Reducing batch size is the single highest-leverage practice in software delivery. Smaller increments produce faster feedback, lower risk, shorter cycle times, and higher quality. The result is provable from queuing theory (Reinertsen, The Principles of Product Development Flow) and validated empirically by DORA research, which lists “working in small batches” as a core capability that predicts both software-delivery and organizational performance.
Most organizations default to large batches: weekly sprints, feature branches that live for days or weeks, release trains, staged deployments. Each of these introduces invisible queues. Reinertsen proved mathematically that queues are the dominant source of delay and variability in product development. In software, the queues are concrete: open PRs waiting for review, branches waiting for merge, changes waiting for the next deploy window, bugs waiting for triage. Every one of those waiting items is information that has not yet been confirmed or refuted, and the cost of that ambiguity scales with the size of what’s queued.
The mechanism is feedback. Small diffs are easy to review and fast to merge; large PRs are hard to review, slow to merge, and have higher defect rates. Small deployments expose problems where they can be cheaply rolled back; large deployments couple unrelated changes into the same blast radius. Feature flags and dark launching let incomplete features deploy safely, decoupling deployment from release. The “right size” for a batch is empirical: deployable within hours, reviewable in one sitting, testable in isolation. The INVEST acronym formalizes this for stories: Independent, Negotiable, Valuable, Estimable, Small (hours to days), Testable.
DORA’s 2025 framing called small batches “a critical countermeasure” to delivery instability caused by AI-assisted development: AI raises velocity, which raises the entropy in the change stream, which makes large batches even more dangerous than before. The pairing with Trunk-Based Development and Frequent Deployments Improve Stability is the configuration where the practice compounds: short-lived changes integrating to a single trunk, deployed often enough that each batch is small. See Lean Flow Theory for the queuing-theoretic foundations.