Complex products, made obvious.

Context

Two audiences, one menu, and neither was served well.

The platform connects allocators (LPs) and fund managers (GPs). Their goals differ, their workflows differ — but the navigation treated them as one undifferentiated user, and had accumulated entries as features shipped.

What was wrong

The menu described the product's insides, not the user's intent.

• Duplicate entries — several different ways into Meetings.
• Three names for the same thing: Lists, My Lists, Watch List.
• Categories mirroring internal structure rather than mental models.
• Long paths to Funds, Managers, Messages, Meetings and Activity.
• Fragmented, inconsistent experiences between LPs and GPs.

The goal

Not a better menu — a faster path through the core workflows.

Reduce cognitive load, increase confidence while navigating, and stay scalable as features keep arriving. The business goal behind it: become the place managers think of first when looking for allocators.

Decision 01

Organise around the cycle: Discover → Connect → Chat.

The navigation should mirror how the platform is actually used — find a fund, manager or allocator; evaluate and connect; start or continue a conversation; move on to meetings and the relationship. Structure follows the loop, not the org chart.

Decision 02

Two states, because focus and orientation are different needs.

An icon rail for people who already know where they're going, and an expanded state for people who don't. The horizontal bar could offer neither.

Research

Eight steps, because renaming things is easy to get wrong.

• Audit of the existing navigation.
• Hypotheses for architecture and naming.
• Tree testing.
• Internal validation with Customer Success.
• Unmoderated testing, then moderated sessions.
• A/B testing recommended for the rollout.

Each session opened with a recall exercise — which sections do you use, what confuses you, how do you normally find this — so participants judged the new structure against their real habits rather than against a blank slate.

What the tests measured

Not preference. Findability, speed and confidence.

• Whether labels matched the user's mental model.
• Whether the grouping felt logical, and the hierarchy readable.
• Time to complete core tasks, and confidence while doing it.
• Cognitive load, measured against the previous navigation.

Result

A 40% increase in product discovery and exploration.

Fewer duplicated entries, consistent naming, and shorter routes to the three things people came for. The structure also holds room for features that did not exist yet — which was the point.

Context

The person who writes the workout is never there when it is done.

A coach builds the session in advance. The athlete opens it hours or days later, alone, standing at a machine with the phone in one hand. Every question that comes up at that moment has to be answered by the screen, because there is nobody to ask.

The product ships in Portuguese — the screens below are in the language its users actually read. Key terms are translated where they matter.

Problem 01

Five methods, and the card treated every one of them as a label.

Simple, drop set, super set, rest pause, cluster set — five ways to structure a set, and the card treated all of them the same way: a badge carrying a name.

Here a badge said "BiSet" and two exercises sat stacked below it. Nothing said they belong to the same set, or that you do the second one immediately after the first. The label named the method to someone who already knew it, and told everyone else nothing.

The old rest pause card was the same shape: series, rest, load. Three numbers, and no trace of the thing that actually defines a rest pause — the short pauses inside the set.

Problem 02

Resting looked identical to being stuck.

When rest started, the timer took the middle of the card and everything behind it dimmed — exercise name, video, both buttons. There was no visual difference between "you are resting" and "the app has frozen", and people read it as the second one.

How I worked

The rules came from the technical team, not from me.

I went through each method with the platform's training specialists: what defines it, what the athlete must do between one series and the next, where the pause goes and how long it lasts. Design decisions only started once the rules were unambiguous, because a card that explains a method wrongly is worse than one that explains nothing.

Decision 01

The instruction lives in the gap between the series.

What distinguishes these methods is not the reps or the load — it is the transition. So the transition became the thing the card states out loud, as a marker sitting in the gap between one series and the next:

• Sem Descanso — no rest. Go straight into the next drop. This is what makes a drop set a drop set.
• Pausa 15s — pause 15 seconds. The short break inside a rest pause set, not the long one between series.
• Descanso — rest. The round is over; the full recovery starts here.

It teaches the method by describing the next move, so an athlete who has never done a drop set can still do one correctly, and one who does them every week is not slowed down by a tutorial they do not need.

Decision 02

Colour says "this one is not a normal exercise".

Five methods needed telling apart at a glance, but they do not need five colours. Simple is the baseline — it stays neutral, because it is what most of a workout is made of. Colour marks departure from it: blue for a drop set, green for a rest pause, and so on for super set and cluster set.

So the rule is not "each method has a colour". It is "colour means this one is not ordinary" — which keeps a full workout calm, and makes the four exceptions impossible to scroll past.

The badge also gained an information affordance. What opens is not a paragraph — it answers two questions in two labelled sections: how it works, and when to use it. Someone who has never done a rest pause needs the first; someone deciding whether to push through a plateau needs the second, and they are rarely the same person on the same day.

Decision 03

Rest shrinks instead of blocking.

Rest opens as a sheet, with controls to pause, add time or restart.

Collapsed into a bar, the whole card stays readable behind it.

The timer stopped being a state the app puts you in and became something running alongside you. You can keep scrolling, adjust a load, or look ahead at the next exercise while it counts down.

How it got made

The bottleneck was never the idea. It was drawing every state.

This problem multiplies badly. Five methods, each with its own number of series, each series with a different transition, plus the resting state, the completed state, the edge cases where a coach configures something unusual. Drawing all of it by hand is weeks of Figma work before anyone can react to a single screen.

I used Claude to generate the variations instead — describing the rules the technical team had given me and getting back the full set of scenarios to react to. Weeks of drawing collapsed into a few days that covered design, testing and validation together.

What changed was not the drawing speed. It was that testing could start while the thinking was still soft, so the states that survived were the ones that survived contact with people, not the ones I happened to draw first.

Results

Coaches stopped being asked how to do the workout.

Reports of athletes asking their coach how to execute an exercise or a method dropped by more than 70%. That was the exact goal: the question was never supposed to reach a person, because the person is not there when the set is being done.

It is the measure that matches the problem. Not time on screen, not taps — whether the screen answered the question that used to be asked out loud, hours later, to someone who could not see what you were looking at.

The number comes from the coaches themselves. These questions arrive by message or in person, where no analytics reach, so the people who were being interrupted are the only ones who can count them. Self-reported, and worth reading as such — but reported by the people who were paying the cost.