Skip to content

Build a "voice of customer" agent over your support inbox

You don't read every support email the moment it lands — but the themes hiding in them are gold. Build a no-code agent grounded on your support inbox and ask it what customers are really saying, any time.

Stage: Agent Builder · For: Maker, End user · Level: Intermediate · Time: 20 min

When to use this

Your support inbox is the most honest feedback channel you have — and the easiest to drown in. The patterns ("everyone's confused by the new export flow") are there, but spotting them means reading hundreds of messages. This is a perfect personal-scope Agent Builder build: point a declarative agent at your own inbox, give it a "voice of customer analyst" persona, and ask it what's trending whenever you want — no rebuilding the analysis each time.

Use this when support email is a firehose and you want the signal without the daily slog.

What you'll need

  • M365 Copilot license with Agent Builder (Copilot → Create an agent)
  • A support mailbox or a folder/label in your Outlook the agent can ground on (your own, or a shared mailbox you have access to)
  • 20 minutes and the four ingredients: name, instructions, knowledge source, starter prompts

Try it now — the build

In the Agent Builder conversation, describe the analyst you wish you had:

Create an agent called "Voice of Customer" grounded on my support email
[mailbox/folder]. Its job is to surface what customers are actually saying: group
messages into themes, tell me how often each theme comes up, quote a representative
example for each, and flag anything urgent or angry. Always cite the emails it's
drawing from. Stay neutral — report what customers said, don't sugar-coat it. If I ask
about something not in the inbox, say so rather than guessing.

Why this works: it names the agent, sets the knowledge boundary (your support email), defines the analyst behavior (theme + frequency + a real quote + urgency flag), demands citations, and fixes the tone ("don't sugar-coat") plus the refusal rule. A neutral, cited analyst is what makes the output trustworthy instead of cherry-picked.

Step by step

  1. Open Agent Builder. In M365 Copilot, choose Create an agent. Build by chatting or fill the Configure fields directly.
  2. Add the knowledge source. Point it at your support mailbox or folder. That's the boundary — it analyzes this, not your whole mailbox or the web.
  3. Write the instructions (the prompt above, adapted). The "theme + frequency + quote + urgency, neutral tone, always cite" rules are the heart of it.
  4. Add starter prompts people (or future-you) would actually ask: "What are this week's top 3 complaints?", "What's the most common feature request?", "Anything urgent I missed?"
  5. Test in the preview pane. Ask "What are customers frustrated about this month?" Confirm it returns themes with cited examples — then ask something off-topic to confirm it declines.

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

Turn a theme-finder into an early-warning system: - Spot the shifts. "Compare this week's themes to last week's — what's newly trending?" - Quantify it. "Rank the top 10 issues by volume and tag each as bug, confusion, or feature request." - Make it shareable. Once it's good for you, widen the persona and share it with your support or product team so everyone reads from the same signal.

📚 Learn more. Browse Microsoft's Agent Library for templates to crib from, and watch April's AI Agents Academy (April Dunnam, Microsoft) for step-by-step building.

Watch out for

  • Grounding ≠ access control. The agent surfaces anything in the mailbox it's pointed at — only ground it on email you're allowed to analyze, and don't share a personal-inbox agent with others.
  • Customer data is sensitive. Support email contains personal information. Keep the agent personal-scope unless your org's privacy rules clearly allow wider sharing.
  • Themes are a starting point, not a verdict. The agent reports patterns from what it can read; confirm the big claims against the actual emails before you act on them.

Where this leads (the ramp)

You built a personal agent that analyzes — but it still only reads and reports. When you want an agent that takes action (file the urgent ones as tickets, route requests, look something up in another system), declarative building hits its ceiling. That's the cue for Stage 6 · Copilot Studio.

Next: Copilot Studio → Build your first Studio agent with a knowledge source + topic

Was this page helpful?

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