people aren't like that.


Yeah, that statement is so out there and incorrect that it invalidates the whole article even though I agree with the premise.


Today I needed to contact a government service office for a task that only a human could perform. Now unfortunately, the phone tree of this office is broad and deep, you need to authenticate with about 4 data points before continuing, and if you hit the wrong option after authenticating, they will hang up on you without a second thought.

So I authenticated got through to the correct option on the first try, and it seemed like they were going to drop me into a queue to wait for a live operator. And then a recording informed me that "due to a high call volume, nobody is available to take your call." and dropped the call! Can you believe that. No option to leave a callback number or wait 4 hours in a queue, just terminate the connection. I called back and got the same thing.

The website does have a chatbot, but of course it is not capable of connecting to a human being and only capable of relaying the simple FAQs they've programmed into it.

I am pleased and impressed, for the most part, with the proliferation of Live Agent Chat for many services. I really prefer this over a telephone interaction (most of the time.) So it's good when I can easily break past the "virtual assistant" trying to FAQ me and get to a human.

But in this case, today, I suppose there were no humans to be found. Tragic. They have good compensation packages there.


  nobody is available to take your call." and dropped the call!
Alternative endings for the last step:

A) Nobody is available to take your call. Press 1 to leave a voicemail. Mailbox full. Click.

B) A human answers. They're not the right person, but can transfer you to the right person. Click.

C) Same as B, but after they transfer you you're told 'Mailbox full. Click.'.

D) Same as B, but they transfer you to the top of the phone tree.


E) all of the above.

Yeah, when I sense that (B) or (C) is coming up, I stop the person cold and I demand to know the direct-dial number for wherever they are transferring me. And these days I also ask to know the department name as well, so I can figure it out if presented with a tree or org chart or something. It doesn't always help, but it helps me feel a little more secure.


Yup. Ohio unemployment office. They didn't process my application during the pandemic. I had to sit through a 6 minute monologue and then immediately get hung up on after it's finished, many, many times. Several people I have talked to had the same problem. Their applications automatically expired and couldn't be resubmitted because no one on the government side did anything.


HMRC?

I'll bet my tax rebate on it.


I don't think the argument is "customers want chat bots", but "customers will hate our chat bots, just like they hate our (poorly paid and trained) human customer service, and it'll cost less".


I despise most chatbots. Especially the ones bothering me while reading an article. I'm leaving these sites without reading.

But I really dig airline chatbots. I rebooked flights quite easily in the past with that. It gave me clear options, I could scroll the list of flights etc., way better than by phone. And at inconvenient times as well, so no hotline anyway.

That way I changed a very unpleasant being stranded for 2h in a hotel room in Geneva for a nice long stay in a five star hotel. Landing at 1, being in the hotel at 2 at night. Flight was scheduled for 6 something in the morning. This would've been gruelsome. So I rebooked my flight to Frankfurt for lunch and enjoyed the breakfast buffet.


I believe this is a clear example of the Moloch effect [1].

Companies race towards having the shiniest chatbot or _virtual assistant_, without fully understanding all the implications and customer needs. Everyone wants to have their brand name associated with being "the first" or "the best" in their segment for providing such technology, no matter how trivial it may seem -- e.g. as the article points out: just wrapping already available info in a dialogue.

Given these incentives and high stakes, it's also hard to "exit" the bandwagon, due to various factors like reputation, costs, and the uncertainty around your competitor not doing the same.

[1]: https://danielmiessler.com/p/moloch-the-most-dangerous-idea


How did [1] write an entire article on this and not reference Scott Alexander’s Meditations on Moloch from 2014? https://slatestarcodex.com/2014/07/30/meditations-on-moloch/


Sure it's already dire, but who cares what customers want (especially the few 'difficult' ones)?

We didn't want phone menu systems, or to wait minutes, tens of minutes listening to muzak on hold, or to have to decrypt heavy foreign accents and painful latency on calls, etc either.

Customers will get what companies give them, which will be phrased as 'lower prices'.


In my experience they're useless because they're trained / limited to the FAQ and website you've already read to try to resolve your issue. The same with calling a helpline. Yes, I'm aware I can check your website for common questions. I'm calling you and need to speak to a human because the answer isn't there.


Please just understand you are part of the <2% of customers who have done your own legwork.

The overwhelming majority of people are calling in about something in the FAQ because they can't be arsed to do their own search.

The first layer of support at larger companies is usually restricted by their software from sending you anything but a pre-written response, just because you're such a rarity.


To add to that ...

I want to call a human even if my question seems dumb for them and have it taken care of, because I don't even know if I get that one answer there might be 10 other questions hiding behind the corner.


A good chatbot could probably address it just as well as a human going through the motions trying to achieve target.


As soon as somebody creates a good chatbot, we can start deploying it then.


> Please just understand you are part of the <2% of customers who have done your own legwork.

I understand this, but it changes nothing. Bad customer service is bad customer service.


My internet went down, and I couldn't reach customer service. I could only reach (drumroll), a chatbot. The chatbot couldn't confirm that anyone knew about the problem and was working on it. (How do I know that it just doesn't always say "someone's working on it?")

I submitted an FCC complaint.

A few days later, a customer service rep called me. She told me that I could contact the chatbot. I told her I'd never submit an FCC complaint if I could talk to a real person, because I don't trust the chatbot.


Wouldn’t this be a perfect use case for a chatbot?

Can you elaborate on why you didn’t trust the chat it but would trust a human?


> Wouldn’t this be a perfect use case for a chatbot?

It's a horrible case for a chatbot. Basically, when you have a bunch of hopping mad customers, you don't want to rub salt in their wounds. Refusing to answer the phone in such a situation is horrible customer service.

Basically, if you refuse to answer the phone and stick out a chat bot, it looks like you're just hiding from your customers.

(This should be obvious if you've ever seen post mortems on customer service disasters, like the Intel Pentium bug.)

> Can you elaborate on why you didn’t trust the chat it but would trust a human?

Two reasons:

Basically, I don't know what kinds of integrations, or lack of, are in the system. So, for all I know, they could have something that just says, "don't worry, we're working on it," if it detects that my internet is offline. (IE, I don't know if anyone is actually working on it.)

No context: The system would only respond with a short, canned answer. First it said that everything was going to be fixed at 1:00, then a few hours after it kept saying "fixed soon."

(Again, this should be obvious if you've ever dealt with automations, or ever seen an error message.)


What’s obvious to me is that I’m not calling customer service to talk to or yell at a human. I am calling to get my problem solved. Perhaps you and I have different goals?

> So, for all I know, they could have something that just says, "don't worry, we're working on it," if it detects that my internet is offline. (IE, I don't know if anyone is actually working on it.)

Why would you trust this exact communication from a person over a computer?


> I am calling to get my problem solved

That is the only reason why I call customer support in internet outages. I usually wait awhile before I call. (In this case, I called a few hours after their posted ETA.)

> Why would you trust this exact communication from a person over a computer?

This response in this thread explains it very well: https://news.ycombinator.com/item?id=36661196

> Conversely, a weasel worded chatbot could say "we are responding to outages as noted" or something similar, and that doesn't actually provide any evidence the utility is aware of the outage or its scope in the actual monitoring and management systems; your report of a failure could go straight to /dev/null.

Like I said, above, I called a few hours after their expected ETA. This is the situation where it's time to call a human who can call up the manager of a different department and say, "you said it was going to be done three hours ago, what's happening?"

It's basically a "Gall's Law" situation. The AI Chatbot is perfectly fine for a "normal" outage that's solved within the ETA; but once the published ETA has come and gone, we're in a corner case that's better handled by humans.

For a good explanation of this, see this recent article on Gall's Law that explains why it's better to let humans handle corner cases: https://bigthink.com/smart-skills/complex-project-galls-law/...


The local power company (a psuedo-governmental co-op, I think) has a website with a big interactive outage map. When the power goes out, the icon is usually on the map by the time I've pulled up the site on my phone, describing the outage area, and the ETA is usually surprisingly accurate.

100% automated service, but as good if not better than anything I could get from a human.

Conversely, a weasel worded chatbot could say "we are responding to outages as noted" or something similar, and that doesn't actually provide any evidence the utility is aware of the outage or its scope in the actual monitoring and management systems; your report of a failure could go straight to /dev/null.


Accountability for one. Another being hallucinations are still too common.


Accountability for what?

I’ve had human customer service members tell me things that I knew to be false. At least computers should be able to calculate confidence levels, like Watson was doing for Jeopardy a decade ago.


Even the Amazon chatbot is fairly useless, and like many chatbots it is slow too, seemingly to mimic a human typing speed which is really annoying when it's not even pretending to be human.


Completely agree with the content. Side note on that:

> Customer service is one category where I expect a lot of these jobs, honestly, to just go away.

For long I have, and in normal conditions, I would have said hurray. Humanity invented a technology that automated some of its bothersome tasks, one thing less to do, we're collectively all for the better, freed to enjoy life or spend our time in a more productive way, whichever suits best.

But in this world, the fruit of this automation is always reaped by the same people, and they're not those whose job was automated. The latter are now in a shity situation, the former are now marginally richer and won't share with anyone else.


Yeah, lots of people think rich capitalists like exploiting people for their labor to enrich themselves, but the truth is they hate it. It's is an unfortunate necessity. If they can replace bothersome employees with a system that does the job "well enough" they'll do it over night and won't lose a wink of sleep over those who were displaced.


Note that I didn't talk about the motivation, but just observed that this is how the system works.

> but the truth is they hate it

They hate having to rely on other people, or the exploitation part? Because if your thesis is the latter, go explain union busting, benefits stripping, or pretty much the entire state of the social system in the US.


> and won't lose a wink of sleep over those who were displaced.

So then they don't hate it after all.


>>Yeah, lots of people think rich capitalists like exploiting people for their labor to enrich themselves, but the truth is they hate it.

The people who genuinely hate exploiting people for labor to enrich themselves do not become rich capitalists, in the same sense that people who genuinely hate heights do not become airplane pilots, and people who genuinely hate public speaking do not become theater actors.


> but customers can get those answers already from the app or website.

We’d need to have access to the specific company’s call categorization to know this. Just last week I was chatting with someone in their mid-thirties who was telling me how they don’t have an ATM card or access their bank through the app or website. Everything is done in person at a physical branch. They need money? They go to the teller. And I know many older people who are the same way.

This is why many companies already have automated phone systems that walk customers through the easy to solve problems with mostly rote formulas. Add in a chatbot that can be better at handling slight exceptions and now the company can solve customer issues without having to pay a human to do the interaction.

From my time answering phones for an internal IT department for a large-ish company, it’s clear that some people called just to interact with and talk with a human. Even if they could have solved their problem through an app before a human answered the phone.

> Instead, the reason people go to customer service is because of a question that’s so specific, or complicated, or gnarly in some respect, that there’s no way the app will have the answer: you need a human

I’d suspect most of these aren’t all that specific or gnarly, but just that it can’t be done through the customer facing app, yet. When switching from physical SIM to eSIM, Verizon’s website kept erroring and directing me to call. I looked online and many posts said it was quickly solved with a call. I called and they fixed it quickly. That could almost certainly have been handled in an automated way, just not yet. I don’t think they even asked any questions, other than confirmation style ones, which a chatbot could have done. I needed to waste time calling and waiting in a queue to talk to a human. The company needed to pay someone to be waiting to answer that call and spend a few minutes with me. Seems like a great spot for a chatbot.

My expectation from my limited experiences is that the number of calls that a sufficiently large business receives are not the specific or gnarly cases. I’d love to see real data.


I think the author overestimates the technical level of most requests fielded by call centre agents. They also underestimate the potential a well-designed chatbot could have given recent technology developments.

Speaking with colleagues in private banking call centres (wealthy clients who you'd assume are more clued up). Many calls are basically comfort calls, holding their hand while they login into the bank account, they complete all the steps themselves with no input other than reassuring comments like "yes". Some probably see it as making sure the bank is liable if they "break something".


The author is assuming a decently high level of technical competence here. I guarantee you there are people calling into customer service with basic queries already handled by an app or website.


It’s not just the tech industry and this article is spot on.

Every support call without exception I hear the excuse “due to higher than normal/expected call volume your wait time….”


> Instead, the reason people go to customer service is because of a question that’s so specific, or complicated, or gnarly in some respect, that there’s no way the app will have the answer: you need a human.

A couple counter points that are kind of the same point.

1. I actually hate talking to customer support that doesn't speak my language as fluently as I do. It makes communicating the important part of what's wrong in these complex scenarios that much more frustrating.

2. Customer support, even when human, often is useless because their roles get restricted to scripts. Discover is a company that used to have amazing customer support, especially when dealing with the complexities of charge disputes. I listened to a woman refuse to give me any idea of what "evidence" I should submit to a charge disputes with a hotel that was billing me despite no services rendered. Instead she read off what seemed to be a script. I hung up, dialed back in, got someone new and they answered my question in great detail.

I think my point is the same as the authors; there should be a high bar for customer service but I'd take it one point higher and say we've been letting that bar slide for a while.


As a customer, I have nothing against chat bots, if they solve my problems.

Last December I had to engage into long chat sessions with airline rep to enquiry about my flight change. I kid you not, they changed my first leg from morning to evening flight, and I only found out when printed boarding pass an evening before. Then I had to change my flight back due to work requirements.

It was not cheap flight: four legs with two connections from Europe to SE Asia business class, so I believe I received a good service. It took about three hours total. My website auth expired, I had to re-login several times. Fortunately, the dialogue kept going during re-logins. I had to wait a lot while call center employee checked everything, and once I found a mistake right before paying a new fare. I got transcripts to email. I would rate a tech behind this conversation as above avarage, but still the overall experience was painful.

If some LLM chat agend could do the same, but quickier and with less errors - I would just be happy about it. I don't see it as an impossible feat, most of the responses are already scripted anyway.


Most customers don't want to pay for good customer support.

This answers the question of the article. What customers want is offloading responsibility without paying for it. There are many concierge level services at the premium end. Chatbots are an attempt at offering this type of luxury to the masses. And it may succeed.


There was a chatbot I dealt with recently, where the only way to get a human being was to type in graphically vulgar language repeatedly. Unfortunately the human who does eventually join the chat is then usually subject to this language as they "review the chat log" :/


I do this often, as well. I find the trick is to intersperse messages like "sorry human --the bot only responds to this aggressively-abusive language." So far, this has worked like a charm.


Chat bots built as Eliza clones are worse than useless. However a more concise ChatGPT hooked up to APIs to achieve customer service goals would be great. It operates in an abstract abductive semantic space that’s considerably more deep than most first level support can, especially those poor souls working from a different country with a different native language. I would assume most if not all companies would maintain second level and third level human support for the truly mindbendingly challenging inquiries, but those are a very small percentage of the total volume. For that 90%+ of the case volume a well contextualized, fine tuned, and vectordb’ed chatbot LLM of at least GPT4 capability is likely going to our perform almost all situations in almost all dimensions, including identifying situations for escalation to next level support.


Our company, a travel provider, has a "chat bot" but it is just the first stage of a layers customer service experience. It is not an AI backed bot, though I'm sure that will come, eventually.

The bot can handle basic functions like information, change, and cancel but on more complex servicing, it brings in a human agent. The customer always has the option to request a human agent. This lets customers self service for the simple stuff but still get a helping hand when they need it.

In surveys our customers seem to like this approach with some indicating that they like being able to use something like this on their own rather than having to wait for an agent to be available. I think, as long as you give your customers options a structured first layer can be helpful as long as you give them a way out to more hands-on support.


I'm creating bots like this where the ai enhanced bot is fed a clients docs/industry info, and it's trained on closing techniques specific to the industry and it basically keeps following up on a schedule until the customer decides to book an appt, or opt out of further messages.

I'm planning on other automations in the future for other workflows, but lead closing and appointment setting on autopilot 24/7 with a single monthly fee that's way less than hiring staff(even from India), in the end it's about giving automations knowledge and power through that knowledge but still the end result is a booking or action to often speak with someone after being warmed up.

A friendly non pushy ai sales agent might even be better at sales than a seasoned sales person after some trial and error on training it. My agency is in it's infancy but I'm excited to see what our sales pros can/can't do.


I think chatbots can and will replace the useless frontline "Tier 0" support agents that all too many companies have. The ones who are working from a script that they can not substantially deviate from where most of us here are just trying to get through to someone who can really help.


Yeah, Facebook removed their CS. So when my FB account was compromised and they blocked the trusted contacts recovery method (I was told I was trying too often, the first time I tried it and so it went for months. Guess they had a bot to basically dos the trusted contacts system for stolen accounts), there was nobody who could help me.

There just wasn't anyone I could contact. I searched for days. I filled out and submitted multiple large support forms, none of which I ever received an answer to. Eventually I contacted a local Norwegian organization named Slett Meg (Delete me) and asked them to either get my account back or have it deleted. They had it back to me in like two days.


Honestly most customers do not read anything on the website before they contact support and in these cases a chatbot, given that it is trained well is probably the best solution. I personally despise them because I don't want to interact with anyone if I can solve a problem myself so when I actually have a problem it's usually just in the way of getting an answer.

But having worked technical support in the past I know that most customers don't do any sort of research at all and will (not happily exactly, but yeah) sit through two hours of phone que to ask the most basic of questions that is in the manual, on the website or anywhere on the internet really.


Problem is that there are so many questions companies (and government) do not want to give answers to, or they want it split per market, or geographically, or only to citizens, or only to their own employees.

Online you find a lot of information you're just not going to find in chatbots.

This is why Google had such extreme value. Organizations, certainly governments, generally had search functionality, but it was purposefully sabotaged according to each organisation's capricious needs and therefore sucked. The problem is that the information available and searchable was definitely NOT accidentally sabotaged, they actively want to prevent people finding out.

Chatbots are going to suck like company search fields suck. Not because the technology behind them sucks, but because their purpose is to change what you want, not give you what you want. And when this fails, and fails, and fails some more it becomes and endless horror of frustration.

For example: https://news.ycombinator.com/item?id=36624294

This is what companies want. They want to use frustration and sabotage as a way to get what they want. Chatbots will rarely be anything but a new weapon to do so.


Workers don't want open-office plans either but they keep building them.

And there's a vocal minority that want to be able to buy a smaller truck in the United States, but they keep on building monstrosities with beds too short for a sheet of plywood.

Time to start realizing that Capitalism doesn't really offer consumer choice or optimizes for the best outcomes for people, it optimizes for the best outcome for the business.


~"because the chatbots aren't good enough"

Chat bots are annoying due to potential inconsistency, having to guess the types of things you ask for, due to their latency, but they've made major advancements recently, and we are only beginning to see what's possible.

If those companies using chat bots are failing to provide value, then they won't be competitive and they will go away. Why do we want to discourage them from exploring this new area? Experimentation is critical to understanding what will work and to the development of our society.


> If those companies using chat bots are failing to provide value, then they won't be competitive and they will go away.

The difficulty is, this is obscured to the customer.

Let's say I have a broadband account in a country with competitors I can choose between (eg the UK, where I can have BT or Virgin). I've already chosen the best package for me. I'm ringing up after six months of good service to fix my intermittent line drop.

The chat bot is terrible and I have to wait until the problem goes away on its own.

Unfortunately for me, I can't just guarantee that switching to the other service will make a difference, because chat bots are not core service. I can compare broadband options and pick the best, and I'm already on that. Now I have to gamble that some combination of:

(broadband option) + (customer service) is higher than the one I'm on.

Unfortunately for me, I have no way of checking this. There's every chance that I'll end up on a worse broadband option and a worse customer service option.


If you want decent support as part of the core offering you can pay a bit more for Andrews and Arnold, their support staff are in the UK, all technically capable rather than KB parrots, and available by IRC. Lower customer contention for their shared bandwidth so it doesn’t slow down at night is another thing worth paying a bit more for. And they have a decently technical control panel web page.

https://www.aa.net.uk/

(Satisfied customer for years).


I have heard good things about their customer service. But I put in my postcode to their site and the response was:

> For your location, forecast download sync speed is 27-37Mb/s and forecast upload sync speed is 4.2-6.9Mb/s.

I’m currently on ten times that and was considering going higher next time. Unfortunately that’s not currently competitive with Virgin.


Most importantly, the chatbot cannot actually solve anything, it is not plugged into any functionality.


That's my feeling too, but maybe it's outdated? There's no real reason why they can't be.


there is plenty of reason why it shouldn't be. Generally speaking functionality is gated behind support (as opposed to being something you can just click a button on a website to do) because it requires at least some level of human judgement on whether a given action should be done (and there's still a whole class of attacks which are aimed against this process). Chat-bots will be even worse, because there will be an extremely consistent script than can be used for any given purpose.


FTA: A chat bot can handle expected, middle-of-the-road queries (as in, for a bank, “what’s my balance” and so on) – but customers can get those answers already from the app or website.

This is where the author misses the point. Chat bots will be te main interface in the future. Especially for these mundane middle of the road queries. Integrated into voice controlled ai buddies.

The author is spot on for the more complicated requests. Even though these will be triaged by a the ai buddy.


I definitely want and prefer chatbots. Telephone/internet customer service isn't a job that people should be doing, it's soul destroying.

The problem with chatbots is that they make customer service time far cheaper than my time. If they have to pay $15/hr for somebody to argue with me about a $10 refund, they're motivated not to do it. If that cost goes down to $0.15/hr, however, you're incentivizing Kafka.


> Telephone/internet customer service isn't a job that people should be doing, it's soul destroying.

That's true, but it's a different issue.

My gas company has already deployed a number of chatbots and automated phone responses. Now you simply just can't talk to a real person via phone.

These days they double-charged an invoice and I simply had no way to inform them about the problem they created. And this is not the sort of service you just move on to a competitor.


Every bot I've dealt with wasted more of my time to save the bot owners their time. Chat bots are a giant insult to users who are forced to use them.


This is the real threat of "AI". Companies are going to use it to make things that already suck even worse in an attempt to reduce costs.


A little different stance on the topic: AI being able to replace these people at their jobs is not because AI has to be better than these people but because most of the people working at call centres or supply chains are extremely bad at what they are supposed to do. So AI just needs to be at par whatever the industry requires.

Only matters that require a transaction are to be dealt with with caution.


I would be happy talking to a chat not the was a real chat not not just a narrow and shallow decision tree with every leaf ending in connection to a human. Train a LLM on actual subject matter and give it access to customer data. The send a request for a change to a customers account to a human for approval. Example: making a minor change to a phone plan.


Generally I don't want ways to communicate with businesses that leave no trace and resolve them of all accountability.


A lot of chat bots will email a transcript of the interaction to you.


I used to not want chat bots because they were bad. I can imagine situations where I would prefer a chat bots over humans. For example, if it's a niche problem that a human would be unaware of and would have to look up, read, and verify taking a long time to respond, I would definitely want a chat bot or a bot assisted human.


Nobody cares about customers. If you can replace the support department with a chat bot, then it'll be done.


> Nobody cares about customers.

Not even customers care about themselves!

Customers say they want a good support, but nobody buys a good support: customers buy cheap! If ditching the customer support department will let you lower your prices, you'll have customers; is maintaining a decent customer support service will keep your prices a few cents more than your competition, you will have no customers.


The critical problem with chatbots is that companies don't trust them to do things. I generally am reaching out to support because I want something done, not because I want advice. This may not be true of everyone but for many, they have a real need. I need X price matched, I need to get my bill corrected, etc. Users know the chatbot won't do it.


As soon as you put a chatbot in front of customer, it's a strong signal how little you value their business.


Chat bots are the same as the support forms that suggest articles.

Ergo useless.

If I've arrived at your support form it already means I've done everything your articles say they do.

Yes the modem is plugged in.

Yes I've rebooted it.

Yes the cable is good.

Yes the house's electricity is on.

No you didn't report an outage.

files claim.

2 days later they replace a piece of cable 2 neighbors down and things work again.

...


They'll be hackers posting ways to defeat the AI restrictions so it will connect you to a human.


> "The tech industry continues to push forward with chat bots, even though customers don’t want them"

The same was true for the previous explosion of AI. Autonomous driving. That was, and remains being, an almost entirely industry-side push to the marketplace.


OpenAI's ChatGPT has over hundred million users. Customers want chatbots just fine, they just don't want a shitty broken useless chatbot that's a waste of time. If it's a bot then respect the user and give it enough CPU resources to answer quickly.


This:

> OpenAI's ChatGPT has over hundred million users

in no way validates this:

> Customers want chatbots just fine


Can you explain why? I'm not making the same logical connections as you are.


Contextually the topic is "customer support chat bots". Openai's Chatgpt service isn't that sort of chat bot.


thanks. my intuition is that the demand is, actually, there - customers don't want to read through an faq and a knowledge base. they want to go "I <their problem>, can you fix it?". whether that's to a human intelligence or artificial is besides the point, they want someone who's able to understand it first, empowered to fix it, second. where a bot can only regurgigate the docs, psh.


To expand on staticman2's reply, the size of the set of ChatGPT users says nothing about its composition, and it certainly includes AI researchers, representatives of companies looking to build their own bots, and no doubt many people exploring the thing just for fun. The number of users by itself therefore says nothing about whether or not people in general want to interact with bots for customer service queries, since it could be that, say, 95% of ChatGPT users are there because they want to learn more about ChatGPT, not because they generally enjoy fulfilling daily tasks with the help of a computer.


I use Ting as my mobile provider. They use the T-Mobile network so the coverage is not on par with Verizon but it's good enough. I'm willing to sacrifice the coverage because when communicating with Ting I ALWAYS get a very helpful live human being.


I used to use Mint, and their chatbot said customer service was closed and to try back in the morning. Interestingly, I typed "emergency" in all caps three times, and it connected me with a human.


I think the calculus of a company deciding on an imperfect automation (such as self-checkout) is that it either won't prevent a purchase. in the case of chatbots, the purchase already happened, hiring a person to do this is just overhead.


> in the case of chatbots, the purchase already happened, hiring a person to do this is just overhead.

But poor customer service may be costing the company future business. Not to mention just increasing the odds that people will hate the company and start badmouthing it to others.


I want AI to do my plumbing, electricity, service my car, and fix the washing machine. Actually do these things, not create some abstract walled garden mobile app-based layer. Chat is the most useless superfluous and irritating application.


chatbots are the future of customer service.

The past of customer service is the complex phone system where you select a number from a menu, and they automate some tasks (like paying your bill, etc.) and sometimes you end up talking to an agent.

But the youngest generations hate talking on the phone and audio is a terrible medium for a menu. A chatbot can provide similar functionality, but you've got the ability to provide a much richer interface, and when you end up with an agent, they'll chat instead of talk on the phone.

As chatbots get better and better at performing actions, you'll have to talk to fewer agents, and waiting on hold will hopefully be a thing of the past


Chatbots can be awful. I called one yesterday because I needed to set up a new account. Twice I tried different phrases for new accounts and each time it sent me down a path where it would invariably ask me for an existing account number.


Honestly, I've heard that for a regular user chat bots are amazing. Tech people are just bunch of nerds who are used to going through documentation. For a regular grandma it's so much easiser to just write like she's used to.


This is an underrated comment.

The overwhelming majority (like 98% or more) of support requests are from people who have not read the documentation and are asking about something that is well-covered within it.

Chat bots can't (currently) eliminate the agents who exercise critical thought and report bugs, but for any company large enough to outsource "tier one" support to a call center overseas, the chat bot would actually be a big improvement.


> for any company large enough to outsource "tier one" support to a call center overseas, the chat bot would actually be a big improvement.

More likely, a lateral move rather than a big improvement. But you're right -- companies that outsource the support to crappy call centers are companies that barely have customer support at all.


I would qualify this.

There is a distribution of users. Definitely some users are better off (no matter how good the chatbot is) using the FAQ or websites. Some people are not chat interface orientated.

Some people would benefit from a very powerful and effective chatbot. The grandmother you cite for example! But the problem is that almost all chatbots are incompetent and really function to:

1) deflect users

2) bolster the innovation credentials of the organization or sponsors for the project.


Chat bots and automated phone customer service suffer from the same problem - they don't solve the exact problem/questions I'm running into and it requires a real human to get me the answer I need.


Strong disagree. The way I see it, any successful customer service operation needs to have three things in place:

1. Properly defined and articulated customer service policies and procedures

2. A training program that adequately prepares agents to know, understand and apply those policies and procedures

3. Sufficient inherent capability and disposition of the customer service agents to apply that training

My take is that the (relevant) inherent capabilities of LLMs at even at this early stage are as strong or stronger than what I see in the human customer service agents I typically deal with. And the disposition of an LLM can be arbitrarily good through fine-tuning and prompt engineering.

As such, any company that gets 1. and 2. right has a high chance of providing a better support experience by replacing human agents with LLMs.


Even more insulting are the telephone chatbots calling me to sell things while pretending to be government workers. A couple of cases lately.


Chat bots are a way of saying Fuck you....without saying it.


Question: would you rather talk to a GPT-4 chatbot or a human?

Answer: human.

Question: would you rather talk to a GPT-4 chatbot RIGHT NOW or a human after holding for 30 minutes

Answer: chatbot!


There must be a usecase here, allowing telephony and chatbot to filter unknown numbers.

My regular contacts? Straight to human.

An unknown number? Suspicious chatbot verifying who they are.


Why would you need a chatbot for this? When I get an unknown caller, they get my voicemail where they can tell me who they are, what they want, and how I can call them back.


I just want useful bots. A good one is in Bloomberg’s help function. Mostly it’s there to route you the the right human, but it does save time


A little different stance on the topic: AI being able to replace these people at their jobs is not because AI has to be better than these people but because most of the people working at call centres or supply chains are extremely bad at what they are supposed to do. So AI just needs to be at par whatever the industry requires.

Only matters that require a transaction to happen are to be dealt with a little caution.


Not rubbish chatbots anyway that just spit out knowledgebase links.

Something more empowered and smart could be good I think


100% no customers want to chat with a machine, they want to talk to people and be connected!


> Instead, the reason people go to customer service is because of a question that’s so specific, or complicated, or gnarly in some respect, that there’s no way the app will have the answer: you need a human

Frankly, such situations are pretty crap even with a human. Support systems are not set up to handle the high effort corner cases and usually you have to go through a couple rounds of escalation.


The reason why LLMs are truly revolutionary is that they suddenly allow to connect and interface systems that used to require a human in the loop because of their ambiguous interfaces.

LLMs can (and will) do a lot more than providing chatbot-like interfaces. Instead, they will power agents that can manipulate a variety of tools and interface with extremely diverse systems, including end users, SMEs, etc.


No they won't.

You're right that LLMs can do this. Well, I guess you might be wrong about that too, but it doesn't matter. That's not the problem. Companies have very large callcenters with actual humans that can do this ... but they don't. They're not allowed to. The only thing callcenters are these days are a semi-human interface to a bad website. LLMs won't be anything more.


yeah i do, skimming biology literature is tedious and requires extensive effort. i'd love a chat bot trained on scientific literature where i can ask questions about what is out there on very specific topics across articles that might otherwise look irrelevant.


Customers & PMs want guarantees of boundaries for knowledge & QnA tasks


AI customer support indicates low quality services. Simple as that.


Customers don't want crappy chat bots.


I don't think this article is correct - it's not right to say that customers don't want chatbots, it that customers don't want the chatbots of today.

> Instead, the reason people go to customer service is because of a question that’s so specific, or complicated, or gnarly in some respect, that there’s no way the app will have the answer: you need a human.

This is the thrust of the whole thing, and the problem it's not looking quite deep enough at what people want. People don't want a human - people want their problem solved.

A human is probably the best way to do that, but realistically, if you consider the role of a human in these situations, it's as an interface between the customer and an internal system. The human translates natural language into a series of actions in some computer program (or maybe multiple programs and also communication with other humans).

This is something that LLMs will inevitably be better than humans at. I just called my new health insurance company to get my member number. I spoke to a woman who first had to open up my account and spend some time looking at information. An LLM will do this instantly, and the aggregate time savings of this across all customer service calls will be enormous. That's on top of the fact that no one will wait on hold because all agents are busy.

She eventually realized she had to transfer me to another department. She gave me the direct phone number for that department then transferred me. Except instead of transferring me, she dropped the call. Then I called the direct number she gave me, and it was the department for cancer and complex medical claims. We've all had experiences like this.

Humans are not anywhere near perfect at this job. LLMs may not ever be perfect either, but they'll be better. A human at a call center either needs to be trained, which is costly and time consuming, or they need to be given rote instructions for all possible tasks. Neither of these make them particularly well-equipped for questions "so specific, or complicated, or gnarly."

Obviously this won't all happen overnight, but it's not tough to imagine that cases like Wendy's replacing their drive thru employees with chatbots can be successful in the immediate future. A Wendy's employee in that role just hears natural language and translates it to a series of button presses on a screen representing a very much finite number of possibilities. This is well within the capabilities of GPT-4. I think we'll see this more and more often for low-level customer service, with the chatbot passing the user off to a human for more complex things. The chatbot will move further and further up the stack of complexity until it's doing everything.


> that there’s no way the app will have the answer ... Hey, my credit card isn’t working, and maybe it’s because of a weird charge I had on it yesterday, but then something else happened right before this purchase” – that’s the sort of complicated scenario

Hmm, what makes OP think a chatgpt can't answer this question? Given enough tokens passed to a ChatGPT session, a chatbot can eventually find out what's going on and provide a simple to understand explaination.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: