Insurance giant finds claims rep that gives a damn (it's AI)
- Reference: 1739439016
- News link: https://www.theregister.co.uk/2025/02/13/allstate_insurance_ai_rep/
- Source link:
Even if it's fundamentally strings of ones and zeroes, US insurance biz Allstate is finding that generative AI crafts more empathetic and customer-friendly emails than its human claims reps, though anyone who has had dealings with insurance providers might not be surprised.
Allstate has some 23,000 insurance reps sending out around 50,000 communications to claimants every day, gathering information for negotiating settlements and the like.
[1]
"When these emails used to go out, even though we had standards and so on, they would include a lot of insurance jargon. They weren't very empathetic ... Claims agents would get frustrated, and so it wasn't necessarily great communication," Allstate CIO Zulfi Jeevanjee told [2]The Wall Street Journal .
[3]
[4]
Now nearly every one of those emails is written by AI, specifically OpenAI's GPT models, grounded in Allstate's company-specific lingo. For the time being, the AI works alongside its army of reps. "The claims agent still looks at them just to make sure they're accurate, but they're not writing them anymore," Jeevanjee added.
The result has been more polite and less jargon-filled communications – something Allstate's reps likely appreciate, given the repetitive back-and-forth of insurance claims.
[5]
Allstate said meatbag reps tended to use phrases like "first notice of loss" or refer to "CCC Intelligent Solutions," a vendor the insurance giant employs to estimate repair costs, without explanation.
It pointed to an email where a rep doled out the acronym "UPP inventory list," but its AI counterpart helpfully spelled it out: "Unscheduled Personal Property inventory list." Jeevanjee also said that where human reps might question the validity of a claim, Allstate's AI would give customers "the benefit of the doubt."
Indeed, the chatbots appear to have an impeccable bedside manner, making empathetic noises like "Thank you for your message. I understand your concern," and "If you have any other questions, feel free to ask" – phrases that may not cross the mind of a tired, grouchy human who is just trying to make it through the day.
[6]AI summaries turn real news into nonsense, BBC finds
[7]AI agents? Yes, let's automate all sorts of things that don't actually need it
[8]DPD chatbot blasts courier company, swears, and dabbles in awful poetry
[9]UnitedHealthcare's broken AI denied seniors' medical claims, lawsuit alleges
While human reps at Allstate still have jobs for now, it's early days for AI-driven customer service in the insurance industry. The WSJ cites a survey from the International Customer Management Institute, where 35 percent of respondents said they think AI will reduce hiring requirements, while 38 percent didn't see it as a factor.
And insurance's interest in AI doesn't stop at customer service. The sector is looking to employ the tech for risk assessment, policy pricing, and claims review – areas where Allstate is experimenting with AI to analyze photos of vehicle damage.
[10]
Of course, AI deployment includes its own risks for vendors who use it. For example, litigation, successful or not, is expensive. UnitedHealthcare was [11]sued in 2023 for allegedly using AI to automate healthcare claim denials, raising concerns about the growing role of automation in insurance decisions. The case [12]is still ongoing .
Jeevanjee said the goal of the AI implementation is to make Allstate more customer focused. "If I think about the insurance industry in general, we haven't really done a great job of being customer obsessed," he said. "That's really what I'm trying to drive."
Allstate's AI rollout appears to be going smoother than that of courier DPD. A year ago, a customer exploited its chatbot's weak guardrails and forced it into spouting [13]self-flagellating Vogon poetry . ®
Get our [14]Tech Resources
[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Z63Q0ArroCZoV3csRxdxyAAAAJc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[2] https://www.wsj.com/articles/turns-out-ai-is-more-empathetic-than-allstates-insurance-reps-cf5f7c98
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z63Q0ArroCZoV3csRxdxyAAAAJc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z63Q0ArroCZoV3csRxdxyAAAAJc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Z63Q0ArroCZoV3csRxdxyAAAAJc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2025/02/12/bbc_ai_news_accuracy/
[7] https://www.theregister.com/2025/01/27/ai_agents_automate_argument/
[8] https://www.theregister.com/2024/01/23/dpd_chatbot_goes_rogue/
[9] https://www.theregister.com/2023/11/15/unitedhealthcare_ai_medicine/
[10] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offbeat/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Z63Q0ArroCZoV3csRxdxyAAAAJc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[11] https://www.theregister.com/2023/11/15/unitedhealthcare_ai_medicine/
[12] https://www.courtlistener.com/docket/68006832/estate-of-gene-b-lokken-the-v-unitedhealth-group-inc/?filed_after=&filed_before=&entry_gte=&entry_lte=&order_by=desc
[13] https://www.theregister.com/2024/01/23/dpd_chatbot_goes_rogue/
[14] https://whitepapers.theregister.com/
Re: "Thank you for your message. I understand your concern,"
I recently had some dealings with OVO and in the process I suspect their entire complaints response emails are AI-generated or copy-paste policy led:
I hope this email finds you well. My name is
First and foremost, I would like to sincerely apologise for any inconvenience or frustration you may have experienced during this process. Your concerns are incredibly important to us, and I fully understand how unsettling it can be to navigate issues related to your energy services. Please know that I am here to listen and assist you every step of the way.
I want to reassure you that my priority is to resolve your complaint as efficiently and effectively as possible. To ensure we address your concerns thoroughly, I would like to discuss the matter with you directly. If you prefer a callback, please do let me know your preferred date and time, and I will do my utmost to accommodate your schedule.
Your feedback is invaluable to us, and I sincerely appreciate your patience and understanding as we work to resolve your issue. Together, we can find a solution that meets your needs.
Thank you for allowing us the opportunity to address your concerns. I look forward to speaking with you soon.
Warm regards,
---
There then follows a series of emails where various OTHER people at OVO then largely regurgitate my complaint and provide absolutely no useful information or updates whatsoever. One of them "confirmed" I had a smart meter. I know. You fitted it. I told you it's working fine. The complaint has nothing to do with the meter. Each one tries to talk to me like they're a con artist trying to win over some old lady or something, and none of them provide any useful context or updates. It's not even just boilerplate, by the look of it, but they can respond with a bunch of text that takes you a minute or so to read but actually says that they've done nothing so far. I wouldn't be surprised if their agents are using AI to just bat the ticket response further down the queue, in fact.
When even the COMPLAINTS are handled like this, it's no wonder they're regarded as one of the worst companies. It's the one place I expect to speak to an actual human, because to get that far means that the computers aren't doing what they should (what they were told to do, no doubt, but not what they should) and a human needs to actually look at the edge cases and do something about them that's not in a rulebook somewhere.
I hate it and I moved away from them largely because of it.
The irony was that they replaced my old teleswitch meter (which I'd been asking for for TWO YEARS because it's so obsolete, and I couldn't change provider with it in place) with a flash new smart meter. Which worked absolutely flawlessly and fed the correct data instantly back to them and continued to do so. But what happened is that they decided to utterly ignore those numbers for 3 billing cycles. Literally "estimated" bills only when they have the data of every 30 minutes of every day of the last 3 months. So I gave them an ultimatum - the next bill has to be right (not inflated 3 x by imaginary numbers) or I'm leaving. They couldn't get it right, so I left. I'd pretty much had no problem with them until I was actually ABLE to finally leave them, and from that second, they couldn't be bothered to retain me as a customer at all.
It's still ongoing now, 2 months after I transferred to someone else, because despite having refunded me for all that overpayment, they then took ANOTHER payment (even though I'm not supplying them... but apparently I was required to keep the Direct Debit open to get the refund). Which now I've had to file another complaint about to get a refund for. So I've had two emails saying "Your final bill" because of that.
But throughout that process, all I have are what appear to be AI-generated or at least AI-assisted emails that largely waffle about how sorry and warm they are (ironic, for an energy company), but don't actually answer anything. I fear this is the future of complaints handling, even written, and I think we're going to have real problems where these things aren't actually progressing complaints at all but the statistics show that they appear to be.
Like I always say about train delay announcements. You're NOT sorry the train is late. If you were sorry, you'd do something about it and it wouldn't happen again. You can't be sorry 50+ times a day repeatedly on even just a single station, let alone what you're announcing over the whole network. It's faux sentiment while on the backend it's just another date/time statistic that nobody pays any attention to at all.
Just remember
If it can't lie, it ain't "AI"
The inference being you need to lie to be a claims agent that another human recognises
"Of course we've processed your claim"
I would be intrigued to see what faith the companies put in all this frippery. Are "AI" agents allowed to settle claims ? I suspect not. And I suspect when one does, the company will reverse ferret on it's legitimacy.
A fun future awaits.
"The claims agent still looks at them just to make sure they're accurate"
Yeah, sure.
They put them on screen and wait a second before validating, but I'll be damned if they really take any time to check anything.
After all, it's written by AI, right ? If they change anything, then it's their fault.
Does it matter how polite they are ?
If they still deny your claim anyway ?
A new technology. Empathetic weasels.
"Thank you for your message. I understand your concern,"
But... that's the very sort of empathy that tells me 'this empathy is written on a policy document somewhere".
Though to be fair, it's probably better than what might come out a person who's been doing the job for the last six months...