What this automates
The first hour of every new opportunity is the same hour: who is this company, what’s been happening to them, who runs the thing we’d sell into, have we ever talked to them before. Sellers either do it badly under time pressure or skip it and wing the first call. This blueprint hands that hour to Researcher and delivers the result to the seller before they’ve opened the record.
It’s the starter blueprint because there are only four nodes and the hard part — multi-step research with citations across the web and your tenant — is entirely Researcher’s problem, not yours. You’re wiring a trigger to an agent that already knows how to do the job. This is also the pattern Microsoft itself documents for new-opportunity research, so you’re building on a paved road.
The design
| Step | Node | What it does |
|---|---|---|
| 1 | Dynamics 365 trigger | Fires when an opportunity row is created. Any CRM connector works; so does a SharePoint list if that’s where your pipeline lives |
| 2 | M365 Copilot node (the brain) | Agent = Researcher; builds the one-page brief from web + tenant — see prompt below |
| 3 | Update record | Writes the brief into a notes/research field on the opportunity (or attaches it as a file) |
| 4 | Teams — post adaptive card | Card to the opportunity owner: account name, three-line summary, link to the full brief on the record |
The M365 Copilot node, configured
Add the Microsoft 365 Copilot node and set Agent = Researcher. The node runs as the connected M365 user, grounded in that user’s mail, files, calendar, and chats — which is exactly what makes the “our history with them” section work without you building any lookup steps.
One thing to know before you write the prompt: in Copilot Chat, Researcher’s signature move is asking clarifying questions before it starts. In a workflow there is nobody to answer. That back-and-forth doesn’t apply in workflow mode, so your prompt has to pre-answer everything Researcher would have asked — scope, sections, length, and sources. A vague prompt that works fine interactively (because you’d refine it in the follow-up) produces a meandering report here.
Prompt for the node:
Research the company below and produce a one-page sales brief. Do not ask
clarifying questions; use exactly this scope.
Sections, in order:
1. SNAPSHOT — what they do, size, HQ, one line.
2. RECENT NEWS — 3-5 developments from the last 6 months (funding,
leadership changes, product launches, layoffs, regulation). Cite each.
3. LEADERSHIP — the 2-4 people most relevant to a [your product category]
purchase: name, title, anything public about their priorities.
4. LIKELY PAIN POINTS — 3 problems this company plausibly has that
[your product] addresses, each tied to evidence from sections 2-3,
not generic industry talk.
5. OUR HISTORY — from internal mail, meetings, and files: past deals,
contacts, or threads with this company. If nothing exists, say
"No prior contact found" — do not speculate.
Length: max 500 words total. Cite sources throughout. Web sources for
1-4, internal sources for 5.
Company: [Account name token]
Website: [Website token]
Opportunity description: [Description token]The “do not speculate” rule in section 5 is the trust-keeper: the first brief that invents a warm relationship with a cold account is the last brief the sales team reads.
Build it (under an hour)
- Flows → New workflow in Copilot Studio. Trigger: Dynamics 365, when a row is added on the opportunity table (or your CRM/SharePoint equivalent).
- Add the Microsoft 365 Copilot node, choose Researcher, paste the prompt, wire the account name, website, and description tokens.
- Test the node alone with node-level testing on a real account you know well — you can grade the brief instantly. Check section 5 hardest.
- Add the update-record step writing Researcher’s output to the opportunity.
- Add the Teams adaptive card to the opportunity owner with a deep link to the record.
- Important: Researcher is deliberately slow — minutes, not seconds. Enable asynchronous responses so the run survives past the two-minute limit. Then publish.
Cost and governance notes
- Credits: one Researcher call per new opportunity. Cheap at ten opportunities a week; run the agent usage estimator if your pipeline sees hundreds, or add a filter step (deal size, segment) so you only research opportunities worth a brief.
- Identity: the node runs as the connected M365 user — the “our history” section reflects that user’s visible mail and files, not the whole org’s. Connect an account with sensible breadth, and tell sellers whose context the brief is built from.
- Latency is fine here: nobody is waiting on screen. The brief landing fifteen minutes after the opportunity is created is still hours ahead of the seller.
Variants worth building once this works
- Pre-meeting edition: trigger on a calendar event with an external domain instead of a CRM row — same Researcher prompt, scoped to the attendee’s company.
- Tiered research: a condition on deal size picks between a short prompt for small deals and the full five-section brief for enterprise ones — same workflow, half the credits.
- Hand-off to the briefing machine: pair with the Morning Briefing Machine — Researcher does the deep account work once at creation, the daily agent keeps it current per meeting.