The struggle was visible in the recordings. The interviews revealed why.
Collateral health, recalculated manually, every time — while the market kept moving
In crypto lending, collateral is what keeps a loan alive during market volatility. For an institutional lender, managing it well isn't easy — it lowers counterparty risk, speeds up compliant lending decisions, and in a volatile market directly affects financial survival.
The process was error-prone by design:
—Fragmented data — hundreds of constantly shifting sources with no single source of truth
—Complex math — multi-step LTV calculations repeated on every market move
—Moving targets — safety thresholds that differ loan by loan, with no visual signal when one is close to breach
The MVP covered loan creation — not what happened after
The MVP addressed the first step well — create a tranche, save a draft, return as the negotiation progressed. What it missed was everything after: monitoring loan health, knowing when a loan was approaching its threshold, acting on it in time. A loan in danger looked identical to a healthy one.
Screen recordings flagged the form problem before any feedback arrived. When MVP feedback surfaced the LTV calculation as incomplete, I had two signals pointing at the same feature — and used that as the case for one redesign, not two separate fixes.
A loan approaching its limit looked identical to a healthy one.
A form too long to navigate, a threshold that didn't reflect reality
1The form had outgrown its original design — the Create Tranche modal started short and logical. As the product grew, so did the form. Screen recordings showed users scrolling up and down, lost in a structure that was never built to hold that much.
2The LTV calculation was structurally incomplete — accounts weren't linked, balances updated manually across dozens of sub-accounts, currency discounts not factored in. A number that looked accurate but wasn't.
3Users' mental model was urgency-based, not data-based — they look for what needs action first, not a full data overview.
Two surfaces, failing the same workflow — monitoring and acting on a loan
1The form needed a logical structure — mapping which steps were mandatory, which were conditional, and which depended on earlier choices. Account linking deserved its own dedicated step — it had its own complexity and sub-steps that couldn't be collapsed into the rest of the flow.
2The LTV calculation needed to reflect reality before the user submitted — linking accounts and applying the correct currency discounts meant the LTV value could become a live signal: visible while editing, not revealed only after saving.
3Urgency had to be the default read, in both surfaces — the table surfaces the most critical loans first, the form shows the live LTV position as values change. Both communicate risk before the user has to look for it.
Create Tranche: from a static form to a live calculator
The previous modal supported quick edits over days. As the form grew, that mental model broke. The wizard fixed the structure, but at a cost: users leave the table to enter a dedicated flow. A navigable side panel with labeled sub-steps compensates — structure and progress always visible, no matter where they are in the process.
What held, what moved elsewhere
Walked the screens through a structured review with an operations manager who'd run institutional lending desks in both traditional and crypto finance.