Observability rebuild for a foodtech platform

The client is a UK food-delivery platform whose revenue concentrates sharply into dinner-rush peaks. During those windows, every minute of a partial outage translates directly into failed orders and lost revenue.

Their observability was fragmented across disconnected tools, so diagnosing incidents at peak was slow and stressful.

At dinner-rush peaks, exactly when revenue concentrates, incidents turned into a scavenger hunt across disconnected tools while orders quietly failed. The tooling worked against the team at the worst possible moment.

Engineers correlated signals by hand across systems that did not talk to each other, so detection lagged customer impact badly. Problems were felt by customers before they were seen by engineers.

The platform needed to see problems as customers felt them, in real time. Detection had to be tied to customer impact, not to noisy infrastructure signals.

We instrumented services with OpenTelemetry, producing consistent traces, metrics, and logs from one standard. A single, consistent signal set is what makes fast diagnosis possible.

Those signals were unified behind a single pane, so engineers stop hunting and start diagnosing. The scavenger hunt was replaced by a coherent picture.

We set SLOs that page on genuine customer impact (like failed orders) rather than on noisy infrastructure metrics, and tuned alerting against real incidents so pages meant something and fatigue dropped. Alerts now signal problems customers actually feel.