Dark kitchen in Marseille: 3 months of refunds analysed across 8 brands

Between June and August 2026, KonkiCam tracked the delivery activity of a Marseille dark kitchen running eight sushi, burger and poke bowl brands.

Over three months, the business handled 3,296 orders, representing €116,015 in revenue.

In a multi-brand environment where dozens of orders can be prepared at the same time, the goal was simple: to quickly check what had actually been prepared when a customer reported a missing or incorrect item.

86 refunds to check

Over the period, 86 orders led to a dispute – around 2.6% of orders.

In 76 cases, a photo was available to analyse the preparation. Refunds backed by evidence totalled €940.10.

Conversely, 10 refunds had no photo at all, representing €98.82 in losses that could not be properly documented.

This first result illustrates a simple reality: when there is no evidence, the restaurant has very little to go on to check the complaint.

22% genuine errors, 78% disputable cases

Above all, analysing refunds makes it possible to distinguish two very different situations:

• 22% corresponded to an identified preparation error;
• in the remaining 78%, the available evidence did not support the error attributed to the restaurant.

86disputes

78%Disputable complaints

22%Preparation errors

It would be wrong to automatically label these 78% as fraud. The photo shows what was in the order when it was prepared. It does not necessarily show what happened afterwards.

What it does allow the restaurant to establish is a key point: the order appeared correct when it left the preparation area.

49 disputes won, worth €621.79

Over the period, 67 cases were disputed. 49 were successful – a 73% success rate.

49 / 67disputes accepted73% success rate · €621.79 recovered

So the point of evidence is not to dispute every refund. It lets you sort them:

• incorrect order → identify the error and don’t dispute;
• correct order → have what you need to document the dispute.

The success rate rose sharply in three months

The monthly breakdown also shows an interesting trend.

The dispute success rate thus rose from 54% in June to 87% in August.

This progress is particularly relevant for a dark kitchen: the more systematic case qualification and use of evidence become, the more teams can focus their time on complaints that are genuinely defensible.

The restaurant’s own errors remain visible too

Photo evidence isn’t only used to challenge certain complaints. It also confirms internal errors.

Over the three months studied, the observed error rate ranged from 0.28% to 0.76% of orders. In a business running eight different brands, this information can help identify:

• frequently forgotten items;
• errors linked to specific brands;
• orders with several bags;
• problems during peak periods;
• preparation steps that need more checks.

A complaint thus becomes quality-control data, not just a financial loss.

Above all, a photo has to be usable

The dashboard also highlights an important limitation: while 91% of eligible orders have a photo, only 84% of photos are considered fully usable.

Evidence can lose its value if:

• an item is hidden;
• the order is already closed;
• the framing doesn’t cover the whole area;
• the image doesn’t show the contents clearly enough.

A particular challenge for multi-brand dark kitchens

In a dark kitchen running several concepts at once, operational risk is amplified. The same team may have to prepare, within a very short window, a poke bowl for one brand, a burger for another, then several sushi orders under a third.

So the challenge isn’t just photographing orders. You need to be able to quickly find the right order, the right brand, the right time and the right photo.

It’s this link between order data and visual evidence that makes traceability genuinely usable.