How to Research a Prospect Before Outreach (Using AI)

Researching a prospect before outreach means answering five questions before the message exists: does the account fit your ICP, who is the right person, what situation are they in, what proven asset should the touch lead with, and is there any reason not to send at all. AI's contribution isn't deeper research than a human could do — it's that the research actually happens on every account, every night, instead of only when someone has time.

Most cold outreach skips research because research is the expensive part. A human doing an honest job on one account — verifying fit, finding the right person, understanding the situation — burns real minutes, and at any volume the minutes win and the corner gets cut. That's the actual reason your inbox is full of "quick question" emails from strangers who clearly don't know what your company does. This guide is the research layer done properly, using the architecture of Optimus Hunter — the agent that runs outbound for the Optimus ecosystem — as the worked example.

Question 1 — Does this account actually fit?

Fit comes first because everything downstream is wasted if fit fails. Check the account against a profile that can reject it: revenue band, niche, and stage. Hunter's gate is $5–$50M companies inside the target niche — checked against live data at hunt time, not against a list that was true last quarter. The mechanics of building that gate are covered in how to find ideal buyers without buying a list.

Just as important: check your own records. Is this account already a customer? Already in sequence? Did anyone at this company ever opt out? A single cold pitch to an existing client — or worse, to someone who already said no — costs more trust than a hundred good sends earn. Hunter dedupes every night's targets against the entire pipeline and keeps opt-out suppression permanent, tracked by fingerprint so it survives renames.

Question 2 — Who is the right person?

The right person is whoever feels the problem your asset addresses — not whoever was easiest to find an email for. At $5–$50M companies that's usually the founder or a specific C-suite role, which is exactly who Hunter surfaces by name. Two rules of thumb:

• Below a certain company size, go to the founder. They still touch everything, and a relevant asset lands directly.
• Pick one person per account. Carpet-bombing four executives at the same company with the same message guarantees they compare notes — and that the message reads as the mail-merge it is.

Question 3 — What situation is this account in?

This is where "research" usually collapses into flattery — quoting a podcast appearance, referencing a LinkedIn post. Skip it. AI made that species of personalization free to fake, so readers discount it to zero. What still reads as genuine is situational relevance: understanding what a company of this size, in this niche, at this stage is almost certainly wrestling with, and leading with something that addresses it.

Personalize to the account, not to the person's content. "You're a $20M firm in a niche where hiring is the bottleneck — here's the tool we built for exactly that" beats "loved your recent post!" every time it's ever been tried.

An AI model is genuinely good at this synthesis: given the firmographics, it can frame why this asset is relevant to this account in language that doesn't smell templated. In Hunter's pipeline, that framing is drafted by Claude Opus per account — and then a human approves the send.

Question 4 — What should the touch lead with?

Research isn't only about the prospect; it's about matching them to the right asset from your side. The rule Hunter enforces mechanically: the touch leads with a lead magnet that cani-loop has already verified converts under $5 cost-per-lead on live paid spend. The research question becomes "which proven asset fits this account's situation?" — a selection problem, not a creation problem. If you don't yet have a bank of proven assets, that's the prerequisite to fix before scaling outreach at all; the reasoning is laid out in what is signal-based outbound?

Question 5 — Is there any reason not to send?

The last research step is a veto pass. Wrong-fit on closer inspection, ambiguous title, a company in visible distress where the message would land tone-deaf, any hint of a prior opt-out — any of these kills the send. This is also where a human belongs in the loop: Hunter's Phase 1 puts every cold first-touch in front of a person with the account, the reason it was picked, and the exact copy. One click approves; one click suppresses. The agent does the minutes; the human does the judgment.

What does the AI actually automate?

Read the last row again before buying anything in this category. The tools that fail are the ones that automate the judgment and skip the research; the architecture that works automates the research and keeps the judgment. More on the failure patterns in 7 outbound automation mistakes that burn your domain.

FAQ

How long should prospect research take?

Done by hand, a competent job takes real minutes per account — which is exactly why most humans skip it and blast instead. Done by an agent with a live data layer, the fit check, role selection, and account context happen in seconds per prospect, every night. The point of AI here isn't deeper research than a human could do; it's that the research actually happens on every single account.

What's the difference between personalization and research?

Personalization is what shows up in the message; research is what decides whether the message should exist. Quoting a prospect's podcast episode is personalization. Confirming the company sits in your revenue band, the title is current, and the account hasn't already opted out — that's research, and it matters more.

Should I mention the prospect's recent LinkedIn post?

Almost never. AI made "loved your recent post" personalization free to fake, so readers discount it to zero. Personalize to the account's situation — size, niche, the problem your asset addresses — not to flattery. Relevance reads as respect; flattery reads as a template.

Can AI do all of this without a human?

It can execute all of it, but it shouldn't ship without review at first. The pattern that works is an agent doing the research and drafting every first touch, with a human approving each send until the agent has earned looser reins — the way Optimus Hunter holds every cold first-touch for one-click approval in Phase 1.