I've used some great AI support agents/bots, whatever they're called. An example of a great one is Lemonade (the insurance company).
Most of the others are terrible. I'll pick on Dell for this next one. Their website in general, is terrible. Why would their support agent be any better?
We purchased a switch from Dell, that cost more than my SUV. Having a hardware issue with it, can't remember if it was fans or a power supply or what. It needs someone to come out and repair it, and we have the big fancy service plan.
There's no way to find the switch on the Dell support website. It's not listed anywhere.
However, this product doesn't have a service tag, or express service code, so you can't do that.
So you try to use the 'Virtual Assistant', who has no idea what you're trying to tell it, no matter which way you try. They link to a contact page but you just end up getting looped back around. There is no useful information to be extracted from this dumb thing.
I finally found a phone number on the website, and was connected to a woman who sounded like she was smoking cigarettes in a basement somewhere in Alabama. She looked through her notes on the computer, and found out for this particular switch I need to go through the OEM manufacturer. I was relieved to have gotten useful information.
I think there is some confusion in the industry about what AI support bots are for.
They are primarily for making customers go away. The terrible support experience was already standard - they are just automating the "go away" experience.
If it were up to me, I'd drop Dell just because of the hoops they make you go through to get service on a server.
They want you to diagnose the issue, and to be the remote hands. Shuffling around DIMMs and drives and re-seating cables and all kinds of crap. Just send someone to out fix the damn thing like HPE does.
There was a thread on HN a while back and it was mentioned they do this to save on costs by making the customer jump through hoops and hoping they will go away. Also, the field techs are made to jump through hoops as well.
The ironic part is that Dell was once known to provide excellent onsite repairs without too much hassle. Now they are known for the exact opposite. So, someone at Dell is running their customer service into the ground.
AI “customer service” is so incredibly bad. I’ve not had a single interaction that an AI agent could solve. It’s just a huge waste of my time and an increase in aggravation.
It's a really good way to offload costs to the customer. You make the process so full of friction that the customer feels like it's not worth the process to make the return. Works the best with big monopolies
I'm 100% convinced that this is the reason so many customer service jobs are being shelved.
No one who actually wants to provide a good experience for their customers would offshore, nor would they offload this to AI agents. They know that customers have very limited opportunities nowadays, so the company responses seem to be along the lines of, "fine, go somewhere else! Oh, there's nowhere else to go? I guess you'll be staying with us afterall muahaha!"
You think the average consumer is willing to fight firms hard to get stuff fixed? Absolutely not. Things like switching costs is what firms take advantage of -> they know you most likely arent going anywhere so shut up and accept the poor service.
Would people here generally fight and switch? Sure. But you lot are what % of the population..?
A lot of the HN crowd is a lot more technically capable than the average consumer. The more competent customers never see the 80-90% of inbound ticket volume that are items answered in FAQs that can be solved with an AI agent. They only know they had an edge case problem they couldn't resolve after quite a bit of effort. And the AI agent and canned responses were totally useless and frustrating barriers to the real help you needed (a person).
After putting in AI agents this year, our customer service scores went up considerably. And the reason was basically that the average customer was getting way faster answers to queries.
How does the average customer even get a hold of a human in the first place? The bots are so aggressively unwilling to connect you that I sometimes fail.
We only have chat/email based support, no experience in phone trees. If the agent determines the question is outside of defined scope, it puts it into the human queue. We are seeing very low rates of improper filtering (i.e., people never making it to the human queue).
This is not a plug and play customer bot product, we actually built a custom solution on top of Bedrock.
If it is an LLM, it has been prompted to say almost nothing except prewritten responses to avoid legal issues or to avoid free LLM usage. I think it was Chipotle support that would write code if you asked it.
Some of the best uses of AI in call centres that I’ve seen focus on improving human interaction instead of getting rid of it.
For example:
- Finding information in the background / performing an account health check instead of putting the caller on hold for 20 minutes.
- Helping newer consultants find the same offers experienced staff know about amongst 200 pages of active/targeted offers.
- Automatically documenting calls so when someone calls back, they don’t get a “sorry, I can’t find any notes about that on your account”.
- Giving the offshore call center consultant a customized script based on the active discussion instead of a generic one
(Yes, reading from a script is still bad, but at least this makes it somewhat more natural rather than it feeling like you’re talking to a lobotomised employee.)
I don't know how much more cost effective AI bots are for customer service over offshore call centers currently, but I would guess that the industry sees the AI tooling getting much better over the next X years, while call center agents are going to be relatively level in terms of skill, so companies are investing early on in the technology. To be honest, my favorite customer service experience are ones where I tell a bot the issue, and they just refund me my cost, widely seen on delivery apps and Amazon. This is ripe for moral hazard but I'm sure there is monitoring happening for that as well.
"so companies are investing early on in the technology."
But this doesn't make sense. The AI bots are worse at the job, across all metrics, other than being cheaper. They're not "investing in the technology" in the sense of actual investment. They are just trying to save money. That's it.
If they cared about the experience of their customers, they wouldn't touch these things with a barge pole.
> I don't know how much more cost effective AI bots are for customer service over offshore call centers currently
They are already fairly cost effective. A the Corporate VC arms of the major call center and BPO players like Infosys, TCS, and others have already funded a large portion of TTS startups, and in a lot of cases are training on customer calls.
There's a reason ElevenLabs, Deepgram, and others are the giants they are now.
This is published by Bold Agents, a company that appears to provide a platform for AI customer service agents, and is based on their own data. Doesn't this really just mean that they're saying that their own product kind of sucks?
I believe Sierra is the market leader in this space with a few other companies like Salesforce and Decagon also competing. It would be more interesting to see an evaluation of performance by these parties before making sweeping claims about the performance of AI agents as a whole.
1-800 Contacts has my undying love because their customer support is excellent and seamless. I wouldn't say I look forward to calling them, but it's just so simple and effortless.
Some, maybe most, humans outperform AI. I've had some calls where I would have preferred an AI that could not understand me to an agent whom I could not understand AND who could barely understand me.
Had a recall reimbursement through Stellantis and couldn’t figure out if it was the best AI I’ve ever spoken to or the most professional human I’ve ever spoken to.
HN readers - and even more so HN commentators - are not typical though. Literally 99% of people interacting really are asking a question answered by the FAQ, just like 99% of people calling the customer service number for their credit card want to pay their bill or find out their balance.
Doesn’t work anymore. Our apartment building has a few providers and people were trying to report an outage with ASNs specified and suggesting they check BGP advertisements and so on. It turns out our Google Fiber microwave dishes were misaligned.
don't worry, they will make Ai listen to the calls with human agents and in about a year, it will out perform them. Have you heard that "this call will be recorded" blah blah.. Now they will use those recordings... Time is on their side.
The AI era isn't about what performs best. It's about what can be delivered most cheaply and isn't a liability because your competition will be doing the same thing.
Configure agents for consumers at scale to report violations to regulators. Wait time not accurate or communicated? Complaint filed. Human agent not available within time limit? Complaint filed.
Think about all the time and effort for all of the consumers in the aggregate that would be needed to hold corporate counter parties accountable, with the cost being code and some compute time on device. This is where AI agents shine. They can do the mundane work of capturing evidence of compliance violations, time stamping, packaging the violation event, and shipping to an endpoint for reporting them.
Every consumer’s phone becomes a regulator watchdog node. We can even use LLMs to rapidly ship the app(s) needed to do this.
The provider for Chase credit card claims (insurance, price match, etc.) went through a company, EClaimsLine, that had nearly that exact phrase bold and red on their website for many, many years. I'm honestly shocked that their new website after the rebrand to Chase Card Services doesn't say it too.
I wonder how many companies are going to trivially get around this by connecting you to an operator that makes you wait to be connected to the other department because they can't handle anything and only route calls.
Support has costs. Fine. I consider purchases under $500 to not include support, and treat them all as "sold as is"
However, if I need support and cannot reach a real living breathing human in under 30 min regarding any purchase over $500, I hang up, call AmEx and file a chargeback, citing inability to gain support. So far they have backed me every time. If enough people do it, chargeback costs send the message.
The problem is it doesn’t matter if they perform better. If AI is just above “too shitty” and costs less, that’s what businesses will generally choose. And that’s the entire problem. “Good enough” reigns supreme for most businesses. In some cases that’s fine, in a lot of others it just tests our patience/tolerance and makes things painful.
It can be summed up thus : AI agents answer the phone slightly more often, but they get hung up on 7 times more often than humans. Overall they make customer service and your business work much worse.
It is also a bullshit advertorial for a product; In this case, an AI sales agent. There is (almost) no worse idea in human history than an AI agent trying to sell things to people. Their entire website is AI generated, and is so stuffed full of slop that it's impossible to read.
I've used some great AI support agents/bots, whatever they're called. An example of a great one is Lemonade (the insurance company).
Most of the others are terrible. I'll pick on Dell for this next one. Their website in general, is terrible. Why would their support agent be any better?
We purchased a switch from Dell, that cost more than my SUV. Having a hardware issue with it, can't remember if it was fans or a power supply or what. It needs someone to come out and repair it, and we have the big fancy service plan.
There's no way to find the switch on the Dell support website. It's not listed anywhere.
https://www.dell.com/support/home/en-us/products/networking_...
You can't actually contact support, you have to identify your device, and then you get options from there.
https://www.dell.com/support/incidents-online/en-us/contactu...
However, this product doesn't have a service tag, or express service code, so you can't do that.
So you try to use the 'Virtual Assistant', who has no idea what you're trying to tell it, no matter which way you try. They link to a contact page but you just end up getting looped back around. There is no useful information to be extracted from this dumb thing.
I finally found a phone number on the website, and was connected to a woman who sounded like she was smoking cigarettes in a basement somewhere in Alabama. She looked through her notes on the computer, and found out for this particular switch I need to go through the OEM manufacturer. I was relieved to have gotten useful information.
I think there is some confusion in the industry about what AI support bots are for.
They are primarily for making customers go away. The terrible support experience was already standard - they are just automating the "go away" experience.
Wow! That is crazy expensive switch.
Also, a relief we dropped Dell for these exact reasons a few years ago. Seems nothing has changed and it’s good we got away from them.
If it were up to me, I'd drop Dell just because of the hoops they make you go through to get service on a server.
They want you to diagnose the issue, and to be the remote hands. Shuffling around DIMMs and drives and re-seating cables and all kinds of crap. Just send someone to out fix the damn thing like HPE does.
There was a thread on HN a while back and it was mentioned they do this to save on costs by making the customer jump through hoops and hoping they will go away. Also, the field techs are made to jump through hoops as well.
The ironic part is that Dell was once known to provide excellent onsite repairs without too much hassle. Now they are known for the exact opposite. So, someone at Dell is running their customer service into the ground.
AI “customer service” is so incredibly bad. I’ve not had a single interaction that an AI agent could solve. It’s just a huge waste of my time and an increase in aggravation.
It's a really good way to offload costs to the customer. You make the process so full of friction that the customer feels like it's not worth the process to make the return. Works the best with big monopolies
Some customers are smart too, they just use their own agents to flood support agents until the problem is resolved.
I'm 100% convinced that this is the reason so many customer service jobs are being shelved.
No one who actually wants to provide a good experience for their customers would offshore, nor would they offload this to AI agents. They know that customers have very limited opportunities nowadays, so the company responses seem to be along the lines of, "fine, go somewhere else! Oh, there's nowhere else to go? I guess you'll be staying with us afterall muahaha!"
Correct.
People here lack a lot of nuance.
You think the average consumer is willing to fight firms hard to get stuff fixed? Absolutely not. Things like switching costs is what firms take advantage of -> they know you most likely arent going anywhere so shut up and accept the poor service.
Would people here generally fight and switch? Sure. But you lot are what % of the population..?
A lot of the HN crowd is a lot more technically capable than the average consumer. The more competent customers never see the 80-90% of inbound ticket volume that are items answered in FAQs that can be solved with an AI agent. They only know they had an edge case problem they couldn't resolve after quite a bit of effort. And the AI agent and canned responses were totally useless and frustrating barriers to the real help you needed (a person).
After putting in AI agents this year, our customer service scores went up considerably. And the reason was basically that the average customer was getting way faster answers to queries.
How does the average customer even get a hold of a human in the first place? The bots are so aggressively unwilling to connect you that I sometimes fail.
We only have chat/email based support, no experience in phone trees. If the agent determines the question is outside of defined scope, it puts it into the human queue. We are seeing very low rates of improper filtering (i.e., people never making it to the human queue).
This is not a plug and play customer bot product, we actually built a custom solution on top of Bedrock.
Average scores going up, but what about tail? 90th / 95th / 99th percentile
For us, they are still at the same scores as before AI agents.
AT&T & Comcast figured out how to have incredibly bad automated service quite a while before LLMs.
If it is an LLM, it has been prompted to say almost nothing except prewritten responses to avoid legal issues or to avoid free LLM usage. I think it was Chipotle support that would write code if you asked it.
My assumption is that it is there to make the process less convenient so people only use it if they really need to.
I've gotten automatic refunds from Amazon bots. (I also got automatic credit from Lyft, but I consider that a failure because I wanted my money back.)
Some of the best uses of AI in call centres that I’ve seen focus on improving human interaction instead of getting rid of it. For example:
- Finding information in the background / performing an account health check instead of putting the caller on hold for 20 minutes.
- Helping newer consultants find the same offers experienced staff know about amongst 200 pages of active/targeted offers.
- Automatically documenting calls so when someone calls back, they don’t get a “sorry, I can’t find any notes about that on your account”.
- Giving the offshore call center consultant a customized script based on the active discussion instead of a generic one
(Yes, reading from a script is still bad, but at least this makes it somewhat more natural rather than it feeling like you’re talking to a lobotomised employee.)
I don't know how much more cost effective AI bots are for customer service over offshore call centers currently, but I would guess that the industry sees the AI tooling getting much better over the next X years, while call center agents are going to be relatively level in terms of skill, so companies are investing early on in the technology. To be honest, my favorite customer service experience are ones where I tell a bot the issue, and they just refund me my cost, widely seen on delivery apps and Amazon. This is ripe for moral hazard but I'm sure there is monitoring happening for that as well.
"so companies are investing early on in the technology."
But this doesn't make sense. The AI bots are worse at the job, across all metrics, other than being cheaper. They're not "investing in the technology" in the sense of actual investment. They are just trying to save money. That's it.
If they cared about the experience of their customers, they wouldn't touch these things with a barge pole.
> I don't know how much more cost effective AI bots are for customer service over offshore call centers currently
They are already fairly cost effective. A the Corporate VC arms of the major call center and BPO players like Infosys, TCS, and others have already funded a large portion of TTS startups, and in a lot of cases are training on customer calls.
There's a reason ElevenLabs, Deepgram, and others are the giants they are now.
This is published by Bold Agents, a company that appears to provide a platform for AI customer service agents, and is based on their own data. Doesn't this really just mean that they're saying that their own product kind of sucks?
I believe Sierra is the market leader in this space with a few other companies like Salesforce and Decagon also competing. It would be more interesting to see an evaluation of performance by these parties before making sweeping claims about the performance of AI agents as a whole.
1-800 Contacts has my undying love because their customer support is excellent and seamless. I wouldn't say I look forward to calling them, but it's just so simple and effortless.
Some, maybe most, humans outperform AI. I've had some calls where I would have preferred an AI that could not understand me to an agent whom I could not understand AND who could barely understand me.
Had a recall reimbursement through Stellantis and couldn’t figure out if it was the best AI I’ve ever spoken to or the most professional human I’ve ever spoken to.
"AI"
Usually when I chat with a support bot it doesn't feel like an LLM but something that performs keyword matching and a prewritten response pre-LLM era.
I hate it either way, because my typical issue is something bespoke that needs a human to resolve. Or else I wouldn't be reaching out.
HN readers - and even more so HN commentators - are not typical though. Literally 99% of people interacting really are asking a question answered by the FAQ, just like 99% of people calling the customer service number for their credit card want to pay their bill or find out their balance.
So then offer the shibboleet option! https://xkcd.com/806/
Doesn’t work anymore. Our apartment building has a few providers and people were trying to report an outage with ASNs specified and suggesting they check BGP advertisements and so on. It turns out our Google Fiber microwave dishes were misaligned.
But all shibboleths are obsolete now.
don't worry, they will make Ai listen to the calls with human agents and in about a year, it will out perform them. Have you heard that "this call will be recorded" blah blah.. Now they will use those recordings... Time is on their side.
Weird, it’s almost like people hate talking to robots
The AI era isn't about what performs best. It's about what can be delivered most cheaply and isn't a liability because your competition will be doing the same thing.
https://legiscan.com/CA/text/AB1609/id/3460800
- California’s AB 1609 requires large businesses to connect customers with human agents within 15 minutes.
- Businesses exceeding $500 million in revenue must display wait times and disclose chatbot interactions.
- Violations carry civil penalties starting at $5,000, making human access a legal obligation.
> must display wait times
"We're sorry, we're receiving more calls than usual, stand by..."
24/7 for over 30 years!
Configure agents for consumers at scale to report violations to regulators. Wait time not accurate or communicated? Complaint filed. Human agent not available within time limit? Complaint filed.
Think about all the time and effort for all of the consumers in the aggregate that would be needed to hold corporate counter parties accountable, with the cost being code and some compute time on device. This is where AI agents shine. They can do the mundane work of capturing evidence of compliance violations, time stamping, packaging the violation event, and shipping to an endpoint for reporting them.
Every consumer’s phone becomes a regulator watchdog node. We can even use LLMs to rapidly ship the app(s) needed to do this.
The provider for Chase credit card claims (insurance, price match, etc.) went through a company, EClaimsLine, that had nearly that exact phrase bold and red on their website for many, many years. I'm honestly shocked that their new website after the rebrand to Chase Card Services doesn't say it too.
Having to waste 15 minutes with an AI is still crazy though.
Nah, if it becomes law, we’ll get apps which just call a number for you and run out the clock.
I believe Google and Apple phones support this today. “Hold assist” and “Hold for Me.”
https://support.apple.com/en-us/127521
https://support.google.com/pixelphone/answer/10071878?hl=en
Hold Assist doesn’t work if there’s any speech (like an AI agent) on the line, it specifically only works when on hold.
Easily updated existing feature, code path, and user experience is the point. Just has to be good enough to iterate on.
I wonder how many companies are going to trivially get around this by connecting you to an operator that makes you wait to be connected to the other department because they can't handle anything and only route calls.
Keep tightening the law and grinding them with monetary penalties until compliance with the spirit of the law is achieved.
Well you can't legally penalize someone complying with the letter of the law, so the law just needs to be tight.
> AB1609
Status: (Engrossed - Dead) 2016-11-30
> Status: (Passed) 2026-09-28 - Chaptered by Secretary of State - Chapter 733, Statutes of 2026. [AB1609 Detail]
Link fixed! Thank you! Additional citations:
https://leginfo.legislature.ca.gov/faces/billHistoryClient.x...
https://www.gov.ca.gov/2026/09/28/governor-newsom-signs-comm...
Support has costs. Fine. I consider purchases under $500 to not include support, and treat them all as "sold as is"
However, if I need support and cannot reach a real living breathing human in under 30 min regarding any purchase over $500, I hang up, call AmEx and file a chargeback, citing inability to gain support. So far they have backed me every time. If enough people do it, chargeback costs send the message.
Man that annoys me calling some place and it's that openai voice answering
The problem is it doesn’t matter if they perform better. If AI is just above “too shitty” and costs less, that’s what businesses will generally choose. And that’s the entire problem. “Good enough” reigns supreme for most businesses. In some cases that’s fine, in a lot of others it just tests our patience/tolerance and makes things painful.
AI slop article, of course.
It can be summed up thus : AI agents answer the phone slightly more often, but they get hung up on 7 times more often than humans. Overall they make customer service and your business work much worse.
It is also a bullshit advertorial for a product; In this case, an AI sales agent. There is (almost) no worse idea in human history than an AI agent trying to sell things to people. Their entire website is AI generated, and is so stuffed full of slop that it's impossible to read.