Merchant Onboarding
How an acquiring platform cut time-to-live 38% and auto-verified 84% of onboarding documents.
An enterprise web application that gives merchant-services teams a single-screen command center for the full onboarding pipeline — funnel analytics, KYC/KYB automation, live risk signals, and a GenAI layer that gets legitimate merchants live faster.
What the web application does
Six core capabilities turn a fragmented, multi-tool process into one browser-based workspace — built for the dense, decision-heavy work of merchant onboarding.
Unified web workspace
A responsive desktop app-shell — persistent sidebar, global search, and time-range controls — replaces a patchwork of disconnected tools with one browser-based command center.
Live funnel analytics
Every onboarding stage from application to approval is charted with conversion and drop-off, so teams see exactly where merchants stall — in real time, at any range.
KYC / KYB automation
Identity and business verification run automatically as documents arrive, extracting fields, flagging mismatches, and clearing low-risk merchants without manual review.
Risk & compliance layer
Adverse-media, sanctions/PEP, jurisdiction, and velocity signals surface as ranked, color-coded alerts with explainable, audit-ready reasoning.
GenAI copilot
A grounded LLM reads the live funnel and risk signals to write a plain-language narrative plus a ranked set of next actions — every claim linked back to the data.
Case & queue management
Stalled applications route into review queues with full context, so analysts act on the right file at the right moment without re-deriving what the system already knows.
Explore the live web application
A working slice of the product, rendered as the real desktop app-shell. Switch the time range to watch the onboarding funnel, KPIs, live risk signals, and the GenAI narrative respond in real time. Hover any metric for a plain-language definition.
Welcome back, Alex
Here's what's happening across your onboarding pipeline · as of May 2025
Onboarding Funnel
7D · merchants per stageTop Risk Signals
from screening engine- High
Adverse media mentions
128 flagged for review
- Medium
High-risk jurisdiction
98 applications
- Medium
PEP / sanctions match
54 potential matches
- Low
Unverifiable beneficial owner
41 cases
- Low
Velocity / multi-application
25 anomalies
AI Onboarding Narrative
GenAIOver the last 7D, 12,480 merchants entered the pipeline and 4,860 were approved (38.9% end-to-end). The largest drop-off is "KYC Completed" → "Risk Assessment" at 29%, driven by adverse-media hits and incomplete beneficial-owner data. Recommended action: trigger GenAI document-assist earlier and fast-track low-risk retail and e-commerce merchants to protect approval velocity without loosening AML controls.
Recent applications
| Merchant | Business type | Country | Status | Risk score |
|---|---|---|---|---|
| Global Supplies Co. | Wholesale | United States | Under Review | 620 |
| GreenLife Market | Retail | Canada | KYC Completed | 310 |
| Blue Tech Solutions | SaaS | United Kingdom | Risk Assessment | 580 |
| Prime Goods Trading | E-commerce | Singapore | Info Submitted | 260 |
| NorthWave Services | Professional | Australia | Under Review | 670 |
Metrics glossary
plain-language definitions for every metric on this dashboard- Applications Received
- Total merchant applications that entered the onboarding pipeline in the selected period.
- Approval Rate
- Share of applications that clear every stage and are approved to transact.
- Avg. Onboarding Time
- Mean time from application received to approval decision.
- Fraud Blocked
- Applications stopped by KYC, AML, or risk checks before approval.
- KYC / KYB
- Know Your Customer / Know Your Business — verifying the identity and legitimacy of a merchant and its owners.
- PEP / Sanctions
- Politically Exposed Person and global sanctions screening to prevent onboarding prohibited entities.
- Beneficial Owner
- The individual who ultimately owns or controls a business; must be identified for compliance.
- Risk Score
- A 0–1000 composite of fraud, credit, and compliance signals; higher means more risk.
- pp (percentage points)
- The absolute difference between two percentages. +4pp means a rate rose from, e.g., 70% to 74%.
The Challenge
Merchant-services teams onboarded thousands of businesses a month across application, information capture, KYC, risk assessment, review, and approval. Those stages lived across spreadsheets, email, and legacy admin screens, so applications stalled silently, documents were re-requested, and merchants waited days before their first payment. Nobody could say where the pipeline was losing people.
What We Learned
Shadowing onboarding analysts showed the real cost was invisible drop-off. The largest leak sat between 'Information Submitted' and 'KYC Completed', almost entirely due to incomplete or low-quality documents. It also settled a form-factor argument: this is dense analyst work — side-by-side documents, multi-metric charts, long review sessions — not phone work.
The Strategy
We rebuilt the intended mobile app as a responsive web application: persistent sidebar, global search, and a single-screen pipeline overview no phone layout could carry. On top of it, a multimodal model classifies and validates documents on arrival, extracts business details, flags mismatches, and coaches applicants to fix issues pre-review. A grounded LLM reads the live funnel and risk signals and writes a plain-language narrative with ranked, auditable actions.
How It Was Built
The engagement extended the design system with a reusable app shell (sidebar plus top bar), a conversion-funnel charting pattern, semantic risk-signal tokens (low / medium / high), and a GenAI narrative card — all reused across onboarding, disputes, and settlements surfaces.
The Impact
Median time-to-live dropped 38% and approval rates for legitimate merchants rose 16 points, with 84% of documents auto-verified and the entire pipeline visible on one screen. Earlier fraud detection kept risk metrics flat as volume grew, and merchants started transacting days sooner.
What I Carry Forward
Match the form factor to the work: dense, decision-heavy operations belong in a web application, not a phone. And GenAI earns trust in a regulated flow only when every extracted field and every recommendation is grounded, reversible, and logged.
Behind the work
My scope, the guardrails I worked inside, what the research changed, and how the work travelled from sketch to production.
My role & scope
Lead designer for the 360° merchant view: research with operations and support teams, the unified merchant profile IA, cross-team component library additions, and the analytics surface. Two designers reported to me on this stream.
Constraints & guardrails
- Data stitched from four systems with different refresh rates — freshness had to be visible per panel.
- Existing brand and component library; no new patterns without design-system review.
- Support agents work under handle-time targets, so any added step needed to pay for itself.
- PII exposure rules limited what could be shown on a shared screen.
What research changed
- Call shadowing: agents opened an average of 5 tabs to answer one merchant question.
- Ticket analysis showed a third of escalations were caused by missing transaction context, not policy.
- Testing revealed agents scan for anomalies first, so the profile had to lead with exceptions, not attributes.
The pivot
The first layout was a tabbed profile organised by source system. Agents kept getting lost, so we reorganised around the questions they actually ask — who is this merchant, is anything wrong, what changed — and demoted system boundaries entirely.
UX metric → business impact
Card-sorting output and question-led wireframes drawn from call shadowing notes.
Prototype with loading, stale-data and restricted-PII states, reviewed with the design-system group.
Shipped 360° profile with live activity, risk posture and portfolio analytics panels.
Retrospective
Bring the data owners into the earliest workshops — several freshness constraints only surfaced after the IA was set, forcing rework.
Organise enterprise screens around the user's question, not the source system. Alignment across teams gets far easier once everyone reads the same profile.
“Moving it out of the phone concept and into a real analyst workspace was the call that made the numbers move. One screen, the whole pipeline.”