What a viral fintech support story says about the value customers lose their credit points without ever being told.
A post went around fintech circles recently. A senior product executive, someone who has built card products at three well-known institutions, lost her card and tried to get a replacement. It took 25 messages over 25 days.
Along the way she was charged $40 for a metal replacement she never asked for. She was passed between support agents, each of whom needed the whole story retold from the top. At one point She was asked to continue over ordinary email so she could share personal details. And when the account finally closed, her 108,000 reward points were forfeited. No warning. No window to redeem them first.
That last part is the one worth sitting with. The $40 was annoying. The 25 days were exhausting. But the points were value she had already earned, over months of ordinary spending, and they evaporated at the exact moment she had the least attention to spare. She was not told it was about to happen. She was not asked whether she wanted a week to use them.
Four things went wrong in that exchange, and most of them are unremarkable on their own. Every handoff between agents reset the conversation to zero, so she ended up being the only continuous memory in the process. Closing the account was reversible right up until it wasn’t, and somewhere in that flow a single sentence would have changed the outcome: you have 108,000 points, here’s what they’re worth, do you want to use them before we proceed. She was also asked to move to ordinary email to share personal details, the kind of call someone makes when they’re trying to be helpful and the system hasn’t given them a safer option. And a former product manager at the company put the underlying issue well in the comments: there’s a difference between following the letter of an agreement and using judgment about what a situation actually calls for. Only one of those keeps a customer.
Why Most Customers Never Complain, They Just Leave
The comments ended up saying more than the post itself. A chief product officer at a competing fintech admitted she keeps a mental blacklist of brands she’ll never use again after a single bad support experience, which is close to how most people actually make these decisions. But the sharper observation came from a customer experience leader who has run support at two companies most people would recognize: for every person who writes a post like that one, hundreds or thousands more just stop using the product. They don’t complain. They don’t fill out a survey. They open the app less, then not at all, and the card sits in a drawer. It never registers as a complaint. It only shows up later, as attrition, in a number nobody can trace back to a cause.
That’s really the same story told from the other direction. A company loses customers it never learns about. A customer loses money they never learn about. Statement credits reset on schedules nobody memorized, points vanish at closure, subscriptions renew after someone thought they’d cancelled, refunds go unclaimed because a window closed while the receipt sat in an inbox. All of it is real money, and almost none of it comes with a notification.
What we actually build against
We are wary of writing the sentence where AI arrives and fixes customer service. It does not, and in several of these failure modes automation makes things worse rather than better. So here is the more specific version: four mechanisms, matched to the four failures above.
Memory that lives with the account, not with the agent. Our assistant is built so that context does not reset. It reads from connected account data first, then from what you have told us before, and only asks you when neither has the answer. That ordering matters more than it sounds. We also hold a hard line on the failure that lives on the other side of this: when the data is not there, the assistant says so rather than producing a plausible number. Confidently wrong is worse than honestly blank, particularly about money.
Watching the irreversible edges. Most financial value does not disappear randomly. It disappears at predictable moments: a credit that resets at quarter end, a trial that converts on day 31, a closure that zeroes a balance. Those moments are knowable in advance, which means the warning is an engineering problem rather than a customer service one. The design rule we work to is that nothing irreversible should happen quietly. If something is about to expire, convert, or be forfeited, you hear about it while you can still act.
No channel downgrades. Sensitive information stays inside the product. There is no version of a support flow where the fix is “email us your details.” That is a constraint we accepted early because the alternative is a system where the safe path and the fast path diverge, and under pressure people take the fast path. We also do not sell data, which is less a feature than a precondition for any of the rest of this being trustworthy.
Knowing when a machine should stop. This is the failure automation is worst at, and we would rather say so plainly. Rigid policy applied by a person is frustrating. Rigid policy applied by software is frustrating at scale and at speed. The only real answer we have found is to design the escalation path first: some merchants and some situations get handled automatically, some get handled with a human in the loop, and some get handed to a person entirely. Deciding which is which is the actual product work.
The part that should not need saying
The executive in that story eventually got her card. She also got a post with thousands of engagements, which is a form of leverage almost nobody has.
What we keep coming back to is how ordinary the failure was. Just a series of individually reasonable decisions that added up to a customer losing 108,000 points she had earned. Value slipping away at the edges, without anybody deciding it should.
That is the thing we built Amalgamic to stop. Not the dramatic losses. The quiet ones.
Leave a Reply