A release technique used at Datasite alongside Continuous Deployment Cadence: instead of releasing a feature to the entire customer base at once, it is flagged on for a subset of customers or a cohort first, letting the team observe real usage and outcomes at low risk before deciding whether to expand, adjust, or kill it.
Cohort release is the delivery-side complement to Discovery (Product Discovery Process): discovery generates a hypothesis about a customer problem, and cohort release lets that hypothesis be tested against real behavior with a small, contained population rather than a full-scale bet. It closes the loop between what was learned from customers and what gets built and shipped, lowering the cost of being wrong.
Apply: use feature flags/cohorts to validate discovery insights in production with a small population before a full rollout, rather than either shipping to everyone on faith or treating discovery and delivery as disconnected phases.
Two adjacent delivery techniques Cagan pairs with feature flagging: A/B testing, which routes a subset of traffic to a new version and compares it against a control group to measure real impact, and dark launches, which ship code into production hidden behind a flag before any user sees it, letting the team validate infrastructure and performance before full exposure. Both depend on Telemetry / Instrumentation (Three-Level Monitoring) to be worth doing — without instrumentation there's nothing to compare or watch.
Biddle walks through staging a real feature this way: test custom playback speed on mobile-only devices, in New Zealand, restricted to Netflix-owned original content, before deciding on a broader rollout. The narrow slice lets the team gather real adoption data while limiting stakeholder visibility and risk if the idea flops — the same logic behind Introducing Focus as a Reversible Test.
Из тем: Transformation in Practice