Claimscan
API OPERATIONALLog inTry it free
// ABOUT US

The team behind Claimscan — forensic AI for honest return decisions

Claimscan is a tool for e-commerce customer-support teams who see suspicious damage photos in return tickets every day — and need to know quickly whether they really are suspicious. We analyse the image forensically in under three seconds and return a probability. The decision is always made by a human.

// MISSION

We are the assistant, not the judge.

In e-commerce there is hardly anything worse than wrongly accusing an honest customer of fraud. A single false positive destroys trust, leads to public complaints, and costs more in the long run than any individual abusive return.

That is why Claimscan never says “This is fraud” or “Reject this return”. Our API returns a sober verdict — LIKELY_AUTHENTIC, SUSPICIOUS, LIKELY_MANIPULATED or LIKELY_AI_GENERATED — combined with a confidence score and a list of forensic findings. What your team does with it is your call.

“Forensic findings, not verdicts.” This single rule defines every report, every API response and every line of UI copy we ship.
// METHODOLOGY

From the fastest signal to the most thorough

Most manipulations are detectable by local forensics — EXIF inconsistencies, missing MakerNotes, software traces from photo editors. Only when that is not enough do external APIs come into play. This cascade returns a result in under three seconds for most images and reserves the deep analyses for the few cases that need them. These processing stages are grouped into the five forensic layers and twelve detectors documented in the methodology.

  1. 01

    Deep Metadata

    local · 200 ms

    EXIF, IPTC, XMP, Apple MakerNote — author software, AI-editor patterns, timestamp consistency.

  2. 02

    MakerNote validation

    local · 50 ms

    Real cameras leave a rich, vendor-specific MakerNote structure. If it is missing, that is a strong signal.

  3. 03

    Pixel forensics

    local · 500 ms

    Error-Level Analysis, noise consistency, edge continuity — classical manipulation indicators.

  4. 04

    Reverse image search

    API

    Reverse image search across the open web. If the image already exists elsewhere, it is not a new damage photo.

  5. 05

    AI detection

    API

    Multiple independent, specialised detectors for AI-generated images — far more precise than a generic LLM.

  6. 06

    Vision model

    API

    A vision model checks physical plausibility, shadow consistency and context — only when stages 1–5 are inconclusive.

  7. 07

    Duplicate check

    local · 100 ms

    Perceptual-hash lookup against our own corpus. Catches recycled material from other cases.

  8. 08

    Cross-image analysis

    local · 500 ms

    When multiple images are in a case: timeline, GPS consistency, device chain. Surfaces hit-and-run patterns and address discrepancies.

Every finding has a severity, a confidence score and an aggregated contribution to the overall verdict. The aggregation logic is documented in the PDF report — no black-box score.

Full methodology with all 12 detectors

// TEAM

Who builds this

Piyal Ranasinghe

· Founder · Engineering · Product
Nuremberg, Germany

For more than ten years I have been building software for e-commerce operators. WELANDA, my own D2C shop, is the source behind Claimscan: over the years we saw how return fraud became increasingly easy with free photo editors and, since 2024, with generative AI tools — and at the same time how quickly a mishandled case alienates a loyal customer.

Claimscan is the answer: a sober, transparent tool that does the forensic work without taking the decision away from the support team. Code, pipeline architecture and scoring logic come directly from me.

// TRUST

What sets us apart from US SaaS

Hosting in Germany

Hetzner Cloud, Nuremberg. No data leaves the EU. Our own Postgres, Redis and MinIO instances — no US hyperscaler.

GDPR from day one

DPA as PDF in-app, automatic image deletion after 30 / 90 / 180 / 365 days depending on plan, GPS data is never persistently stored.

Forensic language

No words like “fraud” or “fake” in API responses, PDFs or UI. Only structured findings with confidence — you keep interpretive authority.

Transparent cost cascade

The layered forensic pipeline is documented in the report — you see exactly which of the 12 detectors fired and at what weight.

Sole proprietor, no VC

Claimscan is bootstrapped. No pivot pressure, no exit pressure. We build the tool we use ourselves in our own shop.

Open API, your data

Standard-compliant REST API, OpenAPI spec, signed webhooks. You can migrate any time — no vendor lock-in.

See it for yourself.

Free plan: 10 images per month, no credit card required. First result in under 3 seconds.

Start for freeRequest a demo