Menu

Overview

Self Inspection lets a seller inspect the car at home. The bar was not "good enough photos." It was quality close enough to a physical inspection that OLX could price the car with confidence.

Product
AI-assisted flow: a short tap-through for facts, then a camera that tells you how to shoot condition.
Bet
If sellers can do the inspection, OLX stops sending an engineer to every driveway and still gets trustworthy data.
Role
Two designers. I shaped the concept, the visuals, the camera, UATs, and later A/B tests.
Step one of the wizard: pick a car brand from a grid of logos.
Facts: tap through brand, model, variant.
Car photos checklist with required angles before opening the camera.
Condition: a checklist, then the camera.

Problem

Selling online meant a long form, then a wait. An OLX inspector still had to see the car at home or at a store. Slow for the seller. Expensive for the business. Hard to grow.

  • The form was tedious. People abandoned it.
  • Every listing still needed a person on site.
  • Ops cost sat on physical inspections, so volume was capped.
Illustration of booking a store inspection, then an OLX expert checking the car at an OLX Autos store.
Book, then bring the car in.
Illustration of booking a home inspection, then an OLX expert checking the car at the seller's house.
Book, then wait at the door.

Approach

Research and benchmarking pointed at two jobs: a tap-tap path for car facts, and a guided image capture for condition. Storyboards forced the hard split before concepts: is the seller near the car, or away from it?

Two contexts, two paths

Path A

Near the car

Shoot first. Capture condition while you are already next to it, then finish the facts.

Path B

Away from the car

Documents and questions first. Photos wait until you can get to the vehicle.

Storyboard comparing a near-the-car path that shoots first with an away-from-the-car path that starts with documents.
Near: shoot first. Away: documents first, photos when you get to the car.
Diagram showing two paths to a quote: photos first if you are near the car, questions first if you are away.
Same destination. Different order depending on where you stand.

Four concepts, one ship decision

Low-fi ideation put four directions on the table. A Dreamer / Optimistic / Critic workshop weighed timeline, tech, and fit, then picked what could clear the data-quality bar.

Concept 1

Gamified chat

Playful conversation to keep people moving through questions.

Concept 2

OLX as buyer or engineer

A conversational inspection with OLX playing the counterparty.

Concept 3

Camera as the assistant

Guidance lives in the lens: stencils, cues, and feedback while you shoot.

Concept 4

Wizard for questions

A short tap-through for facts, separate from condition capture.

We shipped Concept 4 for questions and Concept 3 for condition. Cute chats did not clear the bar for trustworthy inspection data.

Mascot-guided camera exploration with speech bubbles and stencil overlays on document and car shots.
Chatty camera exploration. Friendly tone; we shipped guidance without the mascot chat.

What to ask

  • Ops and stakeholders listed the data points that build buyer trust.
  • Insurance and ownership sat next to condition, not afterthoughts.
  • Sellers still needed a path that felt short.

Start screen and exit nudges

Asking "are you near the car?" failed when people were inside and the car was in the driveway. Iterations landed on two clear cards: fill details, or take photos.

Three iterations of the start screen, ending with two cards for filling details or taking photos.
Start screen iterations: clarity over a yes/no proximity question.

Nudge A

Generic exit

A confirm on most screens so leaving is intentional.

Nudge B

Offer with a range

Once enough data exists to price, show a range instead of a dead end.

Nudge C

Camera bail-out

A camera-specific nudge if someone exits mid-shoot.

Three exit nudges: generic confirm, offer with a price range, and a camera-specific bail-out.
Exit nudges: confirm, offer a range, or pull people back into the shoot.

How to fill the car

Three ways in. Prefer speed when documents or plates work; fall back to manual when they do not.

Option 1

Registration certificate

Scan the RC and pull the details.

Option 2

Car number

Fetch details from the plate when a scan is not handy.

Option 3

Fill manually

Tap through brand, model, and variant when neither fetch works.

Screen offering fill manually or through RC with a privacy note.
Pick RC scan or fill manually. Privacy note stays visible.
RC photo screen analysing a registration certificate to autofill car details.
RC scan in progress: autofill the facts.
Five screens of the wizard: brand, model, variant, photo briefing, then camera confirm.
Tap-through for facts. Camera only when you need the car.

Camera: landscape on purpose

Layout was not a preference exercise. Portrait trials lost the car. Landscape won for proportion, preview size, and less junk in the frame.

Tried

Portrait camera

Familiar phone orientation. Too little of the car, too much of everything else.

Shipped

Landscape camera

Fits car proportions, bigger preview, controls that flip for left-handed shooting.

Camera UI explorations: portrait and landscape layouts with stencil overlays and shot lists.
Camera layout explorations before the landscape ship.
Landscape camera with a car stencil overlay, a shot list on the left, and a Click Photo button.
Shot list on the side; stencil on the preview. Next angle after each click.
Four landscape camera screens for front, left, back, and right, with stencil overlays and a processing state.
Front, sides, back, then process.

What testing taught us

  • Stencils and error states helped. Continuous shooting felt right.
  • Cars parked against walls were hard to frame.
  • People did not know how to open the bonnet for engine shots.
  • A light toolbar disappeared outdoors.
  • Some cars sit far from home, so the away path mattered.

Camera experiments after launch

After ship we kept testing the camera as answers to the UAT failures, not as polish.

Experiment 1

Angled shots

Help people shoot in tight parking without cropping the car.

Experiment 2

Explain in flow

Pop-ups that teach the next move instead of dumping people into the lens.

Experiment 3

Clearer errors

Cropped, blurry, or off-frame: say so, then retake.

Camera feedback for a good side shot, a cropped photo, a blurry photo, and a seats capture with a stencil.
Cropped, blurry, or off-frame: say so, then retake.
Stencil and icon set for car angles, interiors, engine, and damage grids used in the camera flow.
Stencil and icon set that made the angles teachable.

Outcome

After first launch: more leads, more cars bought, and engineers spent time on cars that needed a second look instead of every driveway.

  • 5xMore leads
  • 2xMore cars bought
  • 3Countries shipped
  • Leads grew faster than ops could buy. Humans stayed in the loop while capacity caught up.
  • That was the point of the camera: fewer wasted visits, not zero humans.