Understanding your Copilot spend
GitHub Copilot is often the first AI tool an organization pays for, and one of the hardest to understand. The invoice shows a single number, but the real questions sit behind it: how many people are actually using Copilot, how much usage the plan’s included credits covered, and how much of the work that shipped did Copilot contribute to. This release brings those answers into the same view as the rest of your AI spend. It also adds the first industry reference point next to your Compass Score, and every signed evidence bundle now includes the address where its verification key is published.
GitHub joins your spend connections
Connect a GitHub organization and Copilot spend shows up the way the rest of your AI spend does: attributed daily, model by model. AI-credit consumption is tracked in three parts — what was consumed, what your plan’s included credit pool covered, and what was billed as overage — so usage covered by the plan stays visible instead of disappearing into a $0 line. Copilot seat fees and GitHub Models usage are captured as well, and the backfill includes premium-request history from before GitHub moved to usage-based billing in June 2026.
How Copilot is actually used
Spend tells you what Copilot costs. The new adoption card on the Spend page shows how it is being used: daily active users, suggestion acceptance rate, lines of code added, and the share of merged pull requests Copilot contributed to — alongside seat counts and credit consumption from the same connection.
There is one limitation, and we enforce it in the product instead of burying it in a footnote. GitHub counts a person’s activity in every organization they belong to. If you connect more than one GitHub organization, PulseAI hides the metrics GitHub cannot deduplicate across them rather than show inflated numbers, and the card explains what was hidden and why.
Compare your Compass Score to your industry
The dashboard now includes an Industry Benchmark card, and the AI Accountability Compass™ radar gains a benchmark overlay — both showing how your organization’s Compass Score compares to organizations in your sector and market.
Benchmarks calculated from PulseAI customer scores are protected by an anonymity floor: a figure is only published when the cohort behind it contains at least five organizations, overall and for every scoring domain, so a small cohort can never reveal an individual organization’s score. Benchmarks can also draw on curated figures from external industry sources, which are aggregate data to begin with, not customer scores. If neither is available for your cohort yet, the card says “Not yet available” — we would rather show nothing than show a number we cannot stand behind.
Evidence bundles carry their own verification instructions
Every signed Evidence Vault bundle now states where PulseAI’s evidence-signing key is published: in the signed manifest, in the summary PDF, and on the in-product reference page. The address, pulseai.now/trust/#evidence-verification, is part of the bundle’s signed bytes, so keeping the page at that address is a commitment we cannot quietly walk back. An auditor holding nothing but the bundle knows where to fetch the key and check its fingerprint. Bundles created before this release continue to verify unchanged.
Also in this release
- The Model Registry’s provider and status filters now come from your own records, so every provider and status you actually use is selectable, and provider entry accepts free text instead of a fixed vendor list.
- The OSFI E-23 model appendix spells lifecycle stages and deployment methods correctly (“Re-evaluation”, “Vendor API”), and the Vendor Impact report offers exactly the providers your model records hold.
- First page load is faster: the application now loads each area on demand.
- Help icons across the application link each page — and individual tabs where it matters — to the relevant documentation articles.
- Onboarding reports a failed load as a failure, with a way to retry, instead of presenting it as an organization with no departments.