The warranty wakes up

Levi’s warranty form does not read like a brand promise. It reads like a small crime-scene protocol.

Take a photo of the defect. Take more than one. Photograph the care label. Photograph the inside label. Make the problem visible enough that a person who has never touched the jeans can decide whether the fabric failed, the customer misused it, or the claim is nonsense. CENTR’s equipment claim flow is even more explicit: receipt, defect photos, issue description, contact details, with bank statements, packing slips, or serial numbers sometimes accepted when the clean proof has gone missing.

That is boring human admin. It is also perfect AI work.

Search the inbox. Find the order. Read the warranty window. Pull the receipt. Check whether the serial number appears in an old installation photo. Ask the customer for one missing image. Package the claim. Submit it. Chase it five days later when nobody replies.

None of this creates a new right. The customer already had the warranty. The promise was already printed on the box, used in the product page copy, and priced into the margin. The only thing missing was memory, patience, and paperwork.

Forgotten paperwork is vendor margin.

The dormant liability layer

Warranties, rebates, service contracts, recalls, price-match guarantees, maintenance promises and return windows all live in the same strange category. They are commercial promises that are technically real but operationally asleep.

A blender has a two-year warranty. A retailer offers a 30-day price-match guarantee. A mattress company advertises a long trial. A manufacturer issues a recall notice. A software vendor allows cancellation before renewal. These promises sound generous because most customers do not act on them.

They forget the date. They lose the receipt. They cannot find the serial number. They stop after the second support email. They do not know whether a failed hinge counts as normal wear. They decide £38 is not worth a Tuesday afternoon.

That friction is not accidental. Sometimes it is just legacy mess. Sometimes it is economic design wearing a customer service uniform.

Rebates have always relied on this. The manufacturer can advertise the lower effective price while keeping the money from every customer who misses the form, mistypes the address, posts late, or gives up. Extended warranties have the opposite shape: customers overpay for cover they rarely use, partly because claim filing is annoying at the exact moment the product has already irritated them.

AI changes the claim rate by removing the small acts that used to kill the claim.

Warrantify already points at the consumer-side version: scan email for receipts, attach warranty information, and move towards AI-assisted claim filing through a purchase memory layer (Warrantify). Cover2 points at the merchant-side version: an AI claim conversation that asks for purchase date, retailer, serial number and evidence, then routes unresolved cases to a human (Cover2).

The interesting bit is not either product. It is the collision.

One clerk wakes every possible claim. Another clerk defends margin without looking like a villain.

The purchase-memory agent

The useful agent here is not a receipt scanner with a nicer UI. It is a purchase-memory agent.

A receipt scanner remembers that you bought something. A purchase-memory agent knows what the object is, what promises attach to it, what proof exists, what proof is missing, and when action becomes profitable.

For a consumer, that might mean:

• The washing machine bought from AO on 14 May 2024.
• Paid with a Monzo card ending in 4921.
• Two-year manufacturer warranty.
• Five-year motor warranty if registered within 90 days.
• Serial number visible in a delivery photo.
• Receipt in Gmail.
• No installation photo.
• Coverage expires in 312 days.

For a facilities manager, the ledger gets more interesting. Three hundred office chairs. Forty laptops. Clinic fridges with temperature logs. A school cupboard full of tablets. Rental properties with ovens, boilers, extractor fans and fitted appliances. A small retailer with shelving, scanners, card terminals and CCTV units.

Weak memory is a hidden vendor subsidy in all of those environments.

The agent’s dashboard is not dramatic. It is Monday morning admin:

• 17 assets with expiring coverage.
• 3 possible claims.
• 2 rebates still recoverable.
• 1 recall match.
• 6 items missing serial photos.
• 4 policies requiring proof of installation.

The admin clicks “prepare packets”. The AI drafts the claim, attaches the invoice, serial photo, defect image, usage note and service log. A phone notification asks: “Take a fresh photo of the cracked hinge. Include the label.” The human spends 40 seconds on the part only a human can do: point the camera at the broken thing.

Then the agent submits through a portal, email form or support chat. It follows up every five days. It hands off when the vendor alleges misuse, asks for physical inspection, conflicts with consumer law, or the value crosses a threshold where a person should read the file. £500 is a decent gut-check number for many small businesses, though each category will set its own line.

The product category hiding here is not “AI support”. It is post-purchase memory as infrastructure.

Ordinary commerce becomes insurance-like

Customer support is the wrong analogy. Insurance is closer.

In insurance, the claim is not the event. The claim is the evidence packet around the event: date, cause, policy wording, photos, receipts, exclusions, adjuster notes, prior condition, chain of custody. The speed of payment depends on the shape of the file.

Ordinary commerce is about to inherit that logic.

A toaster warranty becomes a miniature insurance claim. A gym equipment return becomes a claim file. A school laptop repair starts to feel like subrogation-lite when the device was damaged by a pupil, bought by a council, covered by a vendor, and repaired by a third party. A contractor replacing a stolen tool now needs proof of purchase, site report, serial number and police reference.

AI does not need to understand commerce in some grand semantic sense. It needs to assemble a file that survives rejection.

That will change support queues. Today a lot of inbound volume is messy:

• “I bought this last year somewhere.”
• “The receipt is in my email but I cannot find it.”
• “Your website says two years, the agent says one.”
• “The label rubbed off.”
• “I already sent a photo.”

With customer agents, more claims arrive technically complete. That sounds good until the facts are bad.

Support teams will see beautiful packets for low-merit claims. AI-edited defect photos. Serial numbers scraped from marketplace listings. Receipts reused across households. Customer agents quoting the wrong policy with great confidence. Manufacturer says retailer problem; retailer says manufacturer problem; two bots loop politely forever while a human gets angrier.

The first operational shock will not be fraud. It will be volume plus plausibility.

A sloppy claim is easy to reject. A well-formed weak claim takes time.

Pricing gets less lazy

A promise that nobody redeems is marketing. A promise that agents redeem is liability.

That difference hits pricing first.

Extended warranty pricing has always depended on failure rates, claim rates, admin costs and customer behaviour. AI memory changes the behaviour side. If more covered failures are claimed, reserves need to move. If every £79 countertop appliance claim arrives with receipt, label, defect photo and policy quote, the old model starts to leak.

Rebates are even more exposed. A rebate that worked when 20% of buyers redeemed it breaks if agents push redemption towards 70%. The headline discount becomes the real discount. The finance team will notice before the copywriter does.

Price-match guarantees have the same weakness. They are cheap when customers have to check prices manually, remember the deadline, collect proof and ask. They become expensive when a banking app expands a card transaction into an entitlement card and says: “You may be owed £38.”

The better question for a retailer is not “can we block these agents?” It is “which promises only work because customers are disorganised?”

Those promises are going to be re-priced, narrowed, or redesigned.

Some firms will respond predictably:

• Original receipt only.
• No bank statements.
• Video required.
• Product must be shipped back.
• Photo must include today’s code.
• Manual inspection for all claims.
• Shorter windows.
• More exclusions.

That may work for a while. It also turns each proof requirement into an agent checklist. Today’s anti-fraud ritual is tomorrow’s automation target.

There is a grim product future where warranty forms become CAPTCHA with packaging tape: rotate the toaster, show today’s code, read the serial number aloud, display the broken hinge under natural light, prove you still possess the object. It will reduce fraud. It will also make fair customers feel accused.

The better brands will make fair claims boring and fraudulent claims expensive.

The post-purchase surface

Most product teams obsess over pre-purchase conversion. The post-purchase surface is treated as an email receipt, a help centre article and a return form nobody wants to maintain.

That is going to look dated.

A claim-ready product has different behaviours:

• At checkout: “Register product and store claim evidence.”
• On first use: phone camera prompt to capture serial, installation state and condition baseline.
• In the banking app: return window countdown attached to the transaction.
• On the box: QR code pre-linked to policy, parts list and proof of purchase.
• In the customer account: warranty status, transfer rules and claim endpoint.
• In facilities software: asset ledger, expiry warnings and claim readiness scored green, yellow or red.

This is annoying in the same way two-factor authentication is annoying. Nobody wants it until the cost of not having it shows up.

For consumers, the willingness to pay is uneven. Nobody should pay £8 a month to chase warranty claims on cheap kettles. But one recovered appliance claim can justify a year of subscription. For small businesses, a recovery-share model makes more sense: take 20% of recovered value and leave me alone otherwise. For facilities platforms, claims autopilot becomes an upsell to the asset ledger. For brands, AI intake reduces human support load, scores fraud risk, and resolves clean claims fast enough to save the customer relationship.

The commercial split matters. The same underlying workflow can be sold four ways:

• Consumers buy memory.
• Small businesses buy recovery.
• Facilities teams buy asset control.
• Brands buy cleaner intake and fewer angry escalations.

That is why this is a product problem, not a chatbot problem.

The new procurement signal

Procurement teams will learn this faster than consumers.

A school buying 400 laptops does not only care about unit price. It cares about what happens when 37 screens fail, 12 chargers disappear, three serial numbers are rubbed off, and the reseller says the manufacturer has to authorise the repair.

A clinic buying fridges cares about service logs, temperature records, compliance evidence and replacement speed. A landlord cares about transferable warranties when a property changes hands. A retailer cares about fixtures, scanners and payment terminals being repaired without a week of email archaeology.

Vendors with clean post-purchase data will be cheaper to own.

That means receipt-in-email is not enough. Buyers will start asking for:

• Asset IDs.
• Warranty status APIs.
• Parts lists.
• Claim endpoints.
• Transfer rules.
• Service history export.
• Proof captured at purchase.
• Clear escalation paths.

Messy brands become an admin tax. Tidy brands become the default for anyone managing more objects than they can remember.

This is a subtle competitive shift. The brand promise moves from “we stand behind our products” to “we can prove what you bought, what we owe, and what happens next.”

Tone matters less when the audit trail is clean.

The limit case

There are real limits.

Cheap goods are often not worth chasing. Taste-based returns still need judgement. Misuse is hard. Some faults are ambiguous. A cracked hinge might be bad design, rough handling, poor installation, or all three. Agents will overclaim. Consumers may become petty auditors, encouraged by recovery feeds that turn every purchase into a scratchcard.

Support costs can exceed recovered value. A £19 claim that consumes two agents, a warehouse inspection and a return label is a loss even when the customer is right. Brands will need thresholds, automation, and the courage to say no clearly.

The nastiest failures will come from agents reading policy text like executable law when commerce still depends on judgement. A human support manager may approve a technically invalid claim because the customer is valuable, the defect is embarrassing, or the story feels fair. A rigid counter-agent may reject it and save £42 while losing a customer worth £900.

The opposite risk sits on the customer side. An agent that files every marginal claim trains people to see vendors as adversaries. That may recover money in the short term and poison the relationship over time.

Good design will need restraint. Claim readiness does not mean claim everything.

The promise starts executing

The warranty was never asleep. The customer was.

AI memory changes that. It turns old promises into live operational objects: watched, timed, evidenced, filed and chased. The policy copy that used to sit in a PDF starts behaving like a contract with a calendar and a camera attached.

That will feel small at first. A banking app says a return window closes tomorrow. A facilities tool asks for a serial photo before a chair breaks. A retailer offers to register claim evidence at checkout. A support bot receives a packet so complete that “we lost your paperwork” is no longer a plausible sentence.

Then finance notices the claim rate. Procurement notices the admin tax. Product notices that the post-purchase surface has become part of the buying decision.

The next advantage may not belong to the company with the most generous warranty. It may belong to the one whose promises survive contact with agents.

---