Testing one API took eight steps. I made it three.
Porting five old demos to the cloud portal was the easy answer. I rebuilt the playground instead, around the one question a developer brings: does this API deserve a place in my stack?
Problem
You could only watch the demos.
Developers could look at examples, but they couldn’t type their own input, change a field, or judge fit without leaving the portal. The cloud migration was the right moment to replace that model.
- Fixed inputs, separate maps, and raw responses.
- Nothing on the page came from the developer’s own data.
See the old portal
Emergency Info API, old portal.
Fixed inputs, disabled fields.
What I found
Five demos. One broken evaluation.
I mapped one API test end to end. From the third step on, the developer was working in an external tool.
- Every tryout sent developers back to the docs before they could judge a result.
See the audit map
- Find docs
- Download spec
- Open tool
- Import spec
- Configure auth
- Set inputs
- Send
- Interpret
- Choose API
- Enter real input
- Read result
What I decided
Stop rebuilding the demo. Build the system.
One tryout pattern that bends to each API. Developers learn it once, and a new API becomes a configuration of the pattern.
- Context, inputs, map, and result stay in one workspace.
What it cost
No API got a fully custom demo of its own.
Proof
One page logic runs all five shipped demos.
What I decided
Their data, not our sample.
Developers could check coverage first, send a real request with their own coordinates or filters, and see exactly what came back.
- Categories and availability replaced a blank input box.
- I proved the shape on Reverse Geocode first: one input, one map, one result.
What it cost
Real inputs meant real edge cases showed up mid-evaluation.
Watch the simplest demo
What I decided
One system. Five levels of complexity.
Time Zone and Wi-Fi needed a light map. GeoTAX needed far more setup. The same pattern had to handle both without forking.
- Maps and responses stay legible without changing the core flow.
What it cost
GeoTAX’s extra controls stay hidden until they’re needed, so the simple demos stay simple.
Proof
GeoTAX runs in the same workspace, with no separate interaction model.
What shipped
The workaround disappeared. The pattern survived.
The wider program tested it with eight product managers, developers, and sales engineers, then put API Tryout into production. New APIs reuse the pattern.
To a first API response
First-session API success
Up from about 60%.
Fewer support requests
Onboarding, after API Tryout went live.
Estimated and reported by stakeholders and Solutions Engineering, not instrumented.
The portal went on to win two 2026 Global AI Awards, in Agentic AI and Conversational AI. The award belongs to Precisely’s product, and several designers worked on it, me included.
Looking back
Next, I measure it in the product.
Next time I’d instrument time to a first successful request in the product instead of relying on team estimates. I’d also tree-test the catalog: I left its names and grouping as they were, so a developer who doesn’t know what GeoTAX means can still move fast in the wrong direction.
Beyond the brief, I built a working prototype in Figma Make and GitHub Copilot on live APIs, so the direction was approved on real responses. I added reusable components to the Precisely Design System and, with Jerimy Brown, brought Claude and Copilot workflows into the whole design team.