GAAPT is a forecasting platform for an audit and assurance business: leadership uses it to see revenue, profit and headcount years ahead, and finance teams feed it the data. The legacy tool showed everything at once: one endless page of charts, tables without structure, country filters that disappeared once applied. As the product's only designer after discovery, I talked to users twice a week, cut what nobody used and rebuilt the dashboard, tables and admin tools around how people actually read a forecast.
Enterprise forecasting tool for an audit and assurance business. Legacy system redesigned end to end.
GAAPT forecasts revenue, profit, headcount and resources for an audit and assurance business several fiscal years ahead. Two groups use it: senior leaders who check the forecast for their countries about once a week, and finance teams who enter the numbers.
I was the only designer on the product. A design lead joined discovery and the first delivery; after that, discovery, communication with the client and the redesign of every page were mine, including working through the data to decide what each screen should show.
The data was never the issue: there was a lot of it, before and after. The issue was that all of it arrived at once. The dashboard was one long page of charts in no particular order, so the chart you needed sat somewhere below the ones you didn't. Charts were hard to read: one bridge was split into three separate graphs, and the numbers in the headcount pyramid only appeared on hover. Tables ran sideways past the edge of the screen. And the countries a user picked before opening the dashboard were nowhere on it.
We spoke with users about twice a week in live calls, with both the leaders and the finance teams, going through the problems they had hit that week.
Users relied on a handful of views. The rest only added scroll and load time, so we cut them or moved them down.
Users chose countries before opening the dashboard, then couldn't see the choice. One mistake meant going back and starting over.
The data-entry table was one long sheet, yet it split into clear blocks, and each level of manager needed a different depth.
Countries were chosen on a separate step and disappeared once the dashboard loaded.
The selection now lives in the dashboard header as chips, with Edit next to them. Users see what the numbers include and change it in place, without going back.
Every chart sat on one long page, in no particular order.
We grouped the charts into four tabs by the question they answer: Profit & Loss, Gross Margin, Headcount and Transformation Levers, and removed the charts nobody used. Less scrolling, and each tab loads only its own data.
The table was a wide sheet of near-identical rows that scrolled past the edge of the screen.
Rows are grouped into blocks that open from total to subtotal, offering and country. A Show control sets the depth in one step, so a leader stays at the total and an analyst opens every country. All ten years fit on screen.
Two patterns repeat across the dashboard. Bridges show the start, every change and the end in one waterfall, with signed values above each bar and hatching on decreases. The headcount pyramid puts the number, the share and the change next to every grade, so nothing hides behind a tooltip.
Admins needed a way to warn every user about maintenance, deadlines or outages. Notification Center lets them write a banner, pick its level, schedule it and preview it before publishing. The form checks dates and required fields inline, so errors are fixed before anything goes live.
User management moved to the client side too: site admins now add and remove users themselves, a job that used to go through developers and the backend.
One dark system runs across every screen: a muted six-colour palette for offerings where every colour passes 4.5:1, matte glass in three densities, and Geist with Geist Mono for numbers.
No usage metrics were collected. To measure the impact I would track the time it takes to change the set of countries, the number of access requests sent to developers, and dashboard load time.
Back the cuts with data. We removed charts because users said they didn't use them; usage analytics would have made that call certain.
Test the table with the people who fill it in. Most of our calls were with leaders, while the table is the finance teams' main tool.