Warranty claim photos: a checklist for support teams

1. Collect the materials for one claim

Keep one customer's request, product and purchase record together under the same case reference. Check the manufacturer's instructions before asking for more material. The useful set depends on the product and the reported problem.

For example, Tempur-Pedic's US claim form asks for purchase evidence, a product tag, an overall view and a close-up of the issue. STACYC's UK warranty page requests an invoice, serial-number and product photos, damage photos and a description. These are examples of actual intake requirements; they do not establish a universal checklist or imply that either company uses DeepfakePolicy.

• Tempur-Pedic: materials required for a warranty claim
• STACYC UK: warranty claim photographs and review process

• Purchase record: the receipt, invoice or order document, including the relevant product details and date.
• Whole product: a clear view that gives the close-ups context.
• Product label: a readable model and serial number where the item has them.
• Reported issue: a close-up and enough surrounding detail to locate it on the product.
• Customer's description: what happened, which part is affected and when the issue was noticed. Treat this as a statement to review.

2. Define the questions you can answer from those files

Ask a separate question about each relevant field. Start with the product model, serial number, purchase record and visible issue. A broad question such as 'Is this claim genuine?' combines several decisions that the files may not support.

A receipt may identify the model without recording a serial number. In that case, it can support a model comparison but cannot establish that the photograph shows the individual unit sold. Record the missing link and check your order or service records if they contain it.

• Does the model on the product label match the product line on the receipt?
• Where a serial number appears in more than one source, do the readable characters agree?
• Does the photograph show the part and issue described by the customer?
• Which requested detail is absent, obscured or too unclear to compare?

3. Check extracted values against the source

Open the source before relying on a discrepancy. A glare spot, a cropped label or a small character can change how text is read. Keep the original value, filename and document page beside the extracted value so another reviewer can check the same detail.

Automated text extraction can return uncertain values. AWS's document-review guidance calls for good-quality images and closer human scrutiny where needed. Apply that principle to the review: check the source before treating an extracted value as established.

If an identifier remains unreadable, request a clearer photograph. Keep the submitted original and any additional version distinguishable in your own case record. Do not silently fill in a character from the expected number.

• AWS: document image quality and reviewing uncertain extraction results

4. Separate differences from missing evidence

Use findings that tell the next reviewer what to do. Matching model names, a conflicting serial number and an unseen damaged part are three different outcomes. Combining them into one pass-or-fail label hides the unresolved work.

• Readable match: the same field agrees across the cited sources. Keep the scope narrow; a model match does not identify the individual unit.
• Confirmed difference in the supplied files: the source values are readable and differ. Check for a wrong attachment, transcription error, replacement unit or other explanation.
• Missing detail: a source does not contain the requested value. Ask for the record that could supply it.
• Unclear detail: a value or affected part cannot be assessed reliably. Ask for a better view or an appropriate inspection.

5. Make the follow-up specific

A useful request names the unresolved detail and explains how to photograph or document it. For a serial number, ask for the complete label without glare, with enough of the product visible to locate the label. For a crack, ask for a close-up and a wider view showing where it is. Avoid asking the customer to resend the entire submission when only one detail needs clarification.

Some problems need an inspection, diagnostic test or service record. A photograph cannot establish an intermittent electrical fault or when damage occurred. STACYC's published process explicitly allows for additional information, more pictures or a physical inspection before a coverage decision.

Record what the new material resolves and what remains open. A difference can justify a question; it does not by itself establish dishonesty or determine the customer's entitlement.

• STACYC's published review process

6. Keep image analysis separate from the claim decision

Photo-document comparison checks whether submitted details agree. AI-image detection checks a different question: whether the image contains signals associated with generation or manipulation. Use a separate detector check when that question matters, and record its limitations alongside the result.

A low detector signal does not verify a receipt, connect the photo to a particular sale or confirm the cause of damage. A high signal needs review and can be wrong. Your support team still needs to assess the records, applicable terms and any inspection findings before deciding what happens to the claim.

• How photo and video detection fits into a business workflow
• Read the detection methodology and limitations

7. Save a record another reviewer can use

Keep the case reference, question, source files, compared values, findings and follow-up together. Add the reviewer's decision and its reason separately from the automated findings. That lets a colleague see which evidence informed the decision and which questions were still open.

In DeepfakePolicy, start with the Returns and warranty review template or your company's checklist. Add the photos, supporting documents and description, then review the cited findings in the company workspace. The technical file check does not confirm that every label or damaged area is visible; inspect the previews before starting paid processing.

Download the comparison report as PDF, text or JSON when you need to keep it. Company comparison reports are available for seven days, and source files are removed after processing. Keep your own authorised source copies according to your organisation's retention rules. Report exports do not run the comparison again.

The price is shown before submission. A completed comparison is chargeable even when some findings are inconclusive; an AI-image check is a separate service. Start with a case your team already understands and compare the output with its review before making this a routine step.

• Explore the photo and document cross-check
• Try the illustrative camera-return example
• Read business data processing and retention details