Case 03Farsight · BaxEnergy, a Yokogawa company

Analysis Pages

Seventeen analysis pages, read by engineers and executives on the same screen. Before I could design them I had to learn how a turbine turns wind into power.

RoleLead Designer
ProductREMS · wind analytics
Pages17
Pages live1715 analyses, the first page, the hub
The 15-page push~45 daysfrom Jira story to front end
Time per page2 wk → 2–3 dwith the new AI work loop
Types of graphs+20I wrote an AI graph generator skill
1Two readers

A strip for the executive, charts for the engineer

An asset manager asks why an hour earned less than it should. A performance engineer asks which turbine is off its curve. So every page runs in one order: headline numbers first, for the reader who won't open a chart, then two or three charts per screen, with room between them. Experts can live with simplification, as long as nothing is missing: they'd rather have too much information than too little.

The same order on every page. Five strips from five pages, then the chart under each, redrawn from the Figma file with its sample data. Click to pause.
The Wake Effects page, full screen: header, a four-number strip, turbines ranked by wake loss against the farm average, and a map of the site with each turbine sized by energy potential and coloured by wake loss.
Wake Effects, the whole screen. Strip, ranking, map. Figma, sample data; I blurred the map's place names.
The North Azimuth Offset page, full screen: a four-number strip, the offset trend per turbine, active power over time beside the turbine locations, and a heatmap of daily offset per turbine.
North Azimuth Offset, the whole screen. One strip, then four charts in three rows. Figma, sample data.
2Power curve

I drew the power curve as a bell. It's an S.

Clients used to need a BaxEnergy engineer beside them to understand these analyses. The maths was the CEO's and the data the BI team's; my job was charts a client could read alone. That meant learning the physics first: kilowatts versus kilowatt-hours, turbulence, and why wind reaching a turbine hundreds of metres behind another arrives messy.

For the first chart on the first page, I pictured a bell. The curve is an S: nothing below cut-in, a climb, a flat top, then shutdown when the wind could damage the machine. Too little wind is bad, and so is too much. The best of it sits in a Goldilocks zone in between.

readings, 10-minutemanufacturer's curvewhat I drew first
The bell I drew, and the S it turned out to be. Rebuilt from the Wind Analysis page with sample data: the lower readings come from a demo turbine, the rest are synthesised to match.
3Dynamic Yaw Deviation

Corrected twice, better both times

Yaw deviation is the gap between where the wind comes from and where the turbine points. My first version was one line of angles over time, and to me that was enough.

Version 1 · one chart

Version 1: one line chart, Nacelle versus Zero-point Angle, of the zero-point angle over five weeks of July and August, with a shaded band where yaw deviation was flagged.
The first version. One line of angles over five weeks, the flagged stretch shaded. Figma, sample data.

A developer asked, Why not show the nacelle turning? I didn't think it was needed. We talked it through, and I changed my mind.

Version 2 · the nacelle, turning

Nacelle versus wind angle over time, with a polar chart on the right that draws the nacelle and the wind direction and shades the deviation between them.
Shipped. Pick a moment, see wind and nacelle. After launch, we learned that one moment wasn't enough.
In this case, the developer was right.

Version 3 · every angle

Nacelle and wind angle over a month as two lines, with severity markers, and a wind rose showing mean absolute yaw deviation by wind sector.
Where it is now. Every wind angle in the window, so you see which directions this turbine handles badly. Figma, sample data.
4AI Assisted Process

Two weeks became two or three days

The first page took one to two weeks, most of it learning, and I showed the team every draft early. Then a client pulled fifteen analyses forward and I ended up in charge of all of them, with the BI team. The design system was in code by then, so I changed how I worked:

  1. BriefI write it; Claude explains the concepts I don't know yet.
  2. PrototypeClaude builds it in HTML, already on the design system.
  3. Into FigmaClaude writes the screen into Figma through MCP.
  4. Fine passI adjust by hand: spacing, hierarchy, detail.
  5. Back to codeClaude reads the frame and corrects the HTML.
  6. Two artefactsFigma and HTML at the same fidelity, with the error states Figma hides.

All seventeen pages are live. Clients now read these analyses without an engineer beside them. When a turbine stops reporting, we leave the gap in the chart. We don't invent data.

5Hindsight

What I'd do differently

  1. Sit with an engineer before the first sketch.My first power curve was wrong because I hadn't understood it yet. An hour with someone who reads these curves every day would have saved the draft.
  2. Test each chart on a month of real data before launch.The nacelle drawing worked for a single moment and fell short once people read a whole month. We found that out after launch, which is the expensive time to find things out.
  3. Decide how to measure it before shipping.Seventeen pages are live and I never saw usage data or test results. I'd agree what to measure with the team before the first page went out.