The story in 30s
The problem. Entromy's report page, the most important surface in the product, behaved exactly like the PowerPoint it exported: over 96% of report views followed a pattern of open, export, leave, no filtering, no drill-downs, nothing the PPT couldn't already do. Report customization was one of the main requests across 20+ analyzed customer calls. Clients called building leader-ready cuts "hard to operationalize." Underneath that, the platform was built on hardcoded, one-off components with no shared design system, so every new interaction or visual improvement had to be built from scratch, and consistency across the product was nearly impossible to maintain.
What I did. I framed this as two problems that had to be solved together. On the business side, usage was infrequent: clients logged in once or twice and left, which put renewals at risk. Executives still needed the PowerPoint export for their own workflows; that wasn't going away, but the platform itself wasn't giving anyone a reason to stay and explore once they had it. On the product side, we needed a foundation that could support real interactivity without fighting hardcoded UI at every turn. So I sequenced the work in two phases: first, migrate legacy components into reusable, design-native building blocks, for example, tables that once lived as one-off hardcoded elements were wrapped in a reusable card component, so new interactions could be layered on without rebuilding from scratch. Second, once that foundation existed, apply one interaction model platform-wide: overview first, zoom and filter, details on demand, and layer in an AI advisor chatbot so users could interact with the data conversationally, not only through manual drill-down. I wrote the platform's first data visualization guide to make this repeatable, mapped real user roles to dashboard types, and redesigned six report sections on the reusable component base.
What changed. The PowerPoint export stayed exactly where it was. Executives still use it. What changed is that clients now also engage directly in the platform alongside it: over half of reporting sessions now run past ten minutes, someone drilling into detail, asking the AI advisor a question, and working the data instead of just exporting and leaving. Dashboard views per user rose 41%, average session time rose 144%, and client-facing teams now call the redesigned UI "a game changer" for client conversations.
The thinking: a system before any screens
The biggest risk wasn't visual design: it was building on top of the same hardcoded foundation that caused the problem in the first place. So the first real decision wasn't a visual one: it was getting product, design, and engineering aligned on migrating legacy elements into reusable, design-native components before adding any new interactivity. That took real back-and-forth to land, engineering needed a migration path that didn't stall other work, and design needed the resulting components to actually support the interaction model we wanted, not just look consistent. Once that foundation existed, the data visualization guide gave the redesign a shared vocabulary:
- → One model, not six. Every surface follows Shneiderman's mantra, overview first, zoom and filter, details on demand: users see overall context and patterns before specifics, narrow down with filters, and pull detail only when they ask for it, instead of being overwhelmed upfront.
- → Depth by role, not by guess. Decision-makers (CEOs, CHROs, PE firms) get high-level scorecard views built for fast reads. Analysts (managers, consultants) get detailed views built for exploration. This mapping came from actual behavioral personas, not assumptions about titles.
- → Same data, same chart, everywhere. Gauges and indicators for scores, line charts for trends, stacked bars for distributions, ranked lists and tables for comparisons, so the same kind of data always looks the same way anywhere in the product.
- → Color means something, or it isn't used. Categorical palettes for unordered groups, sequential for progressions like participation, and divergent palettes reserved for data with a meaningful midpoint, like eNPS running from -100 to 100.
Solution overview
The clearest illustration is the eNPS flow, three levels deep, each one earning the next. Learning this pattern once unlocks the entire report. That mattered enormously for a product people open a few times a year, where every visit is close to a first visit.
- → Overview: enough to stop there. The dashboard card shows the score, its quartile, and movement against benchmarks. A decision-maker can stop here, fully informed.
- → Zoom and filter: one click deeper. Clicking through opens the full eNPS page: promoter/neutral/detractor breakdowns, detractor insights ranked by gap, and the dimensions where detractors concentrate.
- → Details on demand: invisible until asked for. From there, individual comments, score calculations, and recommendations open in popovers and modals, present when asked for, invisible when not.
Results
- → Over half of reporting sessions now exceed 10 minutes, up from a pattern of opening the page just long enough to export, someone reading, drilling into detail, and working the data
- → Multiple "strongest week on record" cycles, with report clicks and exports both jumping roughly 90% week over week at peak, driven by admins actually using and distributing the new views rather than just generating them
- → Dashboard views per user increased 41%, average session time increased 144%
- → Org admins now drive 93% of report activity, confirming the redesign landed with its primary audience
Reflection
- 1 The foundation had to come before the features. Migrating hardcoded elements into reusable components, like turning a one-off table into a reusable card, wasn't visible in the final UI, but it's the reason six sections could ship as one consistent product instead of six different ones, and why the pattern still holds as new surfaces get added.
- 2 The goal was never to replace the export, it was to give the platform its own reason to exist. Executives still pull the PowerPoint; that's a real workflow and it wasn't our job to kill it. The win was making the in-platform experience valuable enough, drill-down plus the AI advisor, that clients engage there too, instead of the platform being a one-time stop on the way to a static file.
- 3 Consistency compounds. In a product people open a few times a year, one interaction model that repeats across sections means the second section is already familiar by the time someone reaches it.