E-commerce Use case Agent · App

Fight chargebacks with proof you already have

Your order, payment, delivery scan and policy already exist in five systems. The agent gathers them the day a dispute lands and holds the response for your yes.

The proof is not missing

It is spread across your store, your processor, a carrier scan and a policy page nobody has opened since launch. Losing is an assembly problem.

Connected where the money moved

Stripe and Shopify Payments, both on official APIs. The dispute arrives by webhook and the order, the charge and the fulfilment are already on file.

Your policy as it read then

Give it the link, not the text. It captures the page and keeps every version, so a March order is answered with March terms.

One submission, never a second

Banks accept evidence once and it cannot be edited afterwards. The packet is assembled and held, with the weak cases flagged, until you say send.

When the courier will not hand it over

Signed delivery often sits behind a carrier login no API key reaches. The AI agent signs in with its own browser and files the document itself.

It learns which fights you win

Every outcome lands against the evidence that was sent. Win rates by reason code, by issuer country, and by whether delivery proof changed anything.

How to set up fight chargebacks with proof you already have on Clawnify

  1. Connect your processor

    Authorise Stripe or Shopify Payments. New disputes arrive by webhook from that point, and one sync pulls in the ones already open.

  2. Point it at your policy pages

    Paste the links to your refund and cancellation pages rather than the text. It captures them and keeps every version, so old orders answer with the terms that applied then.

  3. Say what always needs you

    Choose which reason codes may go out on their own, if any, and which always wait for a person. Everything waits until you change it.

  4. Read the case and send

    Each dispute arrives scored, with the missing pieces named and the evidence attached. Approve it and the packet goes to the bank.

Questions we get asked

What does a chargeback actually cost?

More than the order, and in two parts. A dispute fee lands the moment the dispute is raised and you do not get it back, whether you win, lose or never respond. If you then contest it, a second fee applies and that one returns only if you win. So a small dispute can cost more to fight than to concede, which is why every case here is scored before anything is drafted and why the app will tell you to let one go.

Does winning bring my dispute rate down?

No. The card networks count a dispute when it is raised and explicitly ignore how it ends, because waiting months for outcomes would make the programmes useless. Visa flags an account as non-compliant at a 0.5 percent ratio and excessive at 1.5 percent outside CEMEA, and it counts early fraud warnings alongside actual disputes. Recovering money and protecting your ratio are two different problems. This solves the first. Only prevention solves the second, and anyone selling you a win rate as ratio protection is hoping you never check.

Why do merchants lose not-received disputes they should win?

Almost always because they sent a tracking number. A number is not evidence. What an issuer checks is the carrier's own delivery confirmation, the timestamp, and whether the delivery address matches the address the order shipped to. A public tracking page screenshot is the common substitute and it is discounted, because anyone can produce one for any parcel.

How does it get proof of delivery?

Through the carrier's API where your account can actually reach it, and through the AI agent's browser where it cannot. That second path is not a fallback for edge cases. FedEx returns a signature proof only to a request carrying the shipper's own billing account number, so a merchant fulfilling through a third party is locked out no matter how good the integration is, and USPS delivery records are not reliably available by API at all. In those cases the agent signs in to the carrier portal itself and files the document, tagged with where it came from.

How much evidence can I actually submit?

Less than people expect, which is why selection matters more than collection. Banks take one file per evidence type and cap the whole submission at a few megabytes, with Mastercard also limiting page count. They review nothing external, so a link to your tracking page or your policy is worth nothing. The content has to be inside the packet, which is the reason this captures your policy page rather than pointing at it.

My Shopify app said it submitted and nothing happened.

That is a known behaviour rather than a bug in your app. Shopify's evidence call can return success with no errors while the evidence never reaches the bank and the dispute sits waiting. This checks: it submits, then re-reads the dispute and compares. When the two disagree it says so instead of showing you a tick, and hands the last step to your agent to finish in the admin.

When is this the wrong tool?

If you see a handful of disputes a year, the deadline is the only hard part and a calendar reminder handles it. And it will not win everything. Issuers vary by country, and merchants regularly submit signed delivery on a parcel that was genuinely delivered and lose anyway. What you get is the strongest honest version of your case, on every dispute, without anyone assembling it by hand, plus a record of which fights are actually worth having. It is not a prevention product and it does not decide whether your policy is any good.

More for E-commerce

3 ready