Consumer AI / Commerce

AI Trial Room

Turn fashion browsing into confidence, not homework.

Product: AI shopping and virtual try-on
Role: Product design lead
Scope: Concept, interaction model, AI states, shopping journey

The tension

Fashion grids show products. Shoppers still have to imagine the outfit, the fit, and whether any of it feels like them.

What was happening

Online fashion forces users to imagine fit, proportion, mood, and full-outfit compatibility from a product grid.

What changed the brief

The real product was not virtual try-on. It was confidence: memory for what the user likes, context for why a look works, and honest recovery when the AI is unsure.

Success looked like

A stronger experience would make users feel advised, not automated: clear recommendations, visible style memory, and calm error recovery.

The constraint

AI output can be uncertain, body and face inputs are sensitive, and shopping decisions need confidence before checkout.

24%try-on intent lift

Projected lift from moving styling into the browse path.

18%full-look saves

Estimated gain from styling complete outfits, not loose items.

32kguided sessions

Concept target for shoppers asking AI before checkout.

The product story

The interface, one decision at a time.

01 / Entry

Start inside the shopping habit.

The stylist enters from the product feed, so asking for help feels like the next shopping action, not a detour.

AI Trial Room home screen with AI Stylist entry point — **AI appears where intent already exists.**A persistent entry sits inside discovery, close to products and the moment doubt begins.

AI Trial Room AI Stylist bottom sheet — **A conversation, not a control panel.**Four plain-language starting points replace an empty prompt and help users say what they need.

02 / The insight

Remember style. Do not make users repeat it.

A lightweight profile captures durable signals once, then lets returning shoppers refine only what changed.

New users teach the minimum.The profile asks for one useful input, then builds confidence progressively instead of front-loading a questionnaire.

03 / The payoff

Advice should end in a complete decision.

The recommendation keeps the look, price, and next action together so inspiration can become a cart without rework.

AI Trial Room add to cart full look state — **The full look becomes one cart decision.**Users can review every item, total cost, and checkout path without reconstructing the outfit themselves.

04 / Recovery / input

Bad input gets a better next move.

The product explains what the camera could not understand, then offers a retake, a guide, or a manual route.

AI Trial Room invalid image recovery screen — **The upload is rejected with a reason.**Specific guidance replaces a generic error, so the next attempt can improve.

AI Trial Room full body photo guidance screen — **The missing frame is made visible.**The screen tells users what the model needs instead of quietly producing a weak result.

AI Trial Room photo guide screen — **A good photo is taught before retry.**Simple examples reduce another failed upload and make the AI requirement feel fair.

AI Trial Room mirror selfie guidance screen — **The preferred capture path stays practical.**Mirror-selfie guidance meets users where full-body photos already happen.

05 / Recovery / confidence

When confidence drops, the UI says why.

Uncertainty is surfaced as a fixable condition, not hidden behind a confident-looking output.

AI Trial Room body type identification failure screen — **Body type can be corrected manually.**Users keep control when the model cannot make a reliable read.

AI Trial Room low lighting recovery screen — **Low light becomes an actionable diagnosis.**The product names the source of uncertainty before asking for another photo.

AI Trial Room multiple people recovery screen — **The model asks for one clear subject.**A precise constraint protects personalization from guessing who the profile belongs to.

AI Trial Room face not visible recovery screen — **Missing identity signal is explained.**The request is tied to recommendation quality, making the retry feel purposeful.

06 / Recovery / commerce

A failed AI moment should not end the shop.

Retry, alternatives, and profile completion keep momentum without pretending the original result worked.

AI Trial Room virtual try-on failed screen — **Try-on failure preserves a useful exit.**Users can retry or continue shopping instead of getting trapped in the feature.

AI Trial Room out of stock alternatives screen — **Stock failure becomes a substitution moment.**Closest alternatives keep the silhouette and intent alive without claiming an exact match.

An incomplete profile gets a checklist.The user sees the minimum missing signals and why finishing them improves the result.

The trust test

The happy path is only half the product.

Photo quality breaks the read

Blurry, dark, cropped, multi-person, or face-hidden inputs explain the issue and guide a better upload.

Try-on or stock fails

The UI offers retry, continue-shopping, or closest alternatives without pretending the match is perfect.

Profile signal is incomplete

The checklist shows what is missing and lets users complete the minimum needed for better styling.

The decisions

Three moves carried the story.

Treat confidence as the product job

The case frames virtual try-on as a trust problem, not a novelty feature. The UI explains recommendations and keeps recovery honest when AI is unsure.

Build a reusable style memory

Instead of remembering clicks, the system captures durable style signals that help future recommendations feel personal without forcing long setup.

Separate first-time and returning behavior

New users get a light onboarding path. Returning users can ask directly and tune occasion, budget, fit, or color mood one variable at a time.

What I would test next

The strongest part is the product model. The next validation step would be measuring whether style memory improves add-to-bag confidence and reduces rework.