Platform engineering pays rent in queue time
The research says platforms help and hurt on different clocks. A guide to reading your own platform investment straight, at month three and month eighteen.
Beyond Delivery Partners2 min readDelivery
Platform engineering spent the decade's middle years as the answer to everything, which is usually the sign a correction is due, and the correction arrived with data. The 2024 DORA report found internal platforms associated with higher individual and team productivity, alongside an initial dip in delivery throughput and stability. Both effects, at once. The platform helps the people and, for a while, slows the system.
The dip has ordinary causes. A platform migration is itself a workload: teams pause feature work to move, the platform team absorbs edge cases it did not anticipate, and the paved path is at first narrower than the dirt roads it replaces. Meanwhile the benefits, less duplicated toil, safer defaults, faster onboarding, compound on a longer clock. The investment is real on both curves. The mistake is measuring it on only one.
The pattern we look for in reviews is not whether a platform exists but where its queues are. A platform is a service with customers, and the honest health check is the same as any service: how long does a team wait for the thing the platform gatekeeps. Environment requests, pipeline changes, new-service scaffolding. When the platform team's ticket queue becomes the org's longest wait, the platform has recreated centrally the bottleneck it was funded to remove, with better branding.
Two practices separate the platforms that pay rent from the ones that collect it. Self-service as the default: anything a team must ask for is a queue, anything a team can do on the paved path is a product. And the platform team measuring itself on its customers' delivery numbers, not its own roadmap completion, which keeps the incentive pointed outward.
If you are eighteen months in and the dip has not reversed, the kind reading is that your platform is still a migration, and the unkind reading is that it has become a department. The way to tell the difference is queue time, team by team, before and after. The data exists in your tooling. Most platform reviews we run are the first time anyone has joined it.