An AI-Powered Dashboard That Turned Scattered Data Into Decisions

My Role

Sole designer, zero to one

Users

AMCs sit between the lender and the appraiser

A lender orders an appraisal. The AMC assigns it to an appraiser from their panel, tracks it through review for quality and compliance, and delivers the report back before the lender's deadline.

The Problem

Dashboards, disconnected

AMCs had to analyze their own performance daily, but the numbers lived across scattered dashboards with no single source of truth. Most people never dug past the headline stat, so metrics moved with no visible cause, insights weren't actionable, and real opportunities went uncaught.

Research

Turning user pain points into validated product metrics

I led stakeholder workshops and interviews with AMC users to identify pain points, define the metrics that mattered, and prioritize what to build. These points surfaced:

Research

A profile for every role

Each role got its own profile: user description, workflow, behavioral traits, motivations and success criteria, and pain points, grounded in real conversations with named stakeholders at AMCs like Equity Solutions, Ascribe, CMG, NAF, and Flagstar.

The Reframe

Not a data problem. A clarity problem.

AMCs weren't missing data. They were drowning in it, scattered across screens, with no path from a number to a next action. The opportunity wasn't more data. It was the right data, consolidated, pointing at what to do next.

Research

Competitive analysis: none of them answered "why," or "what do I do about it?"

Process performance, rule-based operations, workflow visibility: real, but none of it explained why a metric moved, and none of it was predictive or actionable.

Final Designs

Dark mode and light mode

Final Designs

A structured, filterable system

Terms & Conditions

Shipped, but no one could get in

Cogent surfaces sensitive competitive data (fees, margins, operational metrics) across AMC and lender accounts. Explicit consent was legally required before access. Acceptance rates were low, and low acceptance meant no usage.

Terms & Conditions

I pushed back on gaming the acceptance rate

My PM's instinct was to inflate acceptance with a vague CTA and an obscured terms link. I pushed back: if we're going to suggest decisions to users, acceptance has to be genuine, not a UX trick to hit a number.

Terms & Conditions

Two theories, tested in parallel

Rather than betting on one theory and losing time if wrong, I built both paths at once, and let the decline reason tell me which theory was right.

Terms & Conditions

One authorized "yes" for the whole org

A targeted consent flow shown only to organizational decision-makers already active on Core. A single acceptance unlocked dashboard access for their entire team.

Terms & Conditions

Never stuck waiting on one inbox

If a decision-maker hadn't acted yet, individual users could still see the terms and accept directly if authorized, or decline with a reason, turning a dead end into a data point.

Terms & Conditions

Acceptance journey

Terms & Conditions

Decline journey

Testing and Feedback

The strongest signal wasn't a metric

Monitored with Clarity, Hotjar, and Mixpanel, plus direct client conversations. After shipping Revenue, clients started asking for the same clarity in Turn Time and Revision, before those dashboards even existed.

Reflections

Why a modal, not a side panel

Reserved the side panel for a not-yet-built AI workflow agent (operational, in-flow). Cogent Insights is analytical and reflective, a moment to step back from the workflow. The modal reinforces that distinction, even though it costs temporarily obscuring the dashboard.

Reflections

What I'd carry into the next one

Why This Matters

Design + engineering, at the same altitude

This project is the shape of how I work: a collaborative, cross-functional practice; technical and AI-first execution, from Figma through shipped front-end code; and a senior design voice that pushes back when it matters and builds when it counts.