Case study · Collections & hardship
Redesigning the delinquency experience at a top-five US card issuer so that getting help no longer required a phone call, and lifting self-serve completion 30%.
If you read nothing else
The issuer had real help for people behind on their cards: payment plans, reduced-rate programs, ways to bring an account current. But the most meaningful help was reachable mainly through an agent, at the end of a linear, one-size-fits-all flow, and the people who most needed it were exactly the people least likely to call. I owned the redesign that brought help into the digital channel, sequenced friction by who actually needed it, and lifted self-serve completion of those flows by 30%. The premise was simple: on someone's worst month, the phone is not a channel, it is a wall.
Collections is the part of a credit card almost nobody designs for, because it is the part everyone hopes not to see. But at the scale of tens of millions of customers, some meaningful number are behind on their card in any given month, most of them for ordinary human reasons: a job loss, a medical bill, a bad stretch. The bank has programs built for exactly this. My job was the experience through which a person in that month actually finds and enrolls in them, and when I picked it up, that experience quietly assumed the customer would call us. Most of them never did.
The gap was between what the bank offered and what a person in distress could reach. The stronger forms of help sat behind the agent channel, and the digital path that did exist ran everyone through the same linear sequence regardless of their situation: a person who just forgot a payment and a person three months into a hardship saw roughly the same doors in the same order, with the most consequential ones requiring a conversation.
The data said what the design would not: people entered the flow, saw where it led, and left. Some of them charged off months later without ever having touched the help that existed for them. That is the failure mode that gets me, because it is invisible from inside the building. Nobody sees the person who did not call. The business sees a roll rate; the person sees a phone number, and shame does the rest. Anyone who has been behind on a bill knows the specific dread of that call: explaining yourself to a stranger, on hold, during work hours, about the worst thing in your month.
The help existed. The path to it assumed a phone call that the people who needed it most were never going to make.
The old path ran everyone through the same sequence toward a phone call most people in distress will not make. The new one meets them in the channel they are already in, and saves the friction for the steps that genuinely require it.
The bet was that friction should be placed by necessity, not by default. Bring real help into the digital channel where the person already is, at two in the morning, without an audience. Sort the experience by the customer's actual situation instead of running everyone down one sequence. And keep the heavier steps, the ones the programs legitimately require, only where they are genuinely required, instead of letting them gate everything.
What I chose not to do mattered as much. I did not push to strip friction everywhere, because in this domain some of it protects both the customer and the bank, and a reckless one-click version of hardship enrollment would have been wrong and would never have cleared the room. And I did not accept the other comfortable position, that serious help should stay with agents because agents convert better in conversation. That argument quietly optimizes for the people willing to call, and abandons the larger group who will never dial. The senior call was to treat "customer will not phone us" as a design constraint as real as any regulation, and build for it.
The people most likely to need help are the least likely to ask for it out loud. A flow that requires asking out loud is selecting against exactly the people it was built for.
Three rooms had to agree, and they started far apart. Credit risk saw easier access to concessions as more concessions, which is more cost. The operations side saw digital enrollment as volume leaving a channel they are accountable for. Legal and compliance saw any change to these flows as risk in a domain where mistakes are not just bugs.
The argument that moved it was not experience, it was economics told honestly: the alternative to a customer enrolling in a plan digitally is mostly not that customer enrolling through an agent. It is that customer doing nothing, rolling deeper, and eventually charging off, which is the most expensive outcome on the table for everyone, the customer most of all. Framed that way, the redesign stopped being a customer-experience nicety competing against loss numbers and became a loss-prevention case with a better experience attached. Compliance shaped which steps were load-bearing and stayed; ops kept the genuinely hard cases, which are the ones agents are best at; risk got instrumentation showing who was enrolling and how the plans performed. My work was holding one version of the story that all three rooms could sign, and sequencing the build so trust was earned in stages rather than demanded up front.
Self-serve completion of the redesigned flows rose 30%. Behind the percentage, the human translation is the point: meaningfully more people on a bad month reached help without having to say it out loud to a stranger, and fewer quietly dropped out on the way to a phone call they were never going to make. Honest attribution: the programs themselves long predate me, credit risk and compliance shaped what could be offered and how, and engineering and design built it. My part was owning the thesis, the flows, and the argument that got them shipped.
What I would do differently
I would have started measuring the after sooner. Completion tells you someone enrolled; it does not tell you whether the plan held six months later, and enrolling in help you cannot sustain can leave a person worse off than they started, with another failure attached to their name. The instrumentation for plan performance came later than it should have, and I now think of enrollment not as the finish line but as the moment the product's real obligation begins.
Completion is the metric. The person quietly fixing the worst month of their year from their kitchen table, without an audience, is the product.
This is the work I find most interesting in all of consumer finance, and the reason is in the asymmetry: the front of the credit business gets the design attention, and the back gets the people. Every lender will tell you they support customers in hardship. The test is whether that support survives contact with how people in hardship actually behave: quietly, late at night, avoiding the phone, hoping not to be seen. Building for that person is not a niche of payments. As credit expands into every checkout and every wallet, it is quickly becoming the whole job, and it is the one I want.