How to use AI to research prospects in 10 minutes.

Build the research Claude Project (one-time, 20 minutes)

Create a Project called "Account Research — [Your Company]." Load these into the knowledge base:

1. Operational ICP definition. Not a tagline — the actual 15–25 structured fields you care about: company size, industry, growth signals, tech stack patterns, etc.
2. Buyer persona docs. Who you're typically selling to. Title patterns, pain patterns, what they care about, what they ignore.
3. Three to five anonymized winning briefs. Actual research briefs that led to closed deals. The AI learns what "good" looks like from examples.
4. Your value proposition + differentiated positioning. Two paragraphs.

System prompt:

You are an account research analyst for [Company]. We sell [what we sell] to [who].

When researching an account:
1. Verify fit against the ICP fields in the knowledge base
2. Identify signals of timing (recent hires, funding, product launches, churn signals, regulatory changes, leadership changes)
3. Identify the likely buyer persona and what they care about
4. Surface what we'd say in outreach — specifics from the research, not generic value-prop language
5. Flag anything that suggests this is NOT a fit so the rep can disqualify quickly

Output format: 1-page structured brief (sections: Fit · Timing · Buyer · Angle · Risk).
If a key data point is missing, write [Unknown: source needed] rather than guessing.

Gather the inputs (1 minute)

For each account, pull together:

— The company URL
— LinkedIn URL of the likely buyer (or 2-3 candidates)
— Anything recent from a news search ("[Company] funding", "[Company] hires CRO")
— Their most recent press release or blog post if findable

Drop them as URLs into a new chat inside the Project.

If you want to scale this — wire it up

Once you've used this workflow manually for ~20 accounts and you trust the output, the next step is to wire it directly into your sales engagement tool (Outreach, Salesloft, Apollo). The brief gets generated automatically when an account moves to "prospect" stage, drops into the rep's queue, and they review rather than create. This is what Treetop typically builds during the Outbound Intelligence phase of an Implementation engagement.

Run the structured brief prompt (2 minutes)

One sentence:

Research [Company Name] using the URL [url] and the LinkedIn profile [url]. Use the standard brief structure. Flag any data points you can't verify.

Claude returns a structured brief. Roughly 1 page, organized exactly the way your team has agreed to read.

Validate and tag (3 minutes)

This is the only manual work left. Read the brief. Verify:

— Two specific claims about the company. Open one tab, check one fact. If wrong, the rest of the brief is suspect.
— The "Angle" section. Is it specific, or did Claude default to generic value-prop language? If generic, ask Claude to rewrite using only specifics from the research.
— The fit verdict. Trust it cautiously — if Claude says "strong fit" but something in the brief contradicts that, dig in.

Then tag with priority (A/B/C) and route to the right rep. Total time so far: ~8 minutes.

Save back to the system (2 minutes)

The brief needs to live somewhere the next workflow can read it. Three options, in order of how scalable they are:

Manual paste into CRM activity note — fine for early days.
Save to a shared knowledge base or Notion — better once you have 50+ accounts in motion.
Wire to CRM via Zapier or native integration — best, but requires a real workflow build.

The point: don't let the brief die in a Claude chat. The whole compound of AI in sales is that this brief becomes input to the next workflow (sequence personalization, call prep, reply triage). Treat it as data, not a one-off output.