meirlabs
Meir RosenscheinHow Do You Find the Right Person to Contact at a Company?
Stop hunting for more names and start ruling on the ones you have. For each contact ask one question: does the title say this person owns or influences the decision you are pitching? Three answers only, right_contact, wrong_contact, unclear. Two rules make it work: judge from the title alone, never from your hunch, and keep unclear rare, for a bare "Manager" that names no function at all.Launch offer: Early clients get 50% off their first build, so your real cost is about half these figures. Book a free AI plan to lock it in.
Most owners treat this as a sourcing problem. Buy a database, export 500 rows, start sending. Wrong frame. By the time you are actually stuck you already have a name for each target company. What you do not have is a decision about whether that name deserves the send, and nothing in the export tells you that the "Operations Manager" in row 41 has no say whatsoever over the thing you sell.
Why doesn't a bigger list fix this?
Because the list you already have is thinner than it looks. On a recent engagement, a client running outbound for a staffing business handed me 25 accounting leads they had researched by hand. Two of the 25 had a verified email address. Not two bad ones, two verified ones, out of twenty-five. Bolting another 500 rows onto that does not raise the share of names worth writing to, it just gives you more rows in the same condition. And every wrong send costs twice: a slot out of a daily sending budget you cannot safely exceed, and a dent in the reply-rate signal your mailbox reputation runs on. The missing work is not more names. It is a verdict per name.
What are the only three verdicts you need?
One question per contact: does the title say this person owns or influences the decision you are pitching? Three answers, no fourth. Written out for a staffing pitch, where the decision is "who signs off on hiring for the roles this company is advertising":
- right_contact. They plausibly own or influence hiring for those roles. A founder. A CEO or COO at a company small enough that they still sign off personally. The head or director of the function doing the hiring. A People, Talent, or Recruiting lead.
- wrong_contact. Their function is clearly unrelated to buying this for those roles. An individual-contributor engineer, an analyst, a support rep, anything with no people scope and no exec scope in the title.
- unclear. The title itself is genuinely ambiguous. A bare "Manager". A bare "Director" with no function named. That is the entire list of cases.
Three buckets, no fourth. If unclear is more than a small minority of a batch, the rubric is being avoided, not applied.
Why judge only from the title?
Because the alternative is judging from what you hope. The rule is literal: do not guess seniority that is not in the title. An "Operations Manager" at a 2,000-person company is not a buyer just because "operations" sounds adjacent to your pitch. Whoever built the list, you, an assistant, an intern, already had a hunch about why that name got picked. Keep the hunch as context and never let it cast the vote. In the automated version I built, that is enforced in the prompt: the researcher's own note goes in as background, and the model is explicitly forbidden from letting it decide.
Why is "unclear" the verdict that costs you the most?
Because it feels responsible and behaves like a yes. An unclear pile never gets re-reviewed, it gets emailed on the last day of the sprint. So the second rule is a rule against hedging: commit to the call, do not soften an obvious one to unclear. "VP of Engineering" is a verdict, not a mystery. Reserve unclear for the narrow case where the title names no function at all. If a third of your list comes back unclear, the rubric is not being applied, it is being avoided.
What do you do with a wrong contact?
You do not drop the company, you re-aim the email. And here is the guardrail worth keeping when you automate it: the better contact has to come from information already in your own target file. Nothing invents a new person at run time. The moment a system is allowed to produce names, it will produce plausible ones, and a plausible name with a guessed email address is exactly how a sending domain gets burned. Keep address verification as its own separate step for the same reason. Picking the right person and confirming their address are two different jobs, and a tool that blurs them hides the failures of both.
Is it worth automating?
Depends how many companies. By hand this is a couple of minutes of reading per name, fine at 25 companies and miserable at 500. On the staffing engagement I wired the rubric into a pipeline: for one company it scrapes the homepage, the about page, and up to three likely careers URLs to see what that company is hiring for right now, then asks the model for a verdict constrained to exactly those three values, with a written reason attached. It costs $0.18 to $0.40 per company in scraping and model calls, takes 70 seconds to about four minutes, and drafts the email without ever sending it. Be honest about the limits too. There is no measured hit rate, nothing logs whether a verdict was later proven right, and the head-to-head against that client's own human researchers has not been run yet. What the rubric buys is not proven accuracy. It is that every name on the list now carries a stated, reviewable reason for being emailed.
A test you can run on your list today: read the title, say the verdict out loud in one breath. If you need a second sentence to justify a right_contact, it is a wrong_contact. Automating that judgment is one well-scoped workflow, the same shape as any other AI adoption project: pick one decision you already make by hand, write the rule down, and hand over only that.