Delivery order mistakes: what they really cost and how to eliminate them
An item missing from a bag looks like a two-euro problem. Once you add up the five layers of cost it drags along, the sum looks rather different.
When a restaurant looks at what delivery mistakes cost, it almost always looks at a single figure: the month's refunded amount. It is the easiest one to find and the one that says least, because it is the only layer of the cost that appears on a statement.
What does a mistake on a delivery order really cost?
An order mistake has five layers of cost: the amount refunded to the customer, the food that was prepared and wasted, the team's time handling the incident, the visibility penalty on the platform and the value of the customer who does not come back. Only the first appears on your statement, and in most cases it is the smallest of the five.
The five layers, one by one
1. The amount refunded
This is the visible cost: the platform refunds the customer for the item claimed and charges it to you. And if the refund is applied to the whole order rather than the line claimed, the amount multiplies.
2. The wasted food
The item was made with ingredients you had already paid for and it went out of the door. When that order is refunded, the food cost is spent and does not come back. In a tight-margin model, every refund eats the margin from several correct orders.
3. The admin time
Somebody has to read the claim, find the order, ask the shift, decide whether to challenge it and file the case. Those minutes belong to a manager, the most expensive person on the shift, and they get paid for whether or not they are in the budget. Multiply the minutes per incident by the number of incidents in the month: it is nearly always the surprise in the calculation.
4. The platform penalty
The platforms' quality metrics govern your visibility and your access to promotions. A high rate of items not received or avoidable cancellations pushes you out of the top of the listings, and that does not show up as a charge: it shows up as a drop in orders nobody can explain. It is exactly what was blocking Deleito before they corrected their metric.
5. The customer who does not come back
A customer who receives an incomplete order orders somewhere else next time, and does not tell you. It appears in no report, but it is the most expensive of them all in the medium term, because it erodes the repeat business the whole model rests on.
How to work it out with your own data
You do not need a complicated model. Four figures from any given month give you a defensible estimate:
| Figure | Where to find it | What it is for |
|---|---|---|
| Orders with an incident | Each platform's order history, filtered by refund or complaint. | It is the volume everything else gets multiplied by. |
| Average amount refunded | The month's statements: total charged divided by the number of incidents. | Layer 1 of the cost. |
| Food cost per order | Your recipe costing, applied to the average order value. | Layer 2: what gets thrown away with every refund. |
| Minutes of admin per incident | Ask the manager and time it for a week. | Layer 3, the one never in the budget. |
With that, the monthly direct cost is: orders with an incident × (amount refunded + food cost + minutes × the manager's cost per minute). Do not try to quantify layers 4 and 5: just note whether your quality metrics sit outside the range the platform asks for, because if they do, the opportunity cost is greater than everything else combined.
Which levers actually move the needle
There are three, and their order of return is always the same:
- Prevent the mistake before the bag goes out. It is the only lever that tackles all five layers at once. An item added in time generates no refund, no wasted food, no admin, no penalty and no lost customer.
- Recover what is not yours to pay. When the claim is unfair, disputing it with evidence returns layer 1. That is real money, though it does not fix the rest.
- Cut the admin time. If every incident costs twenty minutes of a manager's time, automating it is worth more than it looks.
Bronze's numbers, so you can compare
These are our figures as of July 2026, measured on real customer data:
If you add the mistakes corrected before orders leave the kitchen to the return from disputes, the estimated return rises to 15×. We label it an estimate because it is calculated on the refunds that correcting orders in the kitchen prevents, and that, by definition, is a cost that never happened. When we put a proposal together for you, we show you the full calculation with your data.
The figure that best sums up the effect of prevention: in restaurants that verify every order, fewer than 1% of verified orders end in a net refund.
How the chain is broken in practice
Prevention is not “checking more carefully”. Checking more carefully works on a quiet Tuesday and fails at 21:30 on a Friday, which is exactly when the orders that generate the incidents go out.
Bronze Tracker turns that check into a two-second gesture: showing the order to Bronze. Computer vision compares what was prepared against the order from your POS or delivery platform and flags on screen if anything is missing or extra, in under a second per order. That same photo is documented and is what backs the dispute when a claim does not match reality.
In operational terms: in a case like Nugu Burger, with 99.2% of orders verified, more than 11% of them needed intervention. One in nine orders would have gone out with a mistake had nobody checked.
What to take away
- The amount refunded is the visible layer, and almost never the largest.
- Four figures from one month give you a defensible estimate of the direct cost.
- Preventing it in the kitchen tackles all five layers; challenging it returns only one.
