Food Delivery App Cancellation Case Study
A food-delivery app sees cancellations increase by 25 %. The obvious response? “Give customers a discount". I was not convinced. A discount might help an experience.. It does not answer the more important question: "Why are customers cancelling in the place?" So I treated this as a product diagnosis problem.
I started with the customer journey: Order Placed → Restaurant Accepted → Driver Assigned → Out for Delivery The first thing I would want to identify is where in this journey cancellations are increasing. For this case let us assume the data shows that 70 % of cancellations happen after driver assignment while the average wait for a driver has increased from 3 minutes to 9 minutes. That immediately changes the problem I would investigate. This may not simply be a case of “customers are frustrated”.
The next question becomes: What is happening during those 9 minutes? One hypothesis: Repeated driver reassignments are increasing uncertainty and extending the customers wait even after a driver has initially been assigned. If that hypothesis is supported by the data I would shift the intervention away from compensating customers and toward fixing the mechanism.The question then becomes:Can we reduce driver reassignment than compensate customers after it happens?
My initial idea was fixed driver zones, thinking through the marketplace tradeoff changed it. A rigid radius might improve matching in some situations while reducing driver utilisation in others. Instead I would test dynamic hyperlocal allocation during peak periods combined with a driverperformance system that rewards reliable acceptance and completion rather than relying only on penalties. I would not call the solution successful simply because it sounds logical.
I would test it against: → Cancellation rate after driver assignment → Time from order to delivery → Preacceptance bounce rate as a guardrail
The bigger lesson for me is not about food delivery. It is about how I want to approach product problems. A product case rarely starts with the solution someone proposes. “Give customers discounts" is a solution. “Customers may be cancelling because of repeated driver reassignment and increasing wait time" is a hypothesis, about the mechanism, those are different things.I am learning to approach product problems by asking:
Where is the behaviour changing?
What could have changed at that point?
What evidence would reject each hypothesis?
What is the smallest intervention that could change the mechanism?