Skip to content

Equip Autopilot with a continuous customer-health watch

You already know how to rank your book of business worst-first as a Cowork recipe. This is that same skill — equipped onto an always-on agent so the ranked list stays live, instead of being as old as the last time you remembered to run it.

Stage: Autopilots · For: Manager, Champion, End user · Level: Intermediate · Time: 15 min

This is a lineage skill

The customer-health digest is a Cowork recipe — a skill you describe and run on demand. This walkthrough equips that exact capability onto Autopilot (formerly Microsoft Scout) so it runs continuously. Same skill, one rung up the ramp. That progression — recipe → Autopilot skill → governed Studio tool — is the spine of the Skills Catalog.

When to use this

A worst-first health ranking is only useful the day you run it. By Thursday an account that was green on Monday has gone quiet, a renewal clock started ticking, a support thread escalated — and your list doesn't know. Re-running the digest by hand every few days is exactly the kind of recurring job you shouldn't have to remember. Equip it onto Autopilot as a standing skill and the ranked list re-scores itself as the signals change, surfacing the account that just slipped before your next check-in.

As always, Autopilot works on your behalf, with its own identity in your tenant (permissions, audit and governance behind it) and keeps you in the loop — it ranks and flags; you decide where the week goes.

What you'll need

  • Access to Autopilot (private preview) — or, today, the Frontier desktop app still documented on Learn as Microsoft Scout (needs Frontier turned on by your admin, an Intune policy + admin attestation, a Microsoft 365 Copilot license, and a GitHub Copilot Business or Enterprise license). See Find out if you can get Autopilot.
  • Autopilot connected to where account signals live — Outlook, Teams, calendar, and any OneDrive/SharePoint where account notes or docs sit.
  • A working definition of health for your accounts — the signals that mean "at risk" (gone quiet, escalating thread, slipped date, negative sentiment) so the ranking reflects your reality, not a generic one.

Try it now — the prompt

Equip the watch as a named, standing skill:

Take on a standing "customer-health watch" skill and keep it always-on:
- Rank my accounts worst-first by risk, using signals you can see — threads
  gone quiet, escalations, slipped dates, sentiment, and upcoming renewals.
- Re-score as those signals change; I want the list live, not a one-time snapshot.
- Post me a short worst-first brief every Monday, and ping me mid-week the
  moment an account moves sharply the wrong way — with the why and a link.
Don't message any customer directly without checking with me first.

Why this works: it names a reusable skill ("customer-health watch"), gives it a continuous trigger (re-score as signals change), sets the deliverable (worst-first brief + sharp-move alerts), and draws the human-in-the-loop line on the action that carries weight (customer contact). That's a recipe promoted into an always-on capability.

Step by step

  1. Reach Autopilot where you already work (Teams, Outlook, a chat or a channel) and confirm it can see the surfaces your account signals run on — Outlook, Teams, calendar, and the OneDrive/SharePoint where account context lives.
  2. Equip the skill. Paste the prompt above and name your accounts and what "at risk" means for you. You're handing over a standing watch, not asking for a single ranking.
  3. Read the first ranking as calibration. Autopilot posts a worst-first brief. Check whether its risk signals match your gut — this first pass is where you teach it your definition of health.
  4. Tune the signals. Tell Autopilot what to weight ("a renewal inside 60 days always floats to the top," "ignore the internal thread — that's not the customer"). The skill's ranking sharpens each correction.
  5. Act on the sharp movers. When Autopilot pings you that an account moved hard the wrong way, you decide the move — reach out, loop in the team, or watch it. Autopilot surfaces; you steer.
  6. Leave it running. Once calibrated, the list stays live on its own — no more stale rankings, no more remembering to re-run it. The skill is the always-on version of your Cowork recipe.

Screenshots

We deliberately don't ship screenshots that go stale — the Microsoft Copilot UI changes often. Follow the numbered steps above, which we keep current. Maintainers can regenerate fresh captures with the Playwright tool in tooling/screenshots/.

Make it better

  • Feed the Monday brief into your triage. Pair this with the inbox-triage skill so the at-risk accounts show up in the same morning brief as the rest of what needs you.
  • Capture the definition as a shared recipe. Your tuned "what counts as at-risk" logic is worth reusing — write it up as a Cowork recipe so the team ranks health the same way.
  • Graduate it when the actions get real. The day you want the watch to do something governed — auto-open a save-the-account task, write to the CRM — that's the cue to build it as a Studio tool.

Watch out for

  • The ranking is a lens, not the truth. Autopilot ranks on the signals it can see; a quiet account may be fine and a noisy one may be healthy. Use it to aim your attention, not to replace your judgment.
  • You own any outreach. Autopilot stays behind your sign-off on anything that reaches a customer — it flags and drafts on your behalf, you decide what actually gets sent.
  • Garbage signals, garbage ranking. If Autopilot can't see the surfaces where account reality lives, the watch ranks on partial data. Connect the sources that matter, or scope the skill to the accounts it can actually see.
  • It's private preview. Capabilities shift as Autopilot evolves — verify against the Microsoft Scout desktop app docs if behavior differs from what's described here.

Where this leads (the ramp)

You've now taken a recipe you'd run by hand and equipped it onto an always-on agent — the middle rung of the skill lineage. When the watch needs to take real, governed actions against a system of record — open a task, update the CRM, notify a team channel on a rule — you've outgrown the ready-made skill and you're ready to build the tool that does it.

Next: Stage 6 · Copilot Studio — where the watch becomes a governed tool that acts, not just flags

Was this page helpful?

Found an issue or want to suggest an improvement? Open a GitHub issue  ·  Star on GitHub