We’ve paid for Copilot connectors to ServiceNow and Jira — how do we prove anyone uses them?

That’s a fair question to ask before anyone renews a connector project. Connectors cost build time, sync capacity and someone’s attention when the ACLs drift, so “people seem to like it” isn’t enough.

Here’s what’s going on. Roadmap 519571 (Usage reports – Microsoft 365 Copilot Connectors) is In development on the MRC, with preview from June 2026 and general availability targeted for November 2026 (Worldwide). Learn already documents the report as public preview, under Reports > Usage > Microsoft Copilot > Connectors in the Microsoft 365 admin centre, with 7-day and 30-day views and data that refreshes within about an hour of activity. Learn’s availability table lists it for the public cloud only for now, not GCC, GCC High, DoD or 21Vianet.

What it counts is narrower than “people opened Copilot”, and that’s a good thing. A user is active only when they have a Microsoft Copilot licence, submit a real prompt in Copilot Chat, Copilot Search or an agent, and get back a response that is grounded on (cites) content from one or more of your connectors. Just opening Copilot doesn’t count. The headline metrics are Connections used by Copilot, Connections used by Agents, Active Connector users and Connector responses. Below that are a connector activity table (connection ID, active users, responses provided, last activity date) and a user activity table (connections used, responses received, last activity date). Learn notes that connector use inside Cowork isn’t counted yet.

On names: like other admin centre usage reports, user, group and site details are hidden by default, and the admin centre setting decides whether they’re shown. The user-level table includes everyone licensed for Copilot in the past 180 days, even those with no activity, which is useful for spotting who has never touched a connector.

The plain test: if a connector has no grounded responses after 30 days, it isn’t delivering value, however good the demo looked.

Options that actually work:

1. Run a 30-day review per connector. Use the connector activity table to sort by responses and last activity date; anything at zero goes to its owner with a keep, fix or retire decision. Best Steady path.
2. Pair it with the agent view. Compare Connections used by Copilot with Connections used by Agents; a connector only used through one agent is a candidate to scope down rather than expose tenant-wide.
3. Keep names hidden for the business review. Connector-level numbers answer the value question without naming staff; only reveal user details for a specific, documented purpose.
4. Turn on named reporting and send league tables to managers. Fast way to lose trust, and it measures the wrong thing anyway.

12 Panes action / comms: Tell connector owners: 519571 / Learn: the Connectors usage report counts licensed users who got a connector-grounded answer; we’ll review responses and last activity at 30 days. Tell the DPO: Names are hidden by default; we’ll keep it that way for value reviews. Tell Finance: We’ll have evidence to keep or retire each connector before renewal. Leadership: We measure grounded answers, not clicks.

Benefits / risks: Clear, citation-based definition of use; per-connector and per-user views; near-real-time refresh; still preview until GA (Nov 2026 target); Cowork use isn’t counted yet; unlicensed users don’t appear, so it won’t tell you about Copilot Chat-only staff; named data can drift into performance management if you’re careless.

I’d suggest this month: open the Connectors report, export the 30-day connector table, and book a 20-minute review with each connector owner. Connectors that earn their keep get promoted; the rest get a fix-by date.

DIGITAL SKILL WORKS LTD / 12 PANES · You ask, we answer