Case study · 02
A consultation an AI can run
Fragmented metabolic data turned into a three-act guided consultation that a human coach, or an AI agent, can deliver end to end.
Context
After a metabolic test, a coach or clinician sits down with their client to explain the results and agree on a plan. This conversation is where the data finally gets explained to a person, and where follow-up services are recommended. It is probably the highest-stakes moment in the product. The health-tech company behind the platform wanted the conversation to be well structured and consistent, and deliverable by an AI agent as well as a human. The consultation lives inside the clinic CRM and adds to it without replacing anything. It reorganizes data that already exists into a single, presentable flow.
The problem
There was no consultation surface at all. To walk a client through their results, coaches improvised a tour across individual system pages, biomarker views, report screens and a separate programming section, and stitched the story together while presenting it live. The quality of the conversation depended on each coach's improvisation, and an experience with no fixed structure can't be automated. An AI agent needs a linear, step-by-step interface it can operate, not a scattered set of screens.
- The data is dense and clinical, from VO2max to heart-rate-variability ratios to fat-versus-carb burn, and it has to be explained in plain terms to people who aren't experts.
- Every client's data is different. A resting test only, an exercise test only, both, with or without a body-composition scan: each combination changes which systems and metrics exist.
- Coaches had to keep all of those combinations in their heads while presenting, and every dead end happened in front of a paying client.
Approach
Absorb the science
The brief was a product spec built on about fifteen reference documents of scoring and biomarker logic. Before drawing anything, we had to understand what each test measures, how tests taken within seven days are grouped into one comparable bundle, and which metrics exist under which data conditions.
Structure before screens
The central design move was a story structure. Act one covers who you are: biological age and the main system scores. Act two covers why, system by system. Act three covers what to do: nutrition, training and bio-optimization. Sixteen subsections sit on a two-level stepper, which works like a script that a human or an agent can play scene by scene.
Answer the open questions in mockups
The spec left ten questions open for design, including the page layout, how the detail panel appears, how systems contribute to biological age, how the human figure behaves and what starts AI mode. Explorations sit side by side on the board, with variants for one-test and two-test data and for the AI playing or paused.
Consolidate and extend
The resolved screens were turned into components: stepper bar states, bio-age widgets, metric cards, system scores, and service and workout cards. On that basis, a full mobile version was designed ahead of the MVP scope, with the desktop layout restacked into a vertical, sheet-based flow.
Key decisions
Making you younger, making you older
The spec asked how each system contributes to biological age, but no exact weights exist. Instead of showing false precision, we split the systems by score into two plain-language groups placed on either side of the bio-age clock: what's making you younger and what's making you older. The groups are easy to read emotionally in a single glance, and they stay scientifically honest.
One stage, two drivers
AI mode uses exactly the same screens as a human-led consultation. Only the chrome changes: a voice indicator and pause/play controls appear, the system being discussed is highlighted, and the manual controls are de-emphasized but never hidden, so a person can take over at any step. Building AI control into the interface itself, rather than designing a separate set of AI screens, kept the product coherent and maintainable.
A stable stage with progressive disclosure
Act two uses a fixed three-column layout: the systems list, an inline detail panel and a human figure. Only the middle column changes, so the view never jumps around during a live session. Each panel is ordered the way a consultation is spoken: score and tier, services, influencing factors, effects on other systems and what you're likely experiencing, with raw biomarkers and test history last. The client's bio age stays on screen throughout and becomes a current-to-target figure in act three.
Outcome
Preparing for a consultation used to mean assembling a tour across the product. Now it means pressing Get started. Every coach delivers the same structured story, with client-facing language built in. Every prescribed service, workout and nutrition target is labelled with the system it improves, so the plan always traces back to the reason for it. The experience adapts to each data combination, hides what doesn't exist and shows a clear empty state when there are no tests. Because it is linear and addressable step by step, an AI agent can now run it. This is an MVP-stage design, so no usage figures are claimed here.
Client names and screens are kept confidential; illustrations are abstract recreations.
What we learned
Fragmented data wasn't fixed by a better dashboard. It was fixed by a story structure, with one protagonist metric that stays on screen from the first step to the last. Designing for an AI operator meant treating it as a first-class user of the same interface. We also chose honest data over impressive data: the training chart shows discrete zones instead of a granular curve, because that is what the data can reliably support.