Four days to under an hour on API support

The problem

Northbank's API served 340 integrators across three continents, and every support conversation began with a screenshot. Engineers grepped gateway logs by hand, matched timestamps against a customer's local clock and guessed at the rest.

The cost was not the grep. It was that a third of tickets escalated to the payments team, who were the only people who could read the ledger service's logs — and who were also the people shipping the roadmap.

What they did

• A wake ID in every error

  Northbank surfaced the Flintwake wake ID in the error body and the response headers. Support started every conversation with an identifier instead of a screenshot.

• Consumer-scoped history

  Each integrator got read access to their own wakes through the console. Roughly 40% of tickets stopped being filed at all.

• Replay before escalation

  Tier-2 support replays the failing request against staging before writing a Jira ticket. If it reproduces, the ticket includes the diff; if it does not, the ticket is closed.

What changed

• Median resolution fell from four days to fifty-one minutes over one quarter.
• Escalations to the payments team dropped by 68%, freeing roughly nine engineer-days a month.
• The integrator NPS question about support moved from 22 to 61.

Their stack

• Go origin services
• Kubernetes on EKS
• Flintwake edge collector
• PagerDuty
• Snowflake export

Northbank Payments is a fictional company created for this template. Every figure on this page is demonstration data.