Chargebacks feel like a payments problem and are mostly an operations problem. By the time a dispute arrives, the outcome has largely been decided by things that happened before the parcel shipped.
That is the useful framing, because prevention is achievable and disputing is not reliably winnable.
The three kinds, and why the distinction matters
Genuine fraud: someone used a card they did not own. Prevention is screening, and once it happens the merchant usually loses regardless of evidence.
Merchant error: the item arrived late, damaged, wrong, or never arrived. These are entirely preventable and, when they happen, are better solved by refunding quickly than by disputing.
Friendly fraud: the cardholder made the purchase and disputes it anyway โ sometimes because they did not recognise the charge, sometimes because a return felt harder than a chargeback. This is the largest category in most stores and the most responsive to prevention.
The reason to separate them is that they need opposite responses. Fighting a merchant-error chargeback wastes time you should spend fixing the operation; refunding a friendly-fraud case teaches the pattern.
Preventing the avoidable ones
Make your statement descriptor recognisable. A customer who sees an unfamiliar name on their statement two weeks later disputes it. This is a single setting and it removes a meaningful share of friendly fraud.
Make returns easier than a chargeback. If your return process takes more effort than calling the bank, some customers will call the bank. That is a policy problem wearing a payments costume.
Send tracking and use it. Delivery confirmation is the strongest evidence in a dispute and also prevents the dispute, because a customer who can see where the parcel is does not assume it is lost.
Answer support quickly. A large share of disputes are filed after a customer failed to reach anyone.
Screening without killing conversion
Shopify's fraud analysis flags risky orders. The useful discipline is deciding in advance what you do with a flag rather than judging order by order โ reviewing manually is slow and inconsistent, and inconsistency is what lets fraud through.
Set a rule: what score triggers a hold, what verification you ask for, and how long you wait. Then follow it.
Be careful about over-blocking. Cancelling legitimate orders costs more than the fraud it prevents in most catalogues, and it is invisible in your metrics โ you never see the revenue you refused.
If you do dispute
Evidence that works is specific and verifiable: delivery confirmation with a timestamp, the customer's own messages, the IP and device used at order time, and proof the billing and shipping details matched.
Evidence that does not work: your policies. Attaching your terms of service to a dispute is common and almost never decisive.
Keep a record of outcomes. If a particular product or market produces most of your disputes, that is a product or market decision rather than a payments one.