FinTech UX design, from inside banking products
Most writing on FinTech UX design is written from the outside. This one is not. It comes from twelve years designing the screens people are accountable for — credit risk monitoring for a $12.8B commercial book, merchant onboarding for a national bank, and the review queues behind both. What follows are the principles and patterns that survived contact with regulators, risk officers and real portfolios.
Six principles that hold up in production
1. Design for the decision, not the dataset
Financial screens fail when they present everything with equal weight. Start from the decision the user has to defend to someone else — approve this merchant, downgrade this cohort, release this payout — and let that decision set the hierarchy. Everything that does not move that decision belongs one level deeper.
From the work: On a $12.8B commercial credit book, replacing a static score snapshot with cohort trajectories cut modeled default exposure by 31% and made review cycles 2.1× faster.
2. Latency is a UX problem before it is an engineering one
Overnight batches quietly turn every intervention into a post-mortem. If the data behind a screen is a day old, say so on the screen, and design the fastest available signal into the primary view. Freshness labels are cheap; misplaced confidence is not.
From the work: Risk teams read the book through spreadsheets and nightly jobs — deterioration surfaced only after accounts had already slipped.
3. Make every number traceable
In regulated products, a figure a user cannot source is a figure they will not act on. Every headline metric should link to the ledger, the rule, or the record that produced it. This single pattern does more for adoption than any visual refinement.
From the work: Analyst adoption rose 38% once trend charts were drawn directly from the risk ledger with visible provenance.
4. Onboarding is a compliance funnel — instrument it like one
KYC and underwriting steps are where applicants leave. Split long forms by the source of the answer (what the applicant knows, what a document holds, what a system can fetch), pre-fill everything you can, and never ask twice for the same fact. Measure drop-off per field, not per page.
From the work: Restructuring merchant onboarding around document-driven steps and system pre-fill removed whole rounds of back-and-forth between applicants and reviewers.
5. AI must cite, or it must stay quiet
Generative summaries are genuinely useful in finance — they compress hundreds of documents into a sentence. They are only usable when grounded: retrieval over the actual risk documents (RAG), a visible source list, and a link from every claim to a figure already on screen. Log every accept and dismiss so the model is auditable.
From the work: The grounded agent in Credit Risk 360 links each statement to a visible figure and records analyst accept/dismiss actions.
6. Accessibility is risk management
Dense financial UI is where colour-only status, unlabelled controls and broken focus order concentrate. WCAG AA is the floor: colour never carries meaning alone, every chart exposes a text summary, and keyboard operation covers the full workflow — including any AI panel.
From the work: Both the risk dashboard and the onboarding flows were rebuilt to WCAG AA with text equivalents for every chart.
Four patterns worth stealing
| Pattern | Use it when |
|---|---|
| Cohort trajectory over point score Show movement across periods with a reference band for policy thresholds. A score answers 'where is it?'; a trajectory answers 'where is it going?' — the question people actually act on. | Portfolio monitoring, collections, fraud triage |
| Explainable alert card State the trigger, the contributing factors, the confidence, and one primary action. Include a dismiss reason — dismissals are the highest-value training signal you will collect. | Any threshold breach or model output |
| Document-driven step Ask for the document first, extract what you can, then show the applicant a pre-filled review instead of a blank form. Confirmation is faster and more accurate than entry. | KYC, underwriting, merchant onboarding |
| Reviewer queue with reason codes Give reviewers structured reasons rather than free text. Reason codes turn a queue into an analytics source and shorten every downstream appeal. | Manual review of applications or exceptions |
A working sequence for a new FinTech surface
- Name the decision and its owner. Write the sentence the user must be able to defend. If you cannot write it, the screen has no job yet.
- Map the data behind it. Source, freshness, and who can be asked when it looks wrong. This determines what the screen is honestly allowed to claim.
- Sketch the information architecture before pixels. One primary decision, supporting evidence, then the long tail.
- Prototype the unhappy path first. Rejections, missing documents, stale data and model uncertainty are the majority of real sessions in finance.
- Test with the people who carry the risk. Credit officers and compliance reviewers will find the flaw a usability lab never will.
- Instrument the outcome, not the click. Exposure avoided, cycle time, drop-off per field, dismissal reasons.