Dissatisfied: Three-fourths of AI customer service rollouts are a letdown

If you're thinking you can replace your human call center staff with a server farm of bots, think again. Nearly three-quarters of enterprises that deploy AI customer communications agents later roll them back or shut them down, according to new research suggesting the systems are far harder to manage reliably in production than …

[brep51r]

hype killed the AI star

i hate the hyperscalers as much as the next non sociopath but let’s be honest – there was potential for these bots to make customer support better for everyone, workers and customers. i can imagine them being used as efficiency multipliers for customer support agents, like canned responses on steroids or slightly more interactive forms. and with more capable models more features could be tested and slowly added. it was the over promising and under delivering of the silicon valley biggies that led to this mess.

[GuldenNL]

Re: hype killed the AI star

Bingo. A massive amount of organizations still focus on deflecting inbound calls. DUMB DUMB DUMB!

I always focus on agent assist. Self service would be accomplished if customers could and AI barely improves that.

Focus on making agents' work easier with timely relevant information, which helps customers and improves agents' experiences. Win/win/win. And then turn the AI into root cause radars so the problems that caused the interaction can be addressed.

[Anonymous Coward]

Re: hype killed the AI star

The problem is that a text generator is great at generating text, but there is no way to make that text adhere to any given set of facts about the world or your systems - simply having enough training data to generate comprehensible text means that it will contain enough data to add unwanted extras and it cannot be forced to stick to any given set of options.

The only way I can see it working is having the language model as a kind of interface layer turning user requests into queries against a much more limited decision tree, but even then the chances of inaccuracies in transposition are high enough that you're going to leave a fair proportion of users unhappy.

[brep51r]

Re: hype killed the AI star

it’s used that way around me quite a bit, chat bots redirecting me to relevant FAQ pages or collecting my complaint and case details to smooth over transition to a human. i didn’t mind it. it worked because the translation to the decision tree was transparent and i had the ability to interrupt and choose differently or ask for a human.

[Nate Amsden]

terrible

A recent support experience with Citrix(Netscaler customer for 15 years) was quite terrible with their stupid "AI" bot. Apparently can't call them on the phone anymore(unsure when that changed, I don't need to open cases often), but their bot said I can wait on the line for a live agent so I waited that day for 90 minutes. Nobody live came on, the bot ignored all queries once I was in the queue, never gave another update what position in line I was(initially I was #22).

I tried again a week later and waited THREE HOURS in chat, THAT time the bot didn't even tell me what position I was in and again never gave an update when I may get a live agent.

Citrix was apparently being hammered by 2 things at the same time, a security update issue as well as pending licensing changes. So they were slammed. I can understand being slammed.

I can't understand a simple AI bot not being able to tell me something as simple as what position in line I am in at regularly scheduled intervals (phone systems have done that for decades). And even worse, ignore ALL queries from me while I was waiting. I should be able to ask it at any time what position I am in, how long can I expect to wait etc. It should know all that stuff(average wait times today etc). But it knew nothing. I can only imagine perhaps there was more than 1000 people in line on those days, maybe that overflowed a counter or something in the bot. But if a bot tells me there are 1000 people in line and you may have to wait here for 3 days to get support, then I will know to give up. Many phone support systems in the past offer an option to call the customer back automatically when it's their turn, but with this AI bot at least, no option for that, have to look at the screen and wait to see if anyone types something.

I eventually solved the issues myself(two of them being real bugs, but I found workarounds), though still need to open another case for my original issue now that telemetry is flowing for the first time(originally was rejected to be allowed to open ANY case because I didn't have telemetry setup, something I've never had to do until that point).

[Guido Esperanto]

Re: terrible

Had a similar experience sorting out a payment issue for my Sons Uni.

The payment handler is clearly a US company, the ai bot on the line was manipulated after enough attempts to put me through to somebody.

Now normally you get "if you'd rather us call you back press 1 or stay on the line and we'll get to you"....then perhaps a 2 min time gap then the message again.

Not this service.

No sooner did it end the message..it started the message again. Cycling perpetually

After 10mins of that balls I pressed 1...and was called back almost instantly.

[Pascal Monett]

Re: terrible

Thank you for posting that. I will refer to your post in any future conversations I may have with starry-eyed AI-lusting managers.

Not that it will do much good, those stars are the bonuses they see coming.

[Fuzzy Fitzpatrick]

Three-fourths?

Surely you mean three quarters…

[Jason Bloomberg]

Re: Three-fourths?

Or three-in-four, or 75%, or anything other than that jarring Americanism.

I'd say more but it's a fourth-past-six and I'm already running late.

[coredump]

surprise

> Nearly three-quarters of enterprises that deploy AI customer communications agents later roll them back or shut them down,

I'm less surprised by the AI "agents" being crap, than I am about the "enterprises" throwing in the towel on it rather than doubling down.

"No, you see, it's not working out because we didn't layoff *enough* humans; if we'd just got rid of even more payroll, er, I mean 'individual contributors', then everything would be great!"

[Pascal Monett]

Re: surprise

Good point.

[Filippo]

Re: surprise

That's what the remaining one quarter are doing.

[a_yank_lurker]

Fake It Until You Lose

The greatest salvation to the human race is the AI hype is doing a good job of destroying AI. Unfortunately it will do a lot damage before it's over.

[Anonymous Coward]

AI decision makers

The people who believed AI would save them from the tedious business of understanding the problem or testing the proposed solution?

[Taliesinawen]

How the world was convinced a spicy autocomplete engine was the second coming

ClippyAI: The silicon-scented bubble of AI hype hasn't just popped; it’s currently undergoing a messy, high-velocity unscheduled disassembly that's leaving a trail of scorched balance sheets and bewildered humans in its wake. From the 74 percent of firms desperately yanking their hallucinating chatbots back into the hangar to Apple's $250 million "AI-washing" penance, 2026 has become the year the marketing brochures finally met the cold, hard floor of reality.

[druck]

Re: How the world was convinced a spicy autocomplete engine was the second coming

AI;dr

[Taliesinawen]

Re: How the world was convinced a spicy autocomplete engine was the second coming

SpecisT 15/05/2026: 11:18 BST

[TheServitor]

I wonder about the guardrails argument part though. I'm pretty skeptical about safety and compliance theater at big corporations and law firms, where bureaucracy sometimes causes more trouble than it saves. Which is to say, they're the ones seeing it first, but also the ones with maximum risk aversion and the ones with the most frustrating implementations, so naturally they have the highest pullback rate.

[Anonymous Coward]

AliBaba's…

…AI agent is pretty good, and does a good job of handing off to a human¹ once he's decided he's no longer useful.

¹ Who replies within minutes (and sometimes seconds). It's also been my experience from our Chinese suppliers that they all have amazing customer service (especially nowadays that they speak good English).

[T. F. M. Reader]

Re: AliBaba's…

Not only Chinese: Amazon's customer support bot is pretty good (= pretty quick) to handle off to a human. The human - often, though not always, Asian - takes over really quickly and is usually very helpful. I have an impression that the bot is there mostly to help customers who could not or were too lazy to peruse package tracking page themselves, or something similar. This is a guess, because I never bothered AMZN Support with that, but if it is correct it's a perfectly valid use case for a bot.

A tip for all those big "AI-centric" organizations from someone who has worked closely with Customer Support at multiple smallish companies with really big and important customers: after having a good product the most important thing in achieving customer satisfaction (and renewals, and upsales, etc.) is to always answer the phone when it rings. It matters much less if the person who answers the phone can immediately resolve the problem, as long as he or she knows whom to ask (often me - that's how I know) how to route the ticket for a later follow-up. But do answer the phone (ack the ticket, whatever) when it rings!