AI prospect research: build a prospect list from evidence, not guesses

An AI agent can research a market in an afternoon that would take a person a week. That's real, and it's also the trap. A fast list of plausible-looking companies, with confident descriptions and no sources, is worse than no list: you'll waste outreach on poor fits and embarrass yourself with claims that aren't true.

The method below treats research as evidence collection. It's how we brief agents at Fulcrum Cortex, and it slots straight into an agent-native CRM, where the researched prospects become contacts with their evidence attached.

Start AI prospect research with the question, not the list

'Find me 100 leads' is a bad brief. 'Find Australian independent hotels that are actively marketing to Japanese travellers and still run an old booking system' is a good one, because it tells the agent what evidence to look for. Before you ask for anything, write down:

• Who: the type of organisation, size band and location.
• Signal: the observable evidence that makes them a fit (they advertise in a particular market, they're hiring for a role, they use a certain system).
• Disqualifiers: what rules a company out, such as part of a national chain, or already a customer.
• What you'll do with it: a call list, a proposal, a mail-out. This decides which contact details you actually need.

Here's how a published Cortex customer used exactly that shape of brief:

Individual results vary. The list was 29 companies, not 500, and every row had an evidence column. That is the point.

Columns that make a research spreadsheet useful

Most prospect spreadsheets have too many columns about the company and none about why it's on the list. Use these, in this order:

Leave out anything you won't use. If you're going to send a proposal to the general manager, you don't need the names of every staff member.

Collect only the data you need

Research agents can find a lot about people. That doesn't mean you should keep it. If the Privacy Act 1988 applies to your business, the Australian Privacy Principles require reasonable steps to keep personal information accurate and up to date (APP 10) and to protect it (APP 11), and they put conditions on using personal information for direct marketing (APP 7) (OAIC: APP quick reference). Even if you're exempt as a small business, data you don't hold can't leak and doesn't need maintaining.

Practical rules for the agent: prefer business contact routes published by the organisation itself; don't collect personal mobiles or home addresses; don't scrape sites whose terms prohibit it; and record where each contact detail came from. This is general information, not legal advice.

Check a claim before you use it in outreach

The most damaging research errors are the ones that end up in an email: 'Congratulations on opening your Brisbane site' to a company that closed it, or 'I see you're using X' when they switched last year. Before any claim appears in outreach, it needs a check.

1. Is there a source? If the agent can't link to it, the claim doesn't go in the message.
2. Is the source the organisation itself? Their own website, announcements or listings beat third-party directories.
3. Is it current? Check the evidence date. Anything over a year old gets re-checked or softened.
4. Is it a fact or an inference? 'Their booking page looks dated' is an inference. It can be useful to you; it doesn't belong in a cold email as a statement.
5. Would they be happy to see it? If a claim relies on information they'd find intrusive, leave it out.

A lead is not an opportunity

A researched company is a lead: someone who might fit. It becomes an opportunity only when there's evidence of a need and some engagement from them, such as a reply, a call or a request for a quote. Mixing the two inflates your pipeline and pushes people into deal stages they never entered.

Keep research results in the CRM as contacts with a 'lead' status and their evidence attached. Create a deal only when there's a real conversation. Once outreach starts, every contact needs a next action and opt-outs need to be honoured; that's covered in our article on the CRM follow-up process. If you're importing a research spreadsheet, run it through the contact list cleaning checklist first.

Reviewing the quality of a prospect list

Before anyone contacts a single prospect, spend 20 minutes on a quality check. Pick five rows at random and five top-rated rows, and for each one:

• Open the evidence link. Does it say what the fit signal claims?
• Is the company still operating at that address?
• Does the fit rating's reason make sense to you?
• Is anything on the row a guess presented as a fact?

If two or more of the ten fail, don't fix the rows one by one. Change the brief and rerun the research. The errors are almost always caused by a vague fit signal.

Keeping a prospect list current

Lists decay. People change jobs, companies rebrand, sites close. A list you built in March is partly wrong by September. Rather than rebuilding it, schedule a refresh: once a quarter, ask the agent to re-check the evidence link and contact route for every lead you haven't contacted yet, and mark what changed. Drop anything that no longer fits.

The same applies to channels you might be tempted to mine by hand, such as professional networks. Copying profiles into tabs and spreadsheets doesn't scale, and automated collection may breach the platform's terms, so read them first. A better approach is to research organisations from their own public information, then reach the right person through the organisation's published contact route.

From research to proposal

Good research is the raw material for a strong proposal. The Nitrobots example went on to a client proposal page and ROI calculator, built and approved before going live. We walk through that step in turning market research into a client proposal.

If you'd like researched prospects to land in your CRM with their evidence already attached, see how Fulcrum CRM holds contacts, notes and next actions written by agents.