Research / Product strategy / UX/UI Design / AI-assisted process

FitMatch, a personal size assistant that learns from your purchase history to recommend the right size for any brand you shop online

Project
FitMatch

Role
Product Designer

Project type
Personal project

Year
2026

My role

I led this project end-to-end as a solo designer, from problem framing and research through to information architecture, wireframing, and UI design. I used AI tools across every phase to enhance the process.

1. Overview

Context

Online fashion has a sizing problem no one has fully solved. Customers return items, abandon carts, and stick to brands they already know, because they can't trust that something new will fit.

FitMatch is a mobile app that learns from your purchase history and fit feedback to recommend the right size for any brand, before you buy. Your size profile belongs to you, not to the store.

AI-assisted design process

AI was integrated throughout the project as a working tool, not a shortcut. Claude supported strategic thinking, research structuring and design decisions. UX Pilot and Figma Make were explored during wireframing to accelerate iteration and outputs were always evaluated and refined with Figma based on UX criteria.

2. The problem

What the data tells us

  • · 88% of online shoppers have returned an item due to incorrect sizing.

  • · 73% have abandoned a purchase because they weren't confident about the size.

  • · 73% limit themselves to brands they already know to avoid sizing mistakes.

  • · 68% have no system to keep track of their sizes across brands.

Sources: PowerReviews 2024, Sendcloud 2024, Statista, iF Lastmile 2023.

The core problem

Online shoppers have no reliable way to know what size to order from a brand they don't know. The tools available today are built into the retailer's platform: generic, disconnected, and doesn't account for their personal history or the real behaviour of each garment. The result: avoidable returns, abandoned carts, and an involuntary dependence on familiar brands.

Design question

How might we help online shoppers always order the right size, regardless of the brand?

3. Research

Synthetic approach

With no resources for primary research, I combined secondary sources with a synthetic research method: a survey designed from real market data and answered by AI-generated user profiles based on validated behavioural patterns. This allowed me to simulate research findings grounded in real evidence rather than assumptions.

Every existing solution is built for retailers, not for shoppers

I started by mapping the competitive landscape to understand what solutions already exist and where the real gaps are.

The key finding

True Fit, Zalando Size Advisor, MySizeID: all of them live inside the store. None of them give the user a portable, personal size profile they can take anywhere. That gap became the foundation of FitMatch.

Two shoppers, same problem, different breaking point

Based on research synthesis, I defined two personas representing opposite ends of the same problem.

Laura, 29 — The reactive shopper. Buys frequently, returns often. She already bought wrong, and now she's dealing with the consequences.

Marcos, 38 — The cautious shopper. Abandons the cart if he's not confident about the size. He never gets to buy wrong, he just doesn't buy.

FitMatch needed to work for both: intervening before the purchase for Marcos, and building a history that prevents the next mistake for Laura.

The moment FitMatch had to own

Mapping both journeys side by side revealed the critical moment FitMatch needed to address: the size selection step, where Laura guesses and Marcos gives up.

What the research actually revealed

After synthesizing research data and user stories, we identified four key insights.

1. The problem is not returns, it's uncertainty.

Returns are a symptom. The root cause is that shoppers have no reliable information at the moment of decision.

2. Existing tools don't travel with the user.

Every solution resets when you change retailer.

3. Qualitative fit data is missing everywhere.

Knowing someone ordered an M is not enough, but knowing it ran small and shrank after washing is what actually helps.

4. The cold start problem is real but solvable

A new user with no history can still get value from community data from day one.

4. Product strategy

The market gap that makes FitMatch possible

No existing solution gives shoppers a portable, personal size profile they control. That's the white space FitMatch occupies, not inside a retailer but wherever they shopper buys.

A recommendation built on three layers of data(Data model)

FitMatch's recommendation engine combines three data sources, each solving a different part of the problem.

1. Personal history

Sizes and fit feedback from the user's own purchases, imported from email or added manually. The most accurate signal because it's specific to that person's body and preferences.

2. Open communities

Public data from Reddit, forums, blogs and reviews, processed to extract sizing insights for specific brands and garments. Solves the cold start problem from day one.

2. Combines personal and community data

Two signals are always more reliable than one.

3. FitMatch community

Aggregated fit data from other FitMatch users with similar profiles. Gets stronger as the user base grows. Planned for V2.

Flow 3

Manual size entry

What sets FitMatch apart from the rest of the market?

  • 1. The size profile is portable, across every brand and retailer.

  • 2. Combines personal and community data, more reliable than one source.

  • 3. Qualitative fit feedback is a first-class input.

The value proposition

FitMatch is a personal size assistant that learns from the purchase history and the collective knowledge of online communities to recommend the right size in any brand, before the user buys. Yours, not the store's.

Deciding what MVP actually needed to be(MVP scope)

Not everything could ship in the first version. I used a MoSCoW prioritisation to separate what was essential from what could wait, and to avoid building a product too complex to validate.

———-Insert MVP scope table here. QUIZÁ NO CAL

Flow 2

Size recommendation triggered from any ecommerce

The three flows that define the MVP:

Flow 1

Onboarding with email import

3. Qualitative fit feedback is a first-class input

Knowing a garment runs small or shrinks after washing is as valuable as knowing the size ordered.

What makes it defensible (Differentiators)

Three decisions that set FitMatch apart from anything currently on the market:

1. The size profile is portable

It works across every brand and retailer, not just one.

5. Information architecture & user flows

Three entry points, one coherent system(Information architecture)

FitMatch is built around three tabs that reflect the three things a shopper needs. Simple navigation, no dead ends:

  1. Get a recommendation

  2. Review their history

  3. Manage their profile.

The architecture was designed around a key constraint: the most important action, getting a size recommendation, had to be reachable in as few taps as possible, regardless of where the user enters the app. That meant designing for three distinct entry points: opening the app directly, tapping a push notification, and sharing a product page from any ecommerce via the native share sheet.

Three flows that cover the full user cycle

Rather than mapping every possible path, I focused on the three flows that define the core experience of FitMatch.

Flow 01: Onboarding. The user connects their email, the app imports their purchase history automatically, and they optionally add qualitative feedback for each item. Every decision in this flow was made to reduce friction while maximising the data quality that powers future recommendations.

Flow 02: Size recommendation from ecommerce. The user shares a product page directly from their browser. FitMatch identifies the brand and returns a recommendation combining personal history and community data. Two sources, always visible, so the user understands why the recommendation is what it is.

Flow 03 — Manual size entry. The user adds a size without connecting their email. Designed as a fully valid alternative to the onboarding import: not a fallback, but an equal path.

The design decision that changed the onboarding

The original onboarding ended at import confirmation. After reviewing the flow, I added a qualitative review step: after the user confirms their imported purchases, they're invited to rate the fit of each item.

This single addition transforms the onboarding from a data import into the first meaningful contribution the user makes to their own recommendation engine. The more they share upfront, the better FitMatch works from day one.

6. Wireframes

Structure before style(Approach)

The wireframing process began with Flow 01, the most complex and critical flow of the product. A proper onboarding structure ensured that each next flow had a solid foundation to build upon.

AI as a wireframing accelerator(Process)

This phase was the first where AI tools moved from strategic support to direct design output. I used UX Pilot and Figma Make to generate initial screen structures from detailed prompts, then evaluated and refined each output based on UX criteria.

The process was iterative by design: write a precise prompt, assess what the tool produced, identify what worked and what didn't, and edit directly in Figma. Neither tool replaced design judgment, both accelerated the point at which there was something concrete to react to.

Insert wireframe screens here - 02 y 03

From three flows to twenty-one screens

The wireframing phase produced 21 screens across the three core flows.

Puede unirse a la parte de arriba de los wireframes

Flow 01: Onboarding: 8 screens covering the full journey from splash to home, including the qualitative review step.

Flow 02: Size recommendation: 6 screens from the ecommerce share sheet to the redirect confirmation, with two recommendation states: with and without personal history.

Flow 03: Manual size entry: 5 screens from the history tab to the confirmation state, including the qualitative comment invitation.

Flow 01: Onboarding

8 screens covering the full journey from splash to home, including the qualitative review step.

Flow 02: Size recommendation

6 screens from the ecommerce share sheet to the redirect confirmation, with two recommendation states: with and without personal history.

Flow 03: Manual size entry

5 screens from the history tab to the confirmation state, including the qualitative comment invitation.

7. Design

An identity built around confidence(Visual direction)

FitMatch's visual language has one job: make the user feel certain. Every design decision, color, type, spacing, was made to reduce hesitation and communicate precision.

The primary color is a saturated violet (#6B21D6). In a market dominated by neutral, retailer-safe palettes, violet signals that FitMatch belongs to the user, not the store. It's the color of the recommendation, the moment of confidence.

Two typefaces considered, one chosen(Typography)

Space Grotesk was chosen over Inter for its geometric construction with intentional irregularities. Where Inter disappears into the interface, Space Grotesk holds its own, especially at display sizes where the size recommendation lives. Legible at every scale, distinctive at the ones that matter.

A system, not a collection of screens(UI system)

The component system was built around the three moments that define FitMatch: receiving a recommendation, reviewing your history, and contributing fit data.

Insert UI system / component overview here

The recommendation card is the most important component in the product. It carries the size, the confidence level and the two data sources, personal history and community, in a single glance. Everything else in the system exists to support that moment.

Color with meaning(Color system)

Every color in the interface earns its place.

Violet for primary actions and the recommendation itself. Green (#2E9E6B) for confirmed fit and high confidence (success states only). Amber (#EF9F27) for medium confidence and fit warnings. Surface gray (#F5F4F2) to reduce visual noise in data-heavy screens.

8. Prototyping

Three flows that cover the full user cycle (Scope)

The prototype covers the 3 core flows defined during the product strategy phase. Together they complete a full cycle: a new user sets up their profile, gets a recommendation while browsing, and adds a size manually when needed.

Flow 1: Onboarding (8 screens)

The user connects their email, the app imports their purchase history automatically, and they review and confirm the detected purchases.

The qualitative review turns onboarding into the first step toward building a personalized engine.

Flow 02: Size recommendation from ecommerce (6 screens)

The user shares a product from an ecommerce. FitMatch identifies the brand and returns a recommendation combining personal history and community data, both sources always visible.

Two recommendation states are designed: with personal history and without. The confidence level is never hidden from the user.

9. Key decisions

1. The step that makes the recommendation feel yours

Adding the qualitative review step changed what the onboarding produces. The user tells FitMatch how each item actually felt, and that fit feedback is the most valuable input in the system. When FitMatch recommends a size, it doesn't feel like a guess. It feels like a conclusion based on their own words.

2.Two sources are always better than one hidden algorithm

Early versions showed a single result with a confidence level. The reasoning was invisible: the user had to trust the output without understanding the input. Showing personal history and community data as two separate, named sources turned the recommendation from a black box into a transparent argument. When both sources agree, the user buys with confidence. When they differ, they have the context to decide for themselves.

10. Learnings

What I would do differently

Working with synthetic research means one scenario was never truly stress-tested: a first-time user with no purchase history at all. The medium confidence state in Flow 02 handles it functionally, but real user behavior in that moment (the hesitation, the distrust) is something only live testing would reveal. It's the part of the experience I'd validate first.

On AI tooling: the quality of the output depends entirely on the quality of the input. A vague prompt produces a generic screen. A precise prompt, grounded in real product decisions, produces something worth refining.

My design tools

Notion

Figma

Mobbin

NotebookLM

Forms

ChatGPT

Condens