When I worked at Microsoft, it cost over $20 to have a human customer support agent pick up the phone when someone called in for help. That was greater than our product margin. Every time someone called for help, we basically lost the entire profit on that sale, and then some.
Most common support calls where for things that were explained in the manual, the out of box experience, tutorial documents, FAQ pages, and so on and so forth.
Did we have actual support issues that needed fixing, yes of course. And the insanely high cost of customer support drove us to improve our first use experience. But holy cow people don't realize how expensive support calls are.
Edit: To explain some of the costs - This was back when people worked in physical call centers, so first off we were paying for physical office space. Next up training, each CSR had to be trained on our product. This took time and we had to pay for that training time. We also had to write support material, and update that support material for each new version that came out. All of this gets amortized into the cost of support. Because workers tend not to stay long, you pay for a lot of training.
Add in all the other costs associated with running a call center and the cost per call, even for off shore call centers, is not cheap.
In a reasonable world we'd just raise the price of the product by $x based on what % of people we expect to call in for support (ignore for a minute that estimating that number is hard), but the world isn't reasonable. Downwards price pressure comes from all sides, primarily VC backed competitors who are OK burning $$ to gain market share, and competitors at other FAANGs that are OK burning money to gain market share.
The result is that everyone is going to try and reduce support costs because holy cow per user margins are low now days for huge swaths of product categories (Apple's iPhone being a notable exception...)
> When I worked at Microsoft, it cost over $20 to have a human customer support agent pick up the phone when someone called in for help. That was greater than our product margin. Every time someone called for help, we basically lost the entire profit on that sale, and then some.
This doesn't seem like a bad thing when it comes to aligning incentives (assuming customers actually want a product they don't need help to use).
And if it turns out to be your mistake (faulty product or missing documentation) as opposed to something the user could have reasonably solved by themselves, refund the charge and possibly provide compensation for the inconvenience.
Microsoft is a company that has very little right to complain about support costs. They'd save themselves a fortune if they stopped releasing bad software and updates that required support in the first place. Nobody wants to call Microsoft for support. They do it because they've been forced to, usually by Microsoft. This kind of support can hardly be called "free" because even when Microsoft isn't charging customers to speak with the person on the other end of the line the customer has already paid in time and suffering (and sometimes lost data)
Users are "dumb", and it's a dumb _system_ and dumb business that doesn't plan for that in terms of FTUE, business model, support model, and product flows.
We product makers get to think about our one little product all day, and it's our job to make our product work for the "dumb" users. It's not their job to adapt to us.
Isn't part of why Apple's iPhone can be so expensive is because it's very easy to get actual human support for it when something goes wrong? You probably didn't make the mistake at Microsoft, but I've seen people look at the localized spreadsheet and miss the long term company wide spreadsheet completely. Often because the sales and support departments are so far from each other that they're basically two different companies working in different directions. Maybe Microsoft customer support is a bad place to measure these things because of the size, but around here quite a few banks have tried outsourcing their phone support to everything available and have come back because it cost them customers. Even customers who never phoned them.
That being said. Your example of customers calling for support on things they shpuld be capable of figuring out themselves in is probably where AI is going to shine as first line support. Once (if?) AI voice chat is good enough to replace chatbots we may not even realize we're talking with an AI unless it tells us.
>customers calling for support on things they shpuld be capable of figuring out themselves in is probably where AI is going to shine as first line support.
It certainly won't be cheap to run real-time AI voice chat, or any real-time AI chat. The AI costs that you currently see are heavily subsidized, just like OP's example of "VC backed competitors who are OK burning $$ to gain market share", it's the same. These AI companies are far from profitable, burning billions to insert themselves into customer support pipelines and everywhere else they can, and then the other foot will drop. Uber and Lyft are far more expensive today than when they started, and the price to run "AI" will also inflate when these companies have to pay off all the billions they've spent but didn't earn. I doubt it will end up costing much less if less at all than human support, with worse results.
My last experience with a support chatbot was actually pretty decent. It collected all the information, asked followup details, and then fired that whole thing off to a human to deal with. It was perfectly fine.
my only experiences with chatbots so far have been as instruments for companies to avoid their contractual obligations and just not provide the options that I would have asked a person directly for
obviously not a problem with the technology itself, it was like that with more primitive answering machines as well, often there only to either answer the obvious things, or stonewall people with real problems with the product or service hoping they'd just give up and take the loss
I mean that is also the job of existing call handlers.
"We are experiencing an greater than usual call volume, please wait while an agent becomes available" only to be randomly disconnected has been a thing for most of my life.
Everyone seems to be hyping open claw at the moment soon its just going to be LLMs talking to LLMs.... I wonder if they will develop a short hand and start talking in wingdings.
"smart answering machine" seems like a very apt use case for LLMs, provided the rest of the system works - that a human actually received and acts on the feedback.
This is the thing that drives me crazy. Most of these phone calls should just be emails; I can usually stand to wait a week or two for the company to get back to me. General support funnels like support@example.com have been dead for most consumer-facing technologies for close to a decade at this point. I’m not installing an app for every company I’m forced to interact with when there are already existing, universal technologies available that they could implement if they just priced their products appropriately.
Yeah. I recently had to deal with Amazon's robot. Definitely bird-brained but close enough that the right objective was accomplished even though I don't think it ever understood what happened (but woe to the non-native speaker!) The problem is not chatbot customer support, the problem is bird-brained managers that think a system that solves 99% of issues doesn't need a fallback for that 1%.
Whenever I interact with them I get asked to describe my issue then regardless of what I write I get asked a battery of questions you would expect are getting fed into a form and then on the off-chance I get connected to a human operator (which was my goal to begin with) they end up asking me for all the same information again.
People prefer a pricing model in which support appears free. Free support (that is good) creates the sense that the company stands behind the product and service, and leads to good reviews, so it is a win/win.
They need a place to be, they need to get hired, trained, managed and all the associated general costs of employment (hr, payroll, etc). They need equipment, there's monitoring, evaluations etc.
Then you also have to pay them regardless of whether someone calls.
While I agree with TFA's point that forcing a chatbot isn't a substitute for just having the info available, organized and searchable, the answer to your specific question is that the fully burdened cost of a trained support center human includes a lot more than their gross hourly wage. There's recruiting, interviewing, hiring, training plus space, desk, computer, phone, IT, HR, health care, vacation, sick days, insurance, employer's share of employment taxes.
A rough rule of thumb is the full burdened cost of an hourly office knowledge worker is two to three times the gross hourly wage.
The root problem is that these big companies are not capable of serving the customers that they have but because they have a monopoly, the customers are forced to use them.
All alternatives which are capable of actually serving the customer are systematically driven out of business.
Had they built a better, more intuitive product, they would get fewer support calls and wouldn't be struggling with costs.
> Had they built a better, more intuitive product, they would get fewer support calls and wouldn't be struggling with costs.
As I mentioned, due to high support costs we worked to improve the UX and we ended up dropping our support costs dramatically.
Doesn't change the fact that everyone who did call cost us more than our profit on the sale.
Customer support is expensive.
Microsoft used to charge for customer support back in the day (90s). The way it worked was that if it was your fault, you paid, if it was a product bug, there was no cost for support. While not a perfect system, it at least aligned everyone's incentives in the right direction. (The huge glaring flaw being it was MS that decided if they were going to charge you for the support call or not...)
I don't care if it's a human, a chatbot, or a dog if they fix my problem.
I don't want to contact customer support in the first place, if I'm forced to, it's because something is very wrong and in that case I don't want to be listening to elevator music and "your call is important to us, please hold" for an hour, and get my call disconnected forcing me to call again.
Issue is that I've yet to have a chatbot actually fix my issues, or most 1st contact human operators for that matter.
It matters less to me that the helper is an AI/human than the kind of help I'm getting.
The bigger problem to me is "help" is always framed as my needing to be educated, not a problem with the service. This is especially prevalent for technical customers who are legitimately trying to draw attention to a bug in the platform only to get how-to help articles pasted back to them.
Or technical customers with a case that was not handled properly. I'm thinking of long, long ago, ISP changed the Usenet server and didn't document it--not on their website, not with their tech support. It shouldn't have taken an hour and a third level support person to get we changed providers, here's the new address. First two levels simply couldn't comprehend that it was not a third party system that I was having trouble with.
The "figure out what you want to say" is key. I've started to think of LLMs, at least in a business setting, as misunderstanding amplifiers.
How many times at work have you been talking to someone else where they're using common words as jargon? Maybe it's something like "the online system" or "the platform". And it's perfectly clear to them what they mean, but everyone else in the company either doesn't know what that actually is, or they have a distorted idea based on the conventional definitions of the words. Even without LLMs in the mix, this can lead to people coming out of meetings with completely different understandings of what's going on.
My experience is few people are actually providing the relevant context to the LLM to explain what they mean in situations like this. Or they don't have the actual knowledge and are using the LLM in the hopes it'll fill in for their ignorance. The LLMs are RLHFed to sound confident, so they won't convey that they don't know what a piece of jargon means. Instead they'll use a combination of the common meaning and the rest of the context to invent something. When this gets copy/pasted and sent around, it causes everyone who isn't familiar to get the wrong idea. Hence "misunderstanding amplifier".
To the point of the article, this is soluble if people take the time to actually figure out what they are trying to convey. But if they did that, they wouldn't need the LLM in the first place.
And that people and the systems actually know the relevant terms.
I recently was dealing with the Amazon robot--after correctly identifying the items in the order it then proceeded to use short terms which were wrong, but make sense as what a classifier might have spit out. Instead of understanding being a shared thing it falls entirely on the user. Sufficiently adept user, this is fine. But a lot of users aren't sufficient adept.
I just signed up with Gusto for one of my companies. They charged me for premium support automatically and when I tried to dispute it I had to talk in circles with their AI named Gus. Why am I paying through the nose for premium support just to chat with an AI?
Gusto is a nightmare if your account needs fall out of their happy path. Everything is 100% automated with call center scripts to help you otherwise. You will never reach someone with power to fix anything.
As a customer, I just want the information I need. While I don't want to talk to a chatbot, I also don't want to talk to a human - and for the same reason: they usually don't have the info I need.
That's the aspect I don't understand. The information I want is almost always something some other customers have asked already. I'd much prefer to avoid their customer support maze entirely and help myself on a searchable wiki. Unfortunately, most company's online product support FAQs usually only contain answers to obvious shit on the order of RTFM and "is it plugged in." Why not just post the doc their advanced tier 3 support people share amongst themselves? It can be under a warning label like 'preliminary advanced info for engineers'.
I realize people like me represent only around 2-3% of the customers seeking support but it's 2-3% that is able to self-serve and takes more time than average because we invariably have to work through front-line support to get escalated to someone with the non-obvious info that's still been asked many times before. So maybe we're only ~2% but we suck up 4% of support bandwidth and we probably take up closer to ~20% of Tier 3 support - the most expensive, scarce type.
I mostly agree (although sometimes it is necessary to talk to someone about it); it would be better to actually have good documentation (so that you do not need to talk to someone about it).
A warning label like you mention is a possibility if that is considered to be necessary, although I think it might be better to have a file that you can download and read (or request by mail or telephone or fax, if this becomes necessary in some circumstances; do not assume the computer always works and is compatible with your file), instead of a searchable wiki.
On my team we always ship an agent draft with a short human anchor first. Two sentences that explain the motivation and the checks we ran, then the bot block with a label like “agent draft” for anyone who wants the raw output. That way readers know what we actually think and don’t have to guess whether the chat log is the human opinion. Do you have a checklist for when that human intro is enough versus when the whole thing needs to stay private?
Related: Please don't make me talk to your AI pretend-human complete with Asian accent and background call center sounds. That's even more insulting that a chat bot.
I view the issue of inefficient communication as a problem that will wane as LLMs progress, and a bit idealistic about the efficiency of most human-to-human communication. I feel strongly that we shouldn't be forced to interact with chatbots for a much simpler reason: it's rude. It's dismissive of the time and attention of the person on the other end; it demonstrates laziness or an inability to succeed without cutting corners, and it is an affront to the value of human interaction (regardless of efficiency).
> a bit idealistic about the efficiency of most human-to-human communication.
I don't know if I would call it idealism. I feel like what we're discovering is that while the efficiency of communication is important, the efficacy of communication is more important. And chatbots are far less reliable at communicating the important/relevant information correctly. It doesn't really matter how easy it is to send an email if the email simply says the wrong thing.
To your point though, it's just rude. I've already seen a few cases where people have been chastised for checking out of a conversation and effectively letting their chatbot engage for them. Those conversations revolved around respect and good faith, not efficiency (or even efficacy, for that matter).
I feel like that ship sailed long ago with phone trees and hour-long support wait times becoming normal. Not that it's an ideal state of affairs, but I'd much rather talk to a chatbot than wait for an hour for a human who doesn't want to talk to anyone, as long as that chatbot is empowered to solve my problem.
Have you ever had a chatbot solve your problem? I don't think this has ever happened to me.
As a reasonably technical user capable of using search, the only way this could really happen is if there was no web/app interface for something I wanted to do, but there was a chatbot/AI interface for it.
Perhaps companies will decide to go chatbot-first for these things, and perhaps customers will prefer that. But I doubt it to be honest - do people really want to use a fuzzy-logic CLI instead of a graphical interface? If not, why won't companies just get AI to implement the functionality in their other UIs?
Actually, I have, Amazon has an excellent one. I had a few exchanges with it, and it initiated a refund for me, it was much quicker than a normal customer service call.
Outside of customer service, I'm working on a website that has a huge amount of complexity to it, and would require a much larger interface than normal people would have patience for. So instead, those complex facets are exposed to an LLM as tools it can call, as appropriate based on a discussion with the user, and it can discuss the options with the user to help solve the UI discoverability problem.
I don't know yet if it's a good idea, but it does potentially solve one of the big issues with complex products - they can provide a simple interface to extreme complexity without overwhelming the user with an incredibly complex interface and demanding that they spend the time to learn it. Normally, designers handled this instead by just dumbing down every consumer facing product, and I'd love to see how users respond to this other setup.
I did think about this use-case as I was typing my first message.
I can see it working for complex products, for functionality I only want to use once in a blue moon. If it's something I'm doing regularly, I'd rather the LLM just tell me which submenu to find it in, or what command to type.
I'm happy that LLMs are encouraging people to add discoverable APIs to their products. Do you think you can make the endpoints public, so they can be used for automation without the LLM in the way?
If you need an LLM spin to convince management, maybe you can say something about "bring your own agent" and "openclaw", or something else along those lines?
Yeah as long as the chatbot is empowered to fix a bunch of basic problems I'm okay with them as the first line of support. The way support is setup nowadays humans are basically forced to be robots anyway, given a set of canned responses for each scenario and almost no latitude of their own. At least the robot responds instantly.
The problem is that people are very rude to customer service representatives, so companies spend money training CSRs, who often quit after a short period of being abused by customers. Automated reception systems disallow people from reaching representatives for the same reason.
CSRs are abused by call center managers far more often than they are by the people on the other end of the phone line. The endless push for "better" metrics, the terrible pay, the dehumanizing scripts, bad (or zero) training, optimizing to make every employee interchangeable and expendable, unforgiving attendance policies, treating workers like children, etc. Call centers are brutal environments and the reason churn is often so high has very little to do with abuse from the people calling for help. In fact, the last two call centers I had any insight into (to their credit) had strict policies about not taking abuse from customers and would flag abusive customer's accounts.
I find chatbot conversations to be incredibly similar to dreams.
It's human nature to want to share your dreams, because they are fascinating to you.
However, it's also human nature to want to punch someone in the face when they start talking about this crazy dream they had last night ... because it has nothing to do with you, and doesn't interest you at all.
Similarly, when an AI says something useful to you, in response to your prompts, it's very particular to you. When you try to share it with others ... you get the article.
The worst part isn't that chatbots are bad at their job. It's that teams shipping them genuinely don't know how bad they are. Nobody's reading the conversations, so nobody catches users hitting dead ends, rage-clicking, or dropping off right after a failed session. The data exists, it just sits there unread.
People want to spend as little as possible while getting support for their product as long as possible.
Companies want people to spend as much as possible while doing the minimum work on the product.
Chatbots let companies spend almost nothing while pretending to provide long-term support.
I wonder if something similar to a copyleft license could help. What if there was a contractual "fair business" pledge that companies could add? I imagine that good enough lawyers could craft something that essentially said, "You can only display this contract if you legally guarantee that you do X, Y, Z and do not do A, B C."
I thought this was going to be about (customer support) chatbots, which can be a good thing.
"Don't make me talk to your [customer support] chatbot" reads like "Don't make me go to an ATM for a cash withdrawal". If I can solve a thing quickly and effectively without waiting forever to speak to an overworked customer support agent on another contitent, I would very much like that!
Well, anyways, the post is not about that. It's about posting AI-generated text (blog posts, PR summaries). Which I agree with, although there are a bunch of holes in the argument, such as:
> 1. Figure out what you want to say. 2. Say it. That first figuring-out part is important.
Well, yeah, I can figure out what I want to say, then have the chatbot say it. So looks like the second part is important, too.
The key thing here is not whether it's AI. The key thing is quality and signal. No one wants to read to a low quality human comment either.
If the AI output was actually better than talking to a real human, more useful, more concise, serving the job to be done, then no one would have a problem with it. In fact they would appreciate it. That future is not here in many areas.
The problem is people are wielding AI right now and either [a] the models they are using are not good enough, [b] they aren't being given enough context, or [c] they are deployed in a way that makes it sloppy
(Insert joke about whether this comment is AI. It's not, but joke away)
No. It doesn’t matter how good an llm model is. If a person has something to say and they can give the llm enough context to say it well, they should just write it themselves. Theres 0 reason to bring a llm into it. Doing so simply makes your writing less trustworthy because as a reader I don’t know if what I’m reading is genuine from the writer or simply average of all texts filler.
I disagree. If my colleague can't be bothered to write a PR comment themselves then I can't be bothered to read it. If I can gain the same insights from interfacing an LLM directly then there's no point in this intermediary dance.
> The key thing here is not whether it's AI. The key thing is quality and signal. No one wants to read to a low quality human comment either.
This is so obviously true to intelligent people ... it's sad that you're getting downvoted.
The OP wrote
> When I talk to a person, I expect that they are telling me things out of their head — that they have developed a belief and are trying to communicate it to me.
But when I'm having a conversation about a subject (rather than with a friend, partner, or other person with whom I have a relationship and the conversation is part of the having of that relationship) I don't care what is in that person's head, I care about the truth of the matter, so I'm far more interested in their sources, their logic and the validity of same. Unless I'm a psychologist doing a survey, why should I care about some random person's beliefs? Since I'm a truth seeker, I care about their arguments, and of course the quality of their arguments is of paramount importance. I appreciate people who can back up their arguments, and LLM summaries that are chock full of facts gleaned from the massive training data that includes a vast amount of human knowledge are fully appreciated--while being aware that hallucination is possible so I often double check things regardless of the source. OTOH, the pushback to this is from people I consider worse than irrelevant--they not only are willfully ignorant but they reject knowledge seeking for irrational ideological reasons. (I myself see the LLM industry to be extremely problematic, but as long as LLMs exist and are capable of producing quality signal--which is the given here--then I will use them.)
No it isn't. I really do not care what the LLM has to say. If a person has taken the (substantial) time necessary to fill the context with enough information that something interesting comes out, I would much rather they simply give me the inputs. The middleman is just digested Internet text. I've already got one of those on my end.
That does somewhat depend on the size of the context.
LLMs won't add information to context, so if the output is larger than the input then it's slop. They're much better at picking information out of context. If I have a corpus of information and prompt an extraction, the result may well contain more information than the prompt. It's not necessarily feasible to transfer the entire context, and also I've curated that specific result as suitably conveying the message I intend to convey.
This does all take effort.
My take is also that I am interested in what people say: I have priors for how worthwhile I expect it to be to read stuff written by various people, and I will update my priors when they give me things to read. If they give me slop, that's going to affect what I think of them, and I expect the same in return. I'm willing to work quite hard to avoid asking my colleagues to read or review slop.
I don't know, occasionally there are some funny results. For instance, I have managed to get AWS' support bot to start smack talking their platform and criticising its often needlessly complex and sometimes incoherent design before cheekily offering to help me make my relative simple setup even more complex and enterprise-ready.
I guess part of the advantage of being an extremely long-winded writer who makes lots of typos is that people know that what I'm writing is probably written by a human.
Though maybe people will start supplying context like "no em dashes, and occasionally misspell a word or two", and soon you won't even be able to tell that.
Typos and EM-dashes are not just obvious. They are avoidable. Here's the kicker: Even if you remove those, other AI smells don't just exist. They are obvious.
Maybe, though maybe we're only noticing it when it's obvious? Sort of a survivorship bias thing?
Who knows how much of the comments on any website are written by humans now; yeah there are plenty of tells so it can be obvious, but that might only be for the exceptionally bad posts.
Yeah, I've also see a lot of "Not X, not Y, just Z" style posts on LinkedIn.
LinkedIn has always been a place full of low-effort posts for people trying to self-promote, so I guess it makes sense to have a robot actually do the thinking for something that is and always has been inherently mindless.
Yeah a trend that I've noticed in online comments is people taking LLM generated text and just removing punctuation and making it all lowercase. It's like dude, it's still so obvious xD
I have a shorter, more cynical version of this: if a person doesn’t provide enough input to a chatbot, I’d be better off talking to the chatbot directly.
Reading almost all the comments gives me the sad validation that people truly do not read the article before commenting...
This article is not about support chatbots. It's about clearing up your writing/thoughts and communicating clearly even if you used a chatbot to get there.
It's absolutely miserable isn't it. People see a headline, decide they have an important thing they must tell the world and just blurt it out. Imagine doing this in conversation. You'd overhear a fragment of conversation and just interject with some semi related bullshit that makes you feel smart and then leave.
People on HN spend most of their time congratulating themselves on how they aren't like other social media platforms, but do the exact same crap anyways.
When it comes to technical discussions, there are so many people on here just regurgitating what they read on an earlier thread. Maybe to test if what they heard was true. Maybe because they just want to sound smart. Not a lot of people actually trying things.
I have noticed that my writing ability has atrophied since I was writing essays in school. Now at work most of my writing is slack messages. Writing longer more thoughtful pieces about strategy or performance review has become a slog. I suspect that a lot of people have had a similar experience so offloading the pain of writing to an LLM is appealing.
But frankly LLMs suck at writing. It's not only formulaic, it's uninspired!! So I worry that we're entering an era of mediocre writing. I like the "Have you considered writing?" suggestion. I've been trying to make a habit of writing book reviews so I can counter some of the writing atrophy I've developed. Hopefully it will help me become a better thinker too. As Ray says here: "Understanding your own point of view is an enriching exercise."
I'm reminded of the Air Canada customer service chatbot. It completely made up a refund policy (and there are still people on HN who insist LLMs don't hallucinate) and a court ruled the company had to honor it [1].
The only way to deal with this is to make the implentation not worth it by constantly bypassing it to speak to a human and/or making it cost money by getting it to give you things you're not otherwise entitled to.
I really wonder how these things will handle prompt injection and similar things. I have no confidence any of this is secure.
Wait until this comes to healthcare and it'll be chatbots handling prior authorization calls, wasting even more physician time.
There's been a lot of "the world doesn't work the way I want it to" on HN recently. I suspect this is a function of an aging readership more than anything particularly groundbreaking about hot takes on the up and coming tech.
"Anything invented after you're thirty-five is against the natural order of things." Douglas Adams
When I worked at Microsoft, it cost over $20 to have a human customer support agent pick up the phone when someone called in for help. That was greater than our product margin. Every time someone called for help, we basically lost the entire profit on that sale, and then some.
Most common support calls where for things that were explained in the manual, the out of box experience, tutorial documents, FAQ pages, and so on and so forth.
Did we have actual support issues that needed fixing, yes of course. And the insanely high cost of customer support drove us to improve our first use experience. But holy cow people don't realize how expensive support calls are.
Edit: To explain some of the costs - This was back when people worked in physical call centers, so first off we were paying for physical office space. Next up training, each CSR had to be trained on our product. This took time and we had to pay for that training time. We also had to write support material, and update that support material for each new version that came out. All of this gets amortized into the cost of support. Because workers tend not to stay long, you pay for a lot of training.
Add in all the other costs associated with running a call center and the cost per call, even for off shore call centers, is not cheap.
In a reasonable world we'd just raise the price of the product by $x based on what % of people we expect to call in for support (ignore for a minute that estimating that number is hard), but the world isn't reasonable. Downwards price pressure comes from all sides, primarily VC backed competitors who are OK burning $$ to gain market share, and competitors at other FAANGs that are OK burning money to gain market share.
The result is that everyone is going to try and reduce support costs because holy cow per user margins are low now days for huge swaths of product categories (Apple's iPhone being a notable exception...)
This doesn't seem like a bad thing when it comes to aligning incentives (assuming customers actually want a product they don't need help to use).
And if it turns out to be your mistake (faulty product or missing documentation) as opposed to something the user could have reasonably solved by themselves, refund the charge and possibly provide compensation for the inconvenience.
I doubt it. I suspect the number one tech support call is "I forgot my password" and everything else is a long way below that.
I'll slag on Microslop all day, but users are dumber than dumb.
We product makers get to think about our one little product all day, and it's our job to make our product work for the "dumb" users. It's not their job to adapt to us.
That being said. Your example of customers calling for support on things they shpuld be capable of figuring out themselves in is probably where AI is going to shine as first line support. Once (if?) AI voice chat is good enough to replace chatbots we may not even realize we're talking with an AI unless it tells us.
It certainly won't be cheap to run real-time AI voice chat, or any real-time AI chat. The AI costs that you currently see are heavily subsidized, just like OP's example of "VC backed competitors who are OK burning $$ to gain market share", it's the same. These AI companies are far from profitable, burning billions to insert themselves into customer support pipelines and everywhere else they can, and then the other foot will drop. Uber and Lyft are far more expensive today than when they started, and the price to run "AI" will also inflate when these companies have to pay off all the billions they've spent but didn't earn. I doubt it will end up costing much less if less at all than human support, with worse results.
obviously not a problem with the technology itself, it was like that with more primitive answering machines as well, often there only to either answer the obvious things, or stonewall people with real problems with the product or service hoping they'd just give up and take the loss
"We are experiencing an greater than usual call volume, please wait while an agent becomes available" only to be randomly disconnected has been a thing for most of my life.
Everyone seems to be hyping open claw at the moment soon its just going to be LLMs talking to LLMs.... I wonder if they will develop a short hand and start talking in wingdings.
I would think that's close to an hourly rate for first level support and calls are mostly resolved in ~10 mins?
Then you also have to pay them regardless of whether someone calls.
A rough rule of thumb is the full burdened cost of an hourly office knowledge worker is two to three times the gross hourly wage.
All alternatives which are capable of actually serving the customer are systematically driven out of business.
Had they built a better, more intuitive product, they would get fewer support calls and wouldn't be struggling with costs.
As I mentioned, due to high support costs we worked to improve the UX and we ended up dropping our support costs dramatically.
Doesn't change the fact that everyone who did call cost us more than our profit on the sale.
Customer support is expensive.
Microsoft used to charge for customer support back in the day (90s). The way it worked was that if it was your fault, you paid, if it was a product bug, there was no cost for support. While not a perfect system, it at least aligned everyone's incentives in the right direction. (The huge glaring flaw being it was MS that decided if they were going to charge you for the support call or not...)
I don't want to contact customer support in the first place, if I'm forced to, it's because something is very wrong and in that case I don't want to be listening to elevator music and "your call is important to us, please hold" for an hour, and get my call disconnected forcing me to call again.
Issue is that I've yet to have a chatbot actually fix my issues, or most 1st contact human operators for that matter.
The bigger problem to me is "help" is always framed as my needing to be educated, not a problem with the service. This is especially prevalent for technical customers who are legitimately trying to draw attention to a bug in the platform only to get how-to help articles pasted back to them.
How many times at work have you been talking to someone else where they're using common words as jargon? Maybe it's something like "the online system" or "the platform". And it's perfectly clear to them what they mean, but everyone else in the company either doesn't know what that actually is, or they have a distorted idea based on the conventional definitions of the words. Even without LLMs in the mix, this can lead to people coming out of meetings with completely different understandings of what's going on.
My experience is few people are actually providing the relevant context to the LLM to explain what they mean in situations like this. Or they don't have the actual knowledge and are using the LLM in the hopes it'll fill in for their ignorance. The LLMs are RLHFed to sound confident, so they won't convey that they don't know what a piece of jargon means. Instead they'll use a combination of the common meaning and the rest of the context to invent something. When this gets copy/pasted and sent around, it causes everyone who isn't familiar to get the wrong idea. Hence "misunderstanding amplifier".
To the point of the article, this is soluble if people take the time to actually figure out what they are trying to convey. But if they did that, they wouldn't need the LLM in the first place.
I recently was dealing with the Amazon robot--after correctly identifying the items in the order it then proceeded to use short terms which were wrong, but make sense as what a classifier might have spit out. Instead of understanding being a shared thing it falls entirely on the user. Sufficiently adept user, this is fine. But a lot of users aren't sufficient adept.
That's the aspect I don't understand. The information I want is almost always something some other customers have asked already. I'd much prefer to avoid their customer support maze entirely and help myself on a searchable wiki. Unfortunately, most company's online product support FAQs usually only contain answers to obvious shit on the order of RTFM and "is it plugged in." Why not just post the doc their advanced tier 3 support people share amongst themselves? It can be under a warning label like 'preliminary advanced info for engineers'.
I realize people like me represent only around 2-3% of the customers seeking support but it's 2-3% that is able to self-serve and takes more time than average because we invariably have to work through front-line support to get escalated to someone with the non-obvious info that's still been asked many times before. So maybe we're only ~2% but we suck up 4% of support bandwidth and we probably take up closer to ~20% of Tier 3 support - the most expensive, scarce type.
A warning label like you mention is a possibility if that is considered to be necessary, although I think it might be better to have a file that you can download and read (or request by mail or telephone or fax, if this becomes necessary in some circumstances; do not assume the computer always works and is compatible with your file), instead of a searchable wiki.
The different accents and call center background noise are features in their product.
I don't know if I would call it idealism. I feel like what we're discovering is that while the efficiency of communication is important, the efficacy of communication is more important. And chatbots are far less reliable at communicating the important/relevant information correctly. It doesn't really matter how easy it is to send an email if the email simply says the wrong thing.
To your point though, it's just rude. I've already seen a few cases where people have been chastised for checking out of a conversation and effectively letting their chatbot engage for them. Those conversations revolved around respect and good faith, not efficiency (or even efficacy, for that matter).
As a reasonably technical user capable of using search, the only way this could really happen is if there was no web/app interface for something I wanted to do, but there was a chatbot/AI interface for it.
Perhaps companies will decide to go chatbot-first for these things, and perhaps customers will prefer that. But I doubt it to be honest - do people really want to use a fuzzy-logic CLI instead of a graphical interface? If not, why won't companies just get AI to implement the functionality in their other UIs?
Outside of customer service, I'm working on a website that has a huge amount of complexity to it, and would require a much larger interface than normal people would have patience for. So instead, those complex facets are exposed to an LLM as tools it can call, as appropriate based on a discussion with the user, and it can discuss the options with the user to help solve the UI discoverability problem.
I don't know yet if it's a good idea, but it does potentially solve one of the big issues with complex products - they can provide a simple interface to extreme complexity without overwhelming the user with an incredibly complex interface and demanding that they spend the time to learn it. Normally, designers handled this instead by just dumbing down every consumer facing product, and I'd love to see how users respond to this other setup.
I can see it working for complex products, for functionality I only want to use once in a blue moon. If it's something I'm doing regularly, I'd rather the LLM just tell me which submenu to find it in, or what command to type.
If you need an LLM spin to convince management, maybe you can say something about "bring your own agent" and "openclaw", or something else along those lines?
It's human nature to want to share your dreams, because they are fascinating to you.
However, it's also human nature to want to punch someone in the face when they start talking about this crazy dream they had last night ... because it has nothing to do with you, and doesn't interest you at all.
Similarly, when an AI says something useful to you, in response to your prompts, it's very particular to you. When you try to share it with others ... you get the article.
Companies want people to spend as much as possible while doing the minimum work on the product.
Chatbots let companies spend almost nothing while pretending to provide long-term support.
I wonder if something similar to a copyleft license could help. What if there was a contractual "fair business" pledge that companies could add? I imagine that good enough lawyers could craft something that essentially said, "You can only display this contract if you legally guarantee that you do X, Y, Z and do not do A, B C."
Don't make me talk to a chatbot while there is zero forward progress in solving the problem.
"Don't make me talk to your [customer support] chatbot" reads like "Don't make me go to an ATM for a cash withdrawal". If I can solve a thing quickly and effectively without waiting forever to speak to an overworked customer support agent on another contitent, I would very much like that!
Well, anyways, the post is not about that. It's about posting AI-generated text (blog posts, PR summaries). Which I agree with, although there are a bunch of holes in the argument, such as:
> 1. Figure out what you want to say. 2. Say it. That first figuring-out part is important.
Well, yeah, I can figure out what I want to say, then have the chatbot say it. So looks like the second part is important, too.
If the AI output was actually better than talking to a real human, more useful, more concise, serving the job to be done, then no one would have a problem with it. In fact they would appreciate it. That future is not here in many areas.
The problem is people are wielding AI right now and either [a] the models they are using are not good enough, [b] they aren't being given enough context, or [c] they are deployed in a way that makes it sloppy
(Insert joke about whether this comment is AI. It's not, but joke away)
This is so obviously true to intelligent people ... it's sad that you're getting downvoted.
The OP wrote
> When I talk to a person, I expect that they are telling me things out of their head — that they have developed a belief and are trying to communicate it to me.
But when I'm having a conversation about a subject (rather than with a friend, partner, or other person with whom I have a relationship and the conversation is part of the having of that relationship) I don't care what is in that person's head, I care about the truth of the matter, so I'm far more interested in their sources, their logic and the validity of same. Unless I'm a psychologist doing a survey, why should I care about some random person's beliefs? Since I'm a truth seeker, I care about their arguments, and of course the quality of their arguments is of paramount importance. I appreciate people who can back up their arguments, and LLM summaries that are chock full of facts gleaned from the massive training data that includes a vast amount of human knowledge are fully appreciated--while being aware that hallucination is possible so I often double check things regardless of the source. OTOH, the pushback to this is from people I consider worse than irrelevant--they not only are willfully ignorant but they reject knowledge seeking for irrational ideological reasons. (I myself see the LLM industry to be extremely problematic, but as long as LLMs exist and are capable of producing quality signal--which is the given here--then I will use them.)
(I could have sworn there was a popular HN submission a while back of this or a similar blog post, but damned if I can find it now.)
LLMs won't add information to context, so if the output is larger than the input then it's slop. They're much better at picking information out of context. If I have a corpus of information and prompt an extraction, the result may well contain more information than the prompt. It's not necessarily feasible to transfer the entire context, and also I've curated that specific result as suitably conveying the message I intend to convey.
This does all take effort.
My take is also that I am interested in what people say: I have priors for how worthwhile I expect it to be to read stuff written by various people, and I will update my priors when they give me things to read. If they give me slop, that's going to affect what I think of them, and I expect the same in return. I'm willing to work quite hard to avoid asking my colleagues to read or review slop.
Though maybe people will start supplying context like "no em dashes, and occasionally misspell a word or two", and soon you won't even be able to tell that.
Who knows how much of the comments on any website are written by humans now; yeah there are plenty of tells so it can be obvious, but that might only be for the exceptionally bad posts.
Checked LinkedIn and found four posts in a row that had "here's the kicker".
LinkedIn has always been a place full of low-effort posts for people trying to self-promote, so I guess it makes sense to have a robot actually do the thinking for something that is and always has been inherently mindless.
Forcing a customer anything beyond that is RUDE!
This article is not about support chatbots. It's about clearing up your writing/thoughts and communicating clearly even if you used a chatbot to get there.
When it comes to technical discussions, there are so many people on here just regurgitating what they read on an earlier thread. Maybe to test if what they heard was true. Maybe because they just want to sound smart. Not a lot of people actually trying things.
But frankly LLMs suck at writing. It's not only formulaic, it's uninspired!! So I worry that we're entering an era of mediocre writing. I like the "Have you considered writing?" suggestion. I've been trying to make a habit of writing book reviews so I can counter some of the writing atrophy I've developed. Hopefully it will help me become a better thinker too. As Ray says here: "Understanding your own point of view is an enriching exercise."
Seems to me like you're doing fine so far. (I hope I haven't just been letting my standards go down the drain...)
> It's not only formulaic, it's uninspired!
Heh.
The only way to deal with this is to make the implentation not worth it by constantly bypassing it to speak to a human and/or making it cost money by getting it to give you things you're not otherwise entitled to.
I really wonder how these things will handle prompt injection and similar things. I have no confidence any of this is secure.
Wait until this comes to healthcare and it'll be chatbots handling prior authorization calls, wasting even more physician time.
[1]: https://www.wired.com/story/air-canada-chatbot-refund-policy...
"Anything invented after you're thirty-five is against the natural order of things." Douglas Adams