An engineer argued support loops are intentional cost-reduction tactics.
so... has anyone seen an ai customer support bot resolve an issue firsthand, without escalating to a human or forcing you to give up after a hall-of-mirrors-infinite-loop?
if so, what task (& for which co.) did it complete the above successfully?
Some users share positive examples of AI bots at Klarna, Verizon, and Cursor resolving refunds and issues quickly where humans failed, whereas others claim they have never witnessed bots taking real actions and prefer human interactions.
No Digg Deeper questions have been answered for this story yet.
I did AI for customer support at stripe for a bit before I left, and I learned two things: 1) humans are incredibly, incredibly bad at support 2) support exists to reduce the burden on the core engineering team by redirecting and filtering, and the best way to support is by FIXING THE ACTUAL PROBLEMS IN THE PRODUCT and allowing self serve
so... has anyone seen an ai customer support bot resolve an issue firsthand, without escalating to a human or forcing you to give up after a hall-of-mirrors-infinite-loop?
if so, what task (& for which co.) did it complete the above successfully?
@suchenzang the hall of mirrors is actually engineered, designed to reduce costs. the way you help with "AI" is actually just.. datapipelines and attribution to teams, to create incentives (negative or positive) for engineering to actually fix the core issues and see a line go down
I did AI for customer support at stripe for a bit before I left, and I learned two things: 1) humans are incredibly, incredibly bad at support 2) support exists to reduce the burden on the core engineering team by redirecting and filtering, and the best way to support is by FIXING THE ACTUAL PROBLEMS IN THE PRODUCT and allowing self serve
@suchenzang it's the **same** exact problem as resource usage. if you measure it, and give teams and engineers a hill to climb, they'll climb it
make the problem verifiable in some manner.. testable
@suchenzang the hall of mirrors is actually engineered, designed to reduce costs. the way you help with "AI" is actually just.. datapipelines and attribution to teams, to create incentives (negative or positive) for engineering to actually fix the core issues and see a line go down
(preferably examples that don't involve refunds)
Amazon’s customer service strategy is basically “here’s your refund, please don’t talk to us”
@almmaasoglu why do you think that is? is there not enough trust in the system to take constrained actions?
Finally someone said it. I came across this a few months ago:
“Product & Marketing ARE your business. Every other function (including CS) exists because of gaps that your market and customers experience in your Product & Marketing.
Trace it back: Sales exists because the messaging is weak and the buying process is hard. Support exists because the product is unreliable or complex. CS exists because the product doesn’t make its value obvious.”
@suchenzang It was a weird corner case issue (i.e. the only type of thing you ever ask to talk to someone about) and to my surprise, the AI chatbot understood it and told me the correct answer.
The company was Anthropic; unsurprising in retrospect.
@suchenzang Appointment booking, it did work perfectly fine when i didn't have any special request.
so... has anyone seen an ai customer support bot resolve an issue firsthand, without escalating to a human or forcing you to give up after a hall-of-mirrors-infinite-loop?
if so, what task (& for which co.) did it complete the above successfully?
Mostly product constraints. It needs access to real tools, permission to take actions and a way to handle edge cases safely. Most companies don’t want llms issuing refunds, changing accounts, or making policy exceptions without guardrails.
At my last contract role, we had an internal bot that could do this. I once lost access to an in house tool, tagged the bot in Slack, asked it to reset me, and it actually restored my access.
It could also handle more action oriented internal requests like admin stuff, routing approvals, shipping items to the right address, etc.
@suchenzang Yes. Just need to narrow down to specific issues (ex, ad campaign blocked because x,y,x) vs expecting the bot to resolve all customer issues.
@yacineMTB @suchenzang "FIXING THE ACTUAL PROBLEMS IN THE PRODUCT and allowing self serve" Could not have been better said. user restoration was the leading category for support tickets and i spent like a week automating it to be self service and now support gets 60% of their time back.
@giffmana @suchenzang Wouldn’t a simple calendar ui be better for that? What’s the advantage of a chatbot here?
@suchenzang never run into a functional one, always need to escalate to human.
@suchenzang I refuse to talk to ai at any organization I need help from. I don't call to talk to a robot, I want the phones to route to the first available human FIRST.
@j4nve @suchenzang for you and me yes, for the maaaaaany people from the past two generations, absolutely not. I did it via call only because I wanted to try out the AI, I hate phonecalls :)
@suchenzang verizon AI bot fixes mobile hotspot issues and payment issues decently well. but no other company has had a helpful ai bot
@suchenzang I have never seen one that takes any action ... the most they will do is refer you to a web page or give you a button that says "call" ... I have never seen one take an action on my behalf
@suchenzang klarna’s refund bot handled a duplicate charge reversal in under four minutes without human escalation
@suchenzang Product/surface - Meta’s ads manager
@suchenzang Yep Verizon bot fixed an issue that their actually people I talked to first failed at