← FinOpsAid blog

Is Copilot paying for itself? Reading all five tabs of the Copilot page — FinOpsAid blog

Is Copilot paying for itself? Reading all five tabs of the Copilot page

Adoption funnel, acceptance rate, model mix, seat utilization and the ROI tab — every metric on the FinOpsAid Copilot page, what it measures, and which direction is good.

Copilot is the largest page in FinOpsAid — six KPI tiles and five tabs — because it's answering three separate questions that people usually conflate: are people using it, is it working for them, and what is it costing. Adoption, quality of fit, and spend are different metrics with different failure modes, and the tab structure exists to keep them apart.

The conventions used below are defined in the grammar behind every number.

The KPI row — six tiles, three questions

Tile What it counts Direction
Total Spend Copilot cost for the org. Worse.
Active Users Users with usage, shown as N of M seats · X% engagement. Higher is better. The adoption headline.
Acceptance Rate Mean acceptance across active users. Footnote: "avg across active". Higher is better — measures fit, not output.
Interactions Total Copilot interactions this period. Neutral — raw activity.
Acceptances Interactions × acceptance rate. Footnote: "suggestions kept". Neutral.
AI Credits Total credits consumed. Footnote: "signal, not billed". Neutral — a usage quantity, not money.

Two tiles carry the whole adoption story. Active Users shows both the count and the percentage of licensed seats, so the gap between them is licence waste in one glance. AI Credits is explicitly labelled as a signal, because credits inside your included allowance cost nothing — a large credit number is not a large bill.

If Copilot isn't enabled for the org, the page shows an explicit banner saying so rather than a wall of zeroes. Zeroes read as broken; a banner reads as accurate.

Tab 1 — Overview

Copilot Usage Over Time plots daily active users, scoped by the Period filter. Not spend, not credits — people. It's the adoption curve, and a flattening one after a rollout means you've hit the population that was going to adopt without further intervention.

Adoption Funnel is the most diagnostic card on the page.

The Copilot adoption funnel and where each drop-off points Four stages: licensed seats, invited users, active users, and regular users defined as five or more active days. A drop between licensed and invited means seats were bought but never assigned. A drop between invited and active means people were given access but never started. A drop between active and regular means people tried it and did not adopt it. Each gap has a different remedy. ADOPTION FUNNEL — EACH GAP IS A DIFFERENT PROBLEM Licensed 620 100% Invited 540 87% Active 380 61% Regular 190 31% ← tried, never started: onboarding ← started, never stuck: fit or training ← bought, unassigned: reclaim “Regular” means five or more active days — the page states the definition on the row. Every percentage is against Licensed, not against the stage above. Read the gaps, not the levels.
Buying more seats fixes none of these gaps. Licensed-to-invited is a reclaim problem; invited-to-active is onboarding; active-to-regular is fit. Three findings, three owners, one card.

Beside the funnel, three ranked cards: Usage by Model (a donut of modeled cost share per model), Top Teams by Adoption (ranked by acceptance rate, not by spend), and Top Users by AI Credits (the highest usage signal, explicitly not the highest cost).

Tab 2 — Usage

Four tiles — Interactions, Acceptances, Acceptance Rate, AI Credits — then the daily-active-users chart again, then two model cards.

Model Mix (frontier share) is the card that matters here. It states what percentage of Copilot cost goes to frontier models, then breaks cost and share down by tier with a progress bar each.

Frontier share is the single biggest lever on Copilot spend, because frontier models cost a multiple of standard ones for the same interaction. A high frontier share isn't automatically wrong — some work genuinely needs the better model — but it's the number to check before concluding that "usage went up". Usage may have been flat while the mix got more expensive. The per-developer version of this signal is the $/credit column on the Developers page.

Tab 3 — Adoption

The funnel again, alongside Copilot Seat Utilization: a dial showing active seats as a share of licensed seats, with Active / Idle / Pending counts underneath.

Term Meaning
Active Used the seat within the activity window.
Idle Holds a seat, isn't using it. Reclaim candidate.
Pending Cancellation already scheduled — don't count these as savings twice.

Below that, Top Teams by Adoption in a wide two-column layout, ranked by acceptance rate with a progress bar per team.

Tab 4 — ROI & Licensing

Two sections, and the most directly financial part of the page.

Copilot engagement (ROI) has an Org / Enterprise scope toggle. It shows Active Users, Engaged Users, and Acceptance Rate for the latest day, plus a daily acceptance-rate trend. The card subtitle names the exact GitHub endpoint it reads, which is unusual and deliberate — you can verify the number at source.

Active and Engaged are GitHub's own distinction, not ours: engaged is the narrower measure. If the Enterprise toggle shows nothing, the org isn't part of an enterprise install; that's a fact about your GitHub setup, not a data problem.

Copilot seat breakdown is the licence-waste headline:

Tile Meaning Direction
Licensed Seats Total seats paid for. —
Active This Cycle Seats used in the current billing cycle. Higher is better.
Inactive This Cycle Seats paid for and unused. Footnote: "paid, unused — license waste". Lower is better.
Inactive Share The above as a percentage, with a colour-coded bar. Lower is better.
Pending Cancellation Already scheduled to end. —

The inactive-share bar changes colour by threshold: green under 10%, amber from 10%, red at 25% and above. That's the page telling you when the number stops being normal attrition and starts being a problem worth a meeting. The card header also shows your plan type, seat-management setting, and public-code-suggestions policy.

Tab 5 — Users

Top Users by AI Credits, Recent Copilot Activity (most recently active members with their interaction counts), and Inactive Copilot Seats — a table of idle seats with plan, last active date, days idle, and a per-seat action, headed by the modeled annual savings estimate.

That savings figure is modeled: seat price × idle seats × twelve. It assumes every idle seat is genuinely reclaimable, which is a decision a human has to make — someone on parental leave shows as idle and should not be reclaimed. Treat the list as candidates, not as a work order. The fuller method is in find and reclaim idle Copilot seats.

What people get wrong on this page

The 60-second read

Frequently asked questions

What does Copilot acceptance rate actually measure?

How often a suggestion was kept, which indicates whether the tool fits your codebase — not how productive a developer is. Teams on unusual languages, bespoke frameworks or strict lint rules can get real value at a low acceptance rate, so treat a low number as a prompt to ask them.

What is the Copilot adoption funnel telling me?

Where people drop out, and each gap has a different owner. Licensed to invited is seats bought and never assigned, a reclaim problem. Invited to active is people who never started, an onboarding problem. Active to regular, meaning five or more active days, is people who tried it and it did not stick.

Why did our Copilot spend rise when usage stayed flat?

Check frontier share on the Model Mix card. Frontier models cost a multiple of standard ones for the same interaction, so the mix can get more expensive while the volume of work does not change. Model choice is usually the largest single lever on Copilot spend.

Want your own Copilot numbers? Connect your GitHub org — read-only and free during beta — or explore the demo first. There's a fuller ROI treatment in measuring GitHub Copilot ROI. Next: how to read the Licenses page.