Bodygram app screens on phones: body scan, 3D body model with measurements, and size recommendation.

Bodygram · Body2Fit

Turning a multi-screen sizing form into an instant 3D result

Body2Fit recommends clothing sizes inside retailers' checkouts. People had to work through up to six screens before they saw anything. I redesigned it so the result came first.

Role
Lead Product Designer
Dates
Oct 2024 to Aug 2025
Team
Sole designer, reporting to the CEO and Head of Engineering
Product
Body2Fit, a size recommendation tool businesses embed for their own customers
Status
Designed and prototyped. Not shipped before a company restructure

This work didn't ship, so there is no measured result. It's here for the reasoning: a decision made from the product's own data, and what I would have measured.

The problem

Bodygram is an AI body-measurement company. Its products are used by businesses, who offer them to their own customers. Body2Fit sits inside a retailer's checkout and recommends a size.

Internal data showed that each extra screen in the sizing flow lost about 1 in 10 users. Depending on the route, the existing flow took up to six screens before anyone saw a result.

If each screen loses about 1 in 10 people, a third are gone after four screens. Arithmetic from the internal figure, not measured funnel data.

Constraints

  • Embedded in someone else's product. Body2Fit appears inside partners' checkouts, so it had to respect their branding and the space they gave it.
  • Accuracy. A size recommendation that's quick but wrong leads to returns, which defeats the point.
  • A small team. I was the only designer, across three products.

Decision: put the result on the first screen

Instead of polishing the form, I changed the order. The first screen shows a 3D model based on averages, with a size already recommended. People adjust it with height and weight sliders, and the size updates in real time.

Fewer screens means fewer places to drop out. And a result on screen straight away gives people a reason to carry on, because every change they make visibly improves it.

Before

Before: from Options, three routes. Stats only: enter stats, then processing, then results. Portrait: instructions, take photo, enter height, enter weight, processing, results. Full body: instructions, audio check, device placement, orientation check, front photo, side photo, processing, results.

After

After: enter height and weight shows results straight away. From the results, people can choose more detail: portrait (instructions, take photo) or full body (instructions, orientation check, front photo, side photo), then processing returns them to the results.
The user flows before and after, drawn to the same scale. Before: three routes, each a run of screens before any result. After: height and weight give a result straight away, and photos are an optional step to refine it.
Two phones. Left, the old form: empty height and weight fields and a fit preference slider before a Continue button. Right, the new screen: a 3D body with draggable height and weight markers, and a recommended size of M shown straight away.
The old form, and the new first screen: drag to set height and weight, and the recommended size updates as you go. Designs; not shipped.

Your size

M

Starts from averages. Adjust, and the size updates as you go.

An interactive sketch of the interaction, written for this page. The body shape and size logic are made up; this is not Bodygram's model.
Three phone screens for a pyjama set. First: height and weight set on a 3D body, recommended size M. Second: fit by area, just right at shoulders, sleeves and inseam, slightly tight at the chest, with a tight to loose scale. Third: estimated hip, inseam and waist measurements, and a button to upload a photo for more detailed sizing.
Three screens from the new flow: set height and weight, see how the recommended size fits each area of the body, and open the estimated measurements. A photo is optional, for more detailed sizing. Designs; not shipped.

Designing it as a system

Systems thinking

Most features in this version are built from a small set of reusable elements:

  • Shared tokens and elements throughout
  • A predictable type hierarchy
  • Sensible defaults and clear states

New features could be added without fragmenting the experience.

Brand customisation for partners

We introduced theming for each client:

  • Their own logo
  • Primary and accent colours
  • Token overrides for spacing, border radius and typography

Fashion retailers could embed Body2Fit in their own sites and keep their brand consistent, while still using our core product.

The same buttons in four client themes: a black pill, a teal rounded rectangle, a navy rectangle with small corners, and a square-cornered blue.
Each brand can customise the interface to reflect its own visual identity.
A token table for the button component. For each brand column it lists radius, height, and surface colours for the default, hover and pressed states.
Component tokens, set per brand: radius, height and colours for each button state. Client names hidden.

What I would have measured

  • Completion rate: how many people who open the tool reach a recommendation.
  • Time to recommendation.
  • Size exchanges: whether faster answers held up after delivery.

What I'd do differently

I'd push for a smaller first release that could ship and be measured, such as the result-first screen with only height and weight, rather than the full redesign. A measured slice is worth more than a complete design that waits.