A tech company renewed GitHub Enterprise on its own seat data.

A developer estate sized to the headcount it used to have.

The company had scaled its GitHub Enterprise deployment during a stretch of aggressive engineering hiring. Seats were provisioned generously to keep ahead of onboarding, and a GitHub Copilot rollout followed the same logic, assigned broadly on the assumption that every developer would adopt it. At the time the generosity was rational, because the constraint was hiring speed, not licensing cost, and nobody wanted a new engineer waiting on a seat.

Then the growth curve flattened. A reorganization consolidated several teams, hiring slowed, and attrition reclaimed seats that were never deprovisioned. The active developer base settled well below the provisioned count, but the GitHub Enterprise estate kept paying for the peak. The Copilot rollout carried the same overhang: it had been assigned to a far broader population than the data showed were actually using it day to day. The renewal proposal, predictably, took the existing counts as the starting point and applied an uplift, treating the peak footprint as the established baseline.

The gap between provisioned and active had become structural. Seats provisioned for growth do not deprovision themselves when the growth stops. They just keep renewing.

Renewing the active base, restructuring the AI tooling on evidence.

The practice reconstructed actual activity across the GitHub Enterprise estate, distinguishing the active developer base from seats that had gone dormant through reorganization and attrition. The same analysis was applied to Copilot, measuring genuine day to day usage against the broad assignment to separate the developers who relied on it from those who had been provisioned but rarely engaged. The reconstruction produced the defensible active footprint the renewal had been missing.

From that footprint the practice built two moves. The GitHub Enterprise renewal was resized to the active base, reclaiming roughly nineteen hundred dormant seats so the company renewed what it used rather than what it had peaked at. The Copilot rollout was restructured on the adoption evidence: rather than renewing a broad assignment on faith, the practice scoped it to the population genuinely deriving value, with a measured expansion path tied to demonstrated adoption rather than to an assumption that everyone would eventually use it. Peer benchmarks for developer tooling at comparable scale anchored the pricing so the uplift was contested rather than accepted.

Treating Copilot as a measurable tool rather than a default entitlement was the key. An AI seat renewed on the hope of adoption costs the same as one renewed on the proof of it. Only one of them is worth the money.

The active base renewed, and $4.2M of dormant seats gone.

The renewal closed against the active footprint, reclaiming roughly 1,900 dormant GitHub Enterprise seats and restructuring the Copilot rollout to measured adoption, removing an estimated $4.2M in inactive seat spend. The company renewed the developer estate it actually used, with a Copilot deployment scoped to genuine value and a defined path to expand as adoption proved itself rather than ahead of it.

The lasting outcome was a measurement discipline the developer estate had lacked. Instead of provisioning to peak and renewing on autopilot, the company held an active usage baseline for both GitHub Enterprise and Copilot, with a process to reclaim dormant seats and a Copilot expansion gated on evidence. The next renewal would start from real usage rather than accumulated headcount, and the AI tooling investment would track demonstrated value rather than hope.

The engagement reflects the firm's broader record across Microsoft contracts: more than $420M in cumulative client savings, over 340 engagements delivered, and an average 79 percent reduction in audit financial exposure, built on 20+ years of combined practice depth across the Microsoft estate. The figures above are verifiable on a reference call arranged through the practice.