Start with the arithmetic. A delivery business doing $80,000 a month at a 2% refund rate gives back $1,600. If your net margin is 6%, that $1,600 is a third of your profit, handed back as apologies. The usual reaction is to defend the till: tighter policies, more suspicion, "no refunds after 24 hours." That's the wrong move. Refunds are a pile of separate symptoms, and most of the fixes are operational rather than changes to your policy.
Most refunds trace back to five causes
Tag every refund for a month and a distribution shows up. Operators who bother to do it tend to land near this split:
Read it from the top. Half the leak is missing and wrong items. That's accuracy, something you control at the counter. Another 40% (cold, damaged, late) is logistics. The fraud everyone writes policy against is the smallest slice on the chart.
Accuracy failures usually start with re-typing
Missing items happen at the pass, and they cluster wherever an order gets re-typed: from a marketplace tablet into the POS, from a phone call onto a paper ticket. Every re-entry is a coin flip you lose sooner or later. The fixes, roughly in order of payoff: send orders straight to a kitchen screen through orders management so nobody re-enters them; turn bagging into a checklist against the order, with sauces and drinks on their own lines (those two go missing most); print item counts on the label so the courier can check the bag without opening it. Operators who add nothing but the bagging checklist see the missing-item slice fall by half within a few weeks.
If you can't answer in 10 seconds, that's the first problem. We'll show you the tracking setup.
Cold, damaged, late: refunds as a logistics metric
Food that leaves the kitchen perfect and arrives refund-worthy lost something on the way. Three usual suspects: the bag sat on the pass because no courier was assigned when cooking finished; the courier stacked three orders in a sequence that turned yours into a 25-minute detour; or the delivery zone is simply drawn too wide for hot food. These are dispatching problems, and the timestamps expose them if you track the segments (cook-end to pickup, pickup to door) on a live monitoring panel. A dispatcher who sees "order ready, no courier for 9 minutes" can head off tonight's refund before it happens, and the retention data says the same fix protects next month's orders too.
Packaging earns its own line here. A $0.40 upgrade (vented containers for fried items, sealed cups for sauces, a tamper sticker so the bag reads as unopened) looks like pure cost until you set it against the refunds it heads off. If it saves two $25 refunds a week, it pays for itself at about 120 orders. Most operators who run that number upgrade the box.
Chargebacks cost more than refunds, so handle them separately
A refund costs you the order. A chargeback costs you the order plus a $15–25 bank fee plus a strike against your merchant account, and enough strikes push your processing rates up. Two habits keep them rare. First, make a refund easier to get than a dispute: a customer who can claim a fair refund in two taps has no reason to phone their bank. Second, when a dispute does land, answer it with evidence: order confirmation, delivery GPS trace, doorstep photo, the customer's reorder history. When a platform logs every order event end to end, you win the honest disputes and make the dishonest ones too much work to bother with.
Build a menu of remedies
Cash back is only one way to say sorry. A tiered set of remedies settles most complaints for less, and often leaves the customer happier: redeliver the missing item when a courier is nearby; offer account credit worth a little more than the cash ($8 in credit instead of a $6 refund costs you food margin rather than cash, and it books the next order); keep full cash refunds for real failures. Customers take the credit far more often than owners expect, because most people who complain want the food put right more than they want their money back.
Be generous fast, but log everything
Generosity and discipline work together here. Refund genuine problems on the spot, no friction: the LTV math says a $6 apology that keeps a $400 customer is the cheapest marketing you'll ever buy. Then log every refund against the customer's profile and the cause. The honest majority barely costs you anything. The three customers claiming a "missing item" every single week are a pattern your data should flag on its own, and the answer there is a photo-confirmed handoff, not an argument. Review the cause split each month in reports; it moves as you fix things, and the tallest bar is always your next project.
Do this next
Start tagging today: five causes, one dropdown, every refund. In 30 days you'll have your own version of the chart above, and unlike the industry averages it comes with real names, zones, and shifts attached. Fix the tallest bar first.
Kitchen screens, live dispatch monitoring, and refund analytics in one platform — running in about two weeks.