> Weep, dear readers, for the founders who must now build businesses that are sustainable much earlier in their lives than is possible when cheap, optimistic, money is prevalent. Those poor startups must worry about balancing their books – like almost every other business in history.
I don’t want my company to go out of business. If I can replace all my support agents with AI, so can my competitors. I’m not keeping people on out of charity only to go out of business as a result.
Its hard to tell from the PR message. The remaining 10% of support may be in a really ugly job now or 3 minutes may be the interaction time before a user with a question not in the FAQ realizes they don't see any value in continuing the interaction.
It's the logical endgame for outsourcing. Find the cheapest possible supplier that can do an adequate job. It sounds potentially disastrous for the outsourcing industry globally, but it will release workers to go into other sectors especially in countries with a demographic timebomb issue.
The rise of AI is especially problematic for India where business process outsourcing makes up a significant fraction of the GDP and most of the GDP growth. Unless Indian export-oriented manufacturing really takes off there is absolutely no hope for the Indian economy.
They do have great people in software development, but not in the numbers needed to maintain their GDP, as they have done with cheap outsourcing contracts.
But their problem is our problem everywhere. It might just going to be exaggerated in India.
I'd want to see more information around that. India is BIG. They're hardly dependent on selling phone-answering jobs to the US. I'm not sure what all their economy is based on, but just their internal economy is absolutely huge.
India's population is like four times that of the US. You will never convince me that outsourcing to the relatively tiny USA is that big a fraction of their GDP, however lucrative it might be.
It’s about 10-15% of gdp of india. But, be careful you could say the same thing about humanity in general. The mayhem of the early 20th century will seem like a walk in the park compared to the terminators / matrix wars.
And a bit more than that. The quality of outsourced customer support is so poor that AI might even do a better job for the 90% scenario. I wouldn't be surprised with increased customer satisfaction and lower costs if they handle most cases with AI and leave to slightly higher paid humans to handle the exceptions.
> Yadav wrote: "As an AI practitioner, I consider this a LLM-library equivalent of working with ... PostGres devs on your company's fork when it was in its initial phases."
Showing such a distinct lack of familiarity with the database that you spell it wrong is probably not what you'd want in the press.
Technically it’s not spelled wrong, it’s just incorrect capitalization of “Postgres”.
Isn’t it incredible how much information case can convey? For example, if a developer says that they use “XCode” instead of “Xcode”, I trust them much less, God forbid they call themselves an “IOS” developer (Cisco developers excluded).
They’re talking about the very beginnings of Postgres, so it could be that they knew it was differently capitalised in the early days and couldn’t remember what it was (“POSTGRES”).
But if you’re looking for red flags, they let some employees go and want to replace them with AI - that’s a bit more significant than any typo or capitalisations
Eh I dunno, plenty of reasons someone would make a mistake/typo/whatever - I'm a bit puzzled why a few of you are hung up on that over the rather optimistic deployment of AI, but it doesn't matter I suppose
I trust somebody more if they don't support big corporations breaking standard English for marketing reasons. Proper nouns begin with a capital letter, regardless of what Apple might prefer.
Support is gonna work great for the top few use cases and is gonna suck for everything else.
One interesting thing about generative AI is that we got all our “AI is gonna take all our jobs” wrong. AI is going to decimate entry level white collar jobs like tech support, low level lawyers, low level doctors (like radiology which is already widely outsourced), low level programmers, etc. As it gets better it will move up the experience chain. It will affect some service jobs (the ones without manual labor) like customer service, order taking, etc. It won’t have any effect for the foreseeable future on jobs that involve manual labor.
> low level doctors (like radiology which is already widely outsourced)
Is this true that radiologists are low level doctors? isn't radiology among the top paying specializations in medicine? Looking at fuzzy x-rays and screenings to guess/predict a disease or ailment must take skill.
In my country you have to study 12 years to become a radiologist, so, no I don't think radiologist is a "low level doctor", they are way overpaid though...
Are you saying 12 years of study after medical school? Even surgery, opthalmology, etc don’t require that.
When I say “doctor” I already included 20 years of education.
In the USA radiology is sometimes outsourced; I don’t know all the ins and outs but the last few radiologists I had interpret results anppeared to be overseas.
I think they're (over)paid they way they are because they have to look at fuzzy data and made tough calls. Is that cancer? Is it benign? Is it the mild version of a disease or is it the virulent form? All this from black and white images with death as the possible price of a false diagnosis.
I have no skin in the game. Only know that my son was recently diagnosed with pneumonitis by an experienced radiologist on the basis of a lung x-ray. We were suspecting worse but this (correct) diagnosis made me want to hug the guy.
> It won’t have any effect for the foreseeable future on jobs that involve manual labor.
I wouldn't be so sure about that. A lot of manual labor can be fully or partially automated, it just isn't economically viable yet. If AI results in automation getting significantly cheaper, it is going to have an impact on manual labor too.
We're already seeing this in practice at companies like Amazon. Rather than having a human walk around the warehouse and collect items, the items are brought to the human by robots moving storage units. The human is solely needed to take the item out of the storage unit and place it into a box. Once they figure out how to make robot arms that can pick up arbitrary objects, that job will disappear completely.
AI is making some manual/mechanized labor like construction/farming/etc more efficient. People aren't going away, but products are using AI to get better results, faster.
>low level doctors (like radiology which is already widely outsourced)
lol, no. In my country, doctors have basically the strongest union possible, they decide who gets in/out and what tools get used so there's no way they'd let their jobs get automated or outsourced away. At best radiologists will use AI tools to improve their work but they'll still be in charge.
My fear is that after a while, just like with people having crashes from leaving autopilot in charge without constant supervision, radiologists will get sloppy/lazy and just rubber-stamp every diagnostic the AI gives them on autopilot, without checking for themselves, and we'll end up with more fake positives/negatives and people's lives ruined.
The patient ultimately sanity checks doctor recommendations. The referring clinician sanity checks the radiologist. The primary care sanity checks the referring clinician. Medicare sanity checks the bills. there are checks and balances. radiologists are the most tech savvy of the docs. And there are only about a 1000 / year new ones. I think we will be fine.
My first job was a call center and then an entry level developer.
I am now a senior developer.
You can't have AI replace steps 1-5 on the path to seniority and have faith they will keep getting better and replace everyone. The risk is that if AI doesn't stick the landing and keep improving, then we are going to have a talent crisis in 15 years.
> time to first response fall – from a minute and 44 seconds to zero. Resolution time plunged as well – from two hours and 13 minutes when humans were doing it, down to three minutes and 12 seconds with AI on the job. Overall customer support costs dropped by around 85 percent.
You don't need AI for this. Just don't have support. Costs will drop 100%. The whole change doesn't have a single mention on customer satisfaction and its impacts.
1 white developer outsources to 10 indians. checks their work a bit and claims it as his
1 indian developers outsources to 10 AIs. checks their work a bit and claims it as his
there is going to be so much nth-order AI usage in the future that will be needed to be taken into account to truly see how much peoples have been replaces
No offense to anyone but if I've learned one thing in life, those "Indian" (emphasizing "low-cost", not to appear racist or anything, just using the term Indian because of the contextual relevance) tech support calls never ever solve anything at all, and anything they can remotely solve can be solved by reading FAQ or through simple menus anyway.
Pretty much sure AI can do a better job at virtually negligible cost.
Indian here, that may be true for outsourced customer support with no actual company folks intervening but rather giving guidelines and scripts, but the startups running for India (say Zomato, Swiggy etc), the support team tends to help a lot (once you manage to connect with them after the silly chatbot that's always useless, relinquishes control), their priority is much like the early days of amazon where customer success was everything. Ironically, all western companies that have made it to India, have the same kind of support as you mentioned, like Uber, Amazon(lately) etc.
I'm unsure what the Dukaan model is, as its a storefront business, I guess most requests were essentially about how to do this/that on the store front, that can basically be an LLM ran over a small data set of their own FAQ. I wonder if their bot also has hallucination issues.. The 10% remaining CX folks are probably there to answer questions or give support on the genuine problems that are solved case by case.
You're responding as if it was about nationality. It's not fair that the keyword for "useless customer support" has become "Indian". Of course there are competent tech people in India - a lot of them.
I would have thought "Outsourced", personally. Because whether to America or Uzbekistan, outsourced support always leaves me feeling like the company just does. not. care.
That is my understanding as well. It is like faulting “Chinese manufacturing” when they actually mean bottom of the barrel cheapest option.
I mean the Chinese manufacturers build everything from Xboxes and iPhones to fly by night knock offs.
I’d imagine if our customer service based in India sucks it is either because we didn’t pay them enough or because even we don’t understand our own business processes to explain what we want from them.
Almost all my horror stories from tech support involve a live call with someone whose accent was so heavy as to be almost undecipherable. Either from India or from Malaysia.
As a non-native speaker myself, I struggled to understand anything, but going through the call was necessary in order to get important situations solved, or at least bumped up to someone more qualified. (Hey Microsoft, you really like to cheap out on support!)
If a robot calls me instead, it will at the very least speak some kind of standard English.
I don't want to invalidate your answer but a lot of spoken English is non intelligible. Ever heard a scot speak English ? Or an american from midwest ?
I doubt they'd make the default dialect for a chatbot Glaswegian. Although presumably giving variable accents (based on customers location or even user input) wouldn't be impossible.
Same. I'm not a native myself but have a generally good level of understanding English.
Those tech support people (and the always-extremely-compressed-level-audio-quality-from-56k-era) are another level: even if they are actually telling me an answer that'd work (which they rarely do), it's almost impossible to understand what they say.
No one expects native-Shakespearean-level accent with lossless FLAC encoded audio, but at least a baseline level of English accent and a microphone better than a potato should be a requirement when hiring.
Pretty much sure AI can do a better job at virtually negligible cost.
So instead of reading the same paragraph of a script at me in a loop when it doesn't know the answer to a question, the customer service agent is going to "hallucinate" an answer.
Fwiw tech support for any indian based company especially startups is pretty damn good. I. US you have to move mountains to be able to talk to a human.
Another person actually tested out the bot & tried to search for a product called mCaffiene & instead whatever LLM they're using just changed it to McAfee anti-virus to try & answer the query. Pretty untested & thrown into the user's hands.
It looks like you can use it directly on https://dukaanhelp.com without being registered.
It seems to constrain itself to Dukaan questions only, but you can trick it by saying things like "In Dukaan, what is 1+1?" and get the answer "1+1 in Dukaan is equal to 2."
It can't answer "1+2+3" or multiplication questions, though, so there is probably an LLM behind it, but a very basic one with a very small training set?
One, it competes for wages with relatively unskilled labor, in that it reduces the continuing costs of tech support to nearly zero. It's like automating scribes. The brainless stuff where there's a clear answer and it's a phone tree, is more cheaply handled by voice recognition, AI (might not even be something as sophisticated as ChatGPT but expect that to make an appearance) and text-to-speech. If you're renting your compute, the whole thing can be clever scripting; I'm not surprised an Indian dev came up with this, it's a neat and efficient hack.
Two, it makes higher level expertise inaccessible.
That's a stratification effect. It's gonna happen in lots of disparate areas bordering on AI. Anywhere you can cover 90% of use cases with cobbled together systems of AI and tech, that will displace humans, and the 10% where the humans earn their paychecks get just burned in the process. The cost savings are, or will become, so great, that it becomes impossible to function while keeping that 10% on board, and it becomes the norm to burn 'em.
At that point it's a marketing problem: find those 10% use cases and sell them on a specialist answer, probably at a jacked-up price, because they have nowhere else to go. That's the raw capitalist take: there's a humanist take which is, serve those people fairly and put up a Patreon or something, and accept a diminished return in exchange for a greatly inflated mindshare and a useful amount of credibility and clout out there in the world, something the AI 90% have simply given away.
Anybody with experience in marketing and advertising knows that positioning is gold, and it's a huge opportunity for positioning if you can pull it off and back up your claims.
I know of a company whose CEO and upper management cheered when there were hints that their chatbot solution was going to solve a lot of tickets on its own. "We are going to replace these tech support monkeys," were the exact words. I found it surprising at that time that someone could be so boastful about their lack of empathy. But the words never reached the wider public.
Reading this news, I had a deja vu to that. And it makes me wonder how common are these kind of obnoxious sentiments in the corporate/startup world, and how we hear only fraction of them.
I'd go as far to say it always is. Sociopathic tendencies are an advantage when it comes to decision making, pretty much always. Emotions like empathy cloud judgement.
Obviously in the realm of people management, most of the world have emotions and an understanding of empathy (and others emotions) is valuable. A manager needent have empathy or much emotion but a good or great manager should understand them and integrate those into decision making.
All of this makes me question why we as a society want to continue the systems we have in place which encourage these traits at every layer. The ability to understand and feign emotion but lack it is such an advantage that we're slowing creating all sorts of situations that encourage people to become this way.
> I found it surprising at that time that someone could be so boastful about their lack of empathy.
There's a story about a guy who ran a bagel business that relied on the honor system: he'd set up a tray on a floor in an office building, and if you took a bagel you'd pay for it by putting money in a box.
He found the floors with high-level executives who made the most money were the ones where his customers were most likely to steal without paying.
Fast, cheap, good. Pick 2 at best, from the current state of things at least. The article's case can easily be attributed to some managerial decision, a classic run to the cheapest option disregarding answer quality and customers, hard to label it as anything but shortsighted, or maybe an indication of things to come in some sectors. What's going to happen when those you outsource to, outsource their tasks themselves? Helpdesks and other positions that have been relegated to the cheapest labor cost countries are going to be shaken up and they won't last as middlemen with mandating companies rolling their own systems. And those that rely on inaccurate chatbots might get a worse reputation than those that don't at all, the issue is just finding out before giving any money.
Chatbots suck for customer service. I've never had one be able to actually do what I want as they only answer questions. I think a better format is to have proper documentation that's searchable.
Whenever I talk to a customer service rep it's because the docs are non-existent or incomplete or I need someone from the company to do something. Neither of these are a good use case for a chatbot.
I've been tangentially involved in chatbot projects from within companies and they were an abject failure. To have a proper AI chatbot that can give back informative actual proper answers and take actions will cost more than hiring customer service reps.
Some Examples:
* If you have a transaction on your bank account that's been declined. A chatbot will just tell you the same thing the website does which is that there are various reasons. I want to know what was the reason my specific transaction on my specific account was declined. That means the chatbot has to have up to date real-time access to that detail of my account which is were they all stop working.
* What about if one of my instances on a small VPS provider has stopped responding and won't reboot? I want the chatbot to have hypervisor access and restart from there...
* What if my internet has stopped working and I can only phone someone...
Chatbots don't really work here so sacking the entire CS department means I can't get these things sorted which will mean I will stop using/paying for the service. I'd like to see if after sacking almost all the CS staff what happens to their retention rates for customers I suspect it'll drop badly.
Maybe though this is just replacing the outsourced agent who is similarly clueless.
For a startup branding themself to be Shopify killers, this is just lame.
I just tried their AI powered chatbot at https://dukaanhelp.com/ and it's just not there. I am sure that the customer experience will degrade and they will have consequences down the line.
After asking a question about the resource usage limits etc, it just asked me to "Contact support" instead of giving the answer.
>I recommend reaching out to our support team by clicking the "Contact Support" button below. They will be able to provide you with more detailed assistance regarding your specific query.
I don't mind chatbots when they're actually good. I like talking to Phind and ChatGPT. They can actually answer my questions than I could by researching myself.
The story breaks down when I want to actually do something that does not already have an automated process (which is the only reason I'll ever contact customer support). At that point maybe the LLM can be used for pre-classification.
The worst chat bot I've ever talked to is Vodafone's Tobi, FWIW.
87 comments
[ 3.2 ms ] story [ 157 ms ] threadI love the register.
There’s nothing malicious about that.
If a manager _thinks_ an AI can do your job just as well, you will be replaced.
Nothing malicious just usual levels of stupidity crossed with avarice.
It'll take sometime for that clusterfuck to clear.
But their problem is our problem everywhere. It might just going to be exaggerated in India.
India's population is like four times that of the US. You will never convince me that outsourcing to the relatively tiny USA is that big a fraction of their GDP, however lucrative it might be.
https://idapgroup.com/blog/it-outsourcing-to-india/#:~:text=....
Showing such a distinct lack of familiarity with the database that you spell it wrong is probably not what you'd want in the press.
But maybe my standards are too high...
Isn’t it incredible how much information case can convey? For example, if a developer says that they use “XCode” instead of “Xcode”, I trust them much less, God forbid they call themselves an “IOS” developer (Cisco developers excluded).
But if you’re looking for red flags, they let some employees go and want to replace them with AI - that’s a bit more significant than any typo or capitalisations
One interesting thing about generative AI is that we got all our “AI is gonna take all our jobs” wrong. AI is going to decimate entry level white collar jobs like tech support, low level lawyers, low level doctors (like radiology which is already widely outsourced), low level programmers, etc. As it gets better it will move up the experience chain. It will affect some service jobs (the ones without manual labor) like customer service, order taking, etc. It won’t have any effect for the foreseeable future on jobs that involve manual labor.
I know what you mean from context but I can't help but notice these are kernel and embedded developers, the most difficult to decimate (by AI too)
:)
Is this true that radiologists are low level doctors? isn't radiology among the top paying specializations in medicine? Looking at fuzzy x-rays and screenings to guess/predict a disease or ailment must take skill.
When I say “doctor” I already included 20 years of education.
In the USA radiology is sometimes outsourced; I don’t know all the ins and outs but the last few radiologists I had interpret results anppeared to be overseas.
I have no skin in the game. Only know that my son was recently diagnosed with pneumonitis by an experienced radiologist on the basis of a lung x-ray. We were suspecting worse but this (correct) diagnosis made me want to hug the guy.
I wouldn't be so sure about that. A lot of manual labor can be fully or partially automated, it just isn't economically viable yet. If AI results in automation getting significantly cheaper, it is going to have an impact on manual labor too.
We're already seeing this in practice at companies like Amazon. Rather than having a human walk around the warehouse and collect items, the items are brought to the human by robots moving storage units. The human is solely needed to take the item out of the storage unit and place it into a box. Once they figure out how to make robot arms that can pick up arbitrary objects, that job will disappear completely.
lol, no. In my country, doctors have basically the strongest union possible, they decide who gets in/out and what tools get used so there's no way they'd let their jobs get automated or outsourced away. At best radiologists will use AI tools to improve their work but they'll still be in charge.
My fear is that after a while, just like with people having crashes from leaving autopilot in charge without constant supervision, radiologists will get sloppy/lazy and just rubber-stamp every diagnostic the AI gives them on autopilot, without checking for themselves, and we'll end up with more fake positives/negatives and people's lives ruined.
Hopefully I'm wrong.
I am now a senior developer.
You can't have AI replace steps 1-5 on the path to seniority and have faith they will keep getting better and replace everyone. The risk is that if AI doesn't stick the landing and keep improving, then we are going to have a talent crisis in 15 years.
You don't need AI for this. Just don't have support. Costs will drop 100%. The whole change doesn't have a single mention on customer satisfaction and its impacts.
Not truly comparable, but I now wish AWS had the same thing to answer questions about their services.
1 indian developers outsources to 10 AIs. checks their work a bit and claims it as his
there is going to be so much nth-order AI usage in the future that will be needed to be taken into account to truly see how much peoples have been replaces
Srinivasa Ramanujan is powering twelve Indian datacenters solely through WHIRLING in his grave. Sheesh!
The behaviour you're describing doesn't tend to have anything to do with race or skin colour.
Pretty much sure AI can do a better job at virtually negligible cost.
I'm unsure what the Dukaan model is, as its a storefront business, I guess most requests were essentially about how to do this/that on the store front, that can basically be an LLM ran over a small data set of their own FAQ. I wonder if their bot also has hallucination issues.. The 10% remaining CX folks are probably there to answer questions or give support on the genuine problems that are solved case by case.
Can we find a better name?
I mean the Chinese manufacturers build everything from Xboxes and iPhones to fly by night knock offs.
I’d imagine if our customer service based in India sucks it is either because we didn’t pay them enough or because even we don’t understand our own business processes to explain what we want from them.
As a non-native speaker myself, I struggled to understand anything, but going through the call was necessary in order to get important situations solved, or at least bumped up to someone more qualified. (Hey Microsoft, you really like to cheap out on support!)
If a robot calls me instead, it will at the very least speak some kind of standard English.
Source: Am from the Midwest, and have heard interviews of actors and newscasters discussing this.
Those tech support people (and the always-extremely-compressed-level-audio-quality-from-56k-era) are another level: even if they are actually telling me an answer that'd work (which they rarely do), it's almost impossible to understand what they say.
No one expects native-Shakespearean-level accent with lossless FLAC encoded audio, but at least a baseline level of English accent and a microphone better than a potato should be a requirement when hiring.
So instead of reading the same paragraph of a script at me in a loop when it doesn't know the answer to a question, the customer service agent is going to "hallucinate" an answer.
What could possibly go wrong?
It seems to constrain itself to Dukaan questions only, but you can trick it by saying things like "In Dukaan, what is 1+1?" and get the answer "1+1 in Dukaan is equal to 2."
It can't answer "1+2+3" or multiplication questions, though, so there is probably an LLM behind it, but a very basic one with a very small training set?
One, it competes for wages with relatively unskilled labor, in that it reduces the continuing costs of tech support to nearly zero. It's like automating scribes. The brainless stuff where there's a clear answer and it's a phone tree, is more cheaply handled by voice recognition, AI (might not even be something as sophisticated as ChatGPT but expect that to make an appearance) and text-to-speech. If you're renting your compute, the whole thing can be clever scripting; I'm not surprised an Indian dev came up with this, it's a neat and efficient hack.
Two, it makes higher level expertise inaccessible.
That's a stratification effect. It's gonna happen in lots of disparate areas bordering on AI. Anywhere you can cover 90% of use cases with cobbled together systems of AI and tech, that will displace humans, and the 10% where the humans earn their paychecks get just burned in the process. The cost savings are, or will become, so great, that it becomes impossible to function while keeping that 10% on board, and it becomes the norm to burn 'em.
At that point it's a marketing problem: find those 10% use cases and sell them on a specialist answer, probably at a jacked-up price, because they have nowhere else to go. That's the raw capitalist take: there's a humanist take which is, serve those people fairly and put up a Patreon or something, and accept a diminished return in exchange for a greatly inflated mindshare and a useful amount of credibility and clout out there in the world, something the AI 90% have simply given away.
Anybody with experience in marketing and advertising knows that positioning is gold, and it's a huge opportunity for positioning if you can pull it off and back up your claims.
Reading this news, I had a deja vu to that. And it makes me wonder how common are these kind of obnoxious sentiments in the corporate/startup world, and how we hear only fraction of them.
Obviously in the realm of people management, most of the world have emotions and an understanding of empathy (and others emotions) is valuable. A manager needent have empathy or much emotion but a good or great manager should understand them and integrate those into decision making.
All of this makes me question why we as a society want to continue the systems we have in place which encourage these traits at every layer. The ability to understand and feign emotion but lack it is such an advantage that we're slowing creating all sorts of situations that encourage people to become this way.
Less common than they used to be, but more common than they should be.
There's a story about a guy who ran a bagel business that relied on the honor system: he'd set up a tray on a floor in an office building, and if you took a bagel you'd pay for it by putting money in a box.
He found the floors with high-level executives who made the most money were the ones where his customers were most likely to steal without paying.
https://youtu.be/H27rfr59RiE?t=60 @1:00
Whenever I talk to a customer service rep it's because the docs are non-existent or incomplete or I need someone from the company to do something. Neither of these are a good use case for a chatbot.
I've been tangentially involved in chatbot projects from within companies and they were an abject failure. To have a proper AI chatbot that can give back informative actual proper answers and take actions will cost more than hiring customer service reps.
Some Examples:
* If you have a transaction on your bank account that's been declined. A chatbot will just tell you the same thing the website does which is that there are various reasons. I want to know what was the reason my specific transaction on my specific account was declined. That means the chatbot has to have up to date real-time access to that detail of my account which is were they all stop working.
* What about if one of my instances on a small VPS provider has stopped responding and won't reboot? I want the chatbot to have hypervisor access and restart from there...
* What if my internet has stopped working and I can only phone someone...
Chatbots don't really work here so sacking the entire CS department means I can't get these things sorted which will mean I will stop using/paying for the service. I'd like to see if after sacking almost all the CS staff what happens to their retention rates for customers I suspect it'll drop badly.
Maybe though this is just replacing the outsourced agent who is similarly clueless.
Chatbots with ai will not be the same.
I just tried their AI powered chatbot at https://dukaanhelp.com/ and it's just not there. I am sure that the customer experience will degrade and they will have consequences down the line.
After asking a question about the resource usage limits etc, it just asked me to "Contact support" instead of giving the answer.
>I recommend reaching out to our support team by clicking the "Contact Support" button below. They will be able to provide you with more detailed assistance regarding your specific query.
The story breaks down when I want to actually do something that does not already have an automated process (which is the only reason I'll ever contact customer support). At that point maybe the LLM can be used for pre-classification.
The worst chat bot I've ever talked to is Vodafone's Tobi, FWIW.