A refund is something you decide to give. A chargeback is something that is taken. The customer calls their bank, says the order was wrong or never arrived, and the bank pulls the money back out of your account — often with a dispute fee on top — before anyone has asked you a single question. Then you get a few weeks to prove them wrong, in writing, to someone who has never seen a kitchen.

If you take card payments directly — on your own website, over the phone, at the table — this is your problem, not the delivery app's. Here is how it works and how to fight it well.

Who gets the chargeback

Chargebacks follow the merchant of record: whoever charged the card.

  • Orders through Uber Eats, Deliveroo, Glovo and the like: the platform charged the card, so the platform receives the chargeback. You may see the consequence as a refund they attribute to you, but the bank process is theirs. Our guide on how the platforms handle complaints covers that side.
  • Orders through your own site, app, phone line or POS: you charged the card. The chargeback comes to you via your payment provider (Stripe, SumUp, your bank's terminal contract), and so does the fee.

The rest of this guide is about the second case.

What the timeline looks like

The exact numbers depend on the card network and your provider, so check your provider's documentation — but the shape is always the same:

  1. The cardholder disputes the charge with their bank, choosing a reason: "product not as described", "product not received", "fraudulent transaction". A photo of the "wrong" or "spoiled" food is the usual attachment for the first one.
  2. The funds are withdrawn from your account immediately, plus a dispute fee that most providers keep whether you win or lose.
  3. You are notified and given a deadline to respond — typically a couple of weeks, sometimes less. Miss it and the chargeback stands.
  4. You submit evidence through your provider's dashboard. This is called representment.
  5. The cardholder's bank decides. If you win, the funds return. If you lose, there may be a second round (pre-arbitration) that is rarely worth the cost for a single order.

Nothing in that flow involves anyone looking at your food. It is a documents contest, and the customer opened with a photo.

Why the photo wins by default

Bank dispute analysts handle hundreds of cases a day across every kind of merchant. They are not food experts and they are not image experts. A photo that looks like a bad meal, attached to a claim that says "not as described", is persuasive in the ten seconds it gets. Your representment has to be more persuasive in the same ten seconds — and it has to be evidence, not a story.

"Our food is always fresh" is a story. "Attached: our timestamped photo of the order at the pass; the delivery confirmation; the customer's photo with the edited region highlighted and the metadata analysis showing it was last saved by an editing application" is evidence.

What providers and banks actually accept

Every provider publishes an evidence checklist for each dispute reason. The items that consistently matter for a food order:

  • Proof the order was fulfilled as described: an itemised receipt, the order confirmation, and — the single most valuable thing you can own — a photo of the packed order taken before it left.
  • Proof of delivery or collection: courier confirmation, a signature, a timestamped handover, CCTV of the collection.
  • Your terms, if the customer agreed to them at checkout (refund policy, collection window).
  • Communication with the customer: the complaint, your reply, any refund you already offered. A chargeback filed after you refunded is a common form of double-dipping and is usually won by showing the refund.
  • Analysis of the customer's photo, when the claim rests on it. A short, factual report — this region shows signs of manipulation, the file carries an editor tag, the capture time predates the order — is the kind of specific, checkable statement analysts can act on.

Keep it to one or two pages. Analysts do not read appendices.

Where a photo check fits

When the customer's evidence is a photo, the strongest move is to answer the photo directly. A tool like FraudBite gives you three things you can put straight into a representment: a risk score, a heatmap of the regions that do not match the rest of the image, and a list of the detected features in plain language. Export the PDF, attach it, and reference it in one sentence.

Be precise about what it shows. It shows that the image bears signs of editing or generation; it does not prove who did it or why. Analysts respond well to that precision — and badly to accusations.

The "repeat" problem

Chargeback fraud is habit-forming, and repeat offenders exist. If a card, an address or a phone number disputes twice, block it from ordering directly and note it. Most card processors also let you enrol in the networks' alert programmes, which notify you when a dispute is opened so you can refund pre-emptively and avoid the fee when the claim is genuine — and hold your ground when it is not.

Reducing chargebacks before they happen

  • Photograph every direct order at the pass. Cheap, fast, and it wins more disputes than any argument.
  • Put your trading name on the card statement exactly as it appears on the shop front — a surprising number of "fraudulent transaction" disputes are customers not recognising you.
  • Send an order confirmation with the itemised order, so "not as described" has something to be compared against.
  • Make it easy to complain to you first. A customer who can get a fair answer from you in five minutes rarely calls their bank.

When the next chargeback arrives with a photo attached, check the photo first. If it is genuine, refund gracefully and skip the fee. If it is not, you will know what to write.