FindIt Mobile & AI Feature

Overview

FindIt helps people list and discover local goods. The mobile product needed a faster path from camera to listing — especially for first-time sellers who abandoned uploads mid-flow.

This study covers the AI-assisted upload feature: how we reduced friction in capture, review, and publish without making the system feel opaque or untrustworthy.

Context

FindIt AG was scaling a marketplace where listing quality and speed both mattered. Sellers on mobile were the growth edge; desktop flows already worked better than phone.

I joined as product designer on the mobile surface, partnering with engineering on an AI-assisted path for image capture and listing drafts.

Problem

Upload completion lagged on mobile. Sellers dropped when:

• Framing and photo count felt unclear
• Manual tagging and fields felt like paperwork
• Errors surfaced too late in the flow

Research

• Reviewed session recordings and support tickets around failed uploads
• Shadowed a small set of new sellers on first listing
• Benchmarked assistive capture patterns in adjacent consumer apps (without copying chrome)

Key insight: trust hinged on editable drafts more than on “magic” autofill. People wanted speed and the last word.

Constraints

• Shipping window measured in weeks, not quarters
• On-device and server assist both had cost/latency tradeoffs
• Marketplace trust & safety rules still applied to AI-suggested copy and images
• Localization for primary markets could not slip

Process

1. Mapped the existing upload funnel and drop-off points
2. Prototyped a capture → review → edit → publish spine with AI suggestions inline
3. Defined empty, loading, low-confidence, and failure states before polish
4. Paired with eng on confidence thresholds for when to suggest vs. ask

Iterations

Early builds over-automated: drafts felt “done” and sellers skipped review. We pulled back — suggestions became clearly marked, with one-tap accept/edit and a forced glance at critical fields.

Photo guidance moved from multi-step coaching to lightweight overlays that disappear once the shot is good enough.

Outcome

• 100K+ downloads in the launch window
• ~20% increase in successful uploads vs. the prior mobile path
• Clearer ownership: AI drafts, human publish

Metrics above are directional outcomes from the product’s reported launch window; replace with final signed-off figures when available.

Reflection

Assisted flows only work when the interface makes agency obvious. The win was not “more AI” — it was fewer abandoned listings while keeping sellers in control of what goes live.

Next time: instrument suggestion accept/edit rates earlier, and design the low-confidence path with the same care as the happy path.