LinkedIn lead scoring: how FOUND decides who fits
Two checks before anyone is contacted, a named reason for everyone turned away, a score you can read, and three buttons that teach it your taste.
On this page 8 sections
Most people a LinkedIn signal surfaces will never buy from you. They sell the same thing you do, they're students, recruiters or agencies, or they're simply in the wrong job. A lead score is how an agent decides which of them are worth a message, and most scores are a number with no reason behind it.
This is how FOUND's works, taken from the product's own code, with a real run from our own account. It's also a checklist for any tool that scores leads for you: can it tell you why someone was kept, and why everyone else wasn't?
Why two checks
Reading someone's full profile is the scarcest thing an agent has. Each read comes out of your LinkedIn account's daily allowance, and an account that reads hundreds of profiles a day stops looking like a person. So FOUND checks everyone twice: first for nothing, on what it already knows, and only then, for the people who pass, with one read.
Check one: the headline, for nothing
When someone reacts to a post, their LinkedIn headline comes with the reaction, so the first check costs no read at all. It asks one question: could this person be the buyer?
- Your roles, as patterns. The job titles in your buyer become patterns that match the ways people write them: "VP Sales", "Vice President of Sales", "Head of Sales".
- Forty titles the patterns must never match. Sales and business development reps, account executives and managers, analysts, coordinators, assistants, interns and students, recruiters, specialists, consultants, and past roles such as "Former VP of Sales".
- A second look before a read. In broader mode, and for a senior-sounding headline the patterns missed, the scoring model reads the headline first, so the net widens without paying a read for everyone it catches.
Check two: one profile read, and named reasons
People who pass get one profile read: title, company, location, time in the role, the "about" section. Then come hard filters, and each one that drops somebody writes down why. A drop you can't see is a drop you can't correct.
- Open to work, unless you say otherwise: someone looking for a job isn't buying.
- Location, outside the countries or regions your buyer is in.
- Excluded words: your competitors, your own company, anything you listed.
- Your block list, by company name or by its website's domain.
- Company size or type, outside the bands you picked, or a public company when you didn't want one.
- A "job change" that isn't recent: more than 90 days in the role.
What isn't known never drops anyone. An unknown company size, or a start date LinkedIn didn't give, lets the person through: absence of evidence isn't evidence.
The decision
Then one decision by a scoring model built for exactly this kind of choice. It reads what you sell in one line, who buys it (roles, industries, company sizes, locations, company types), who is excluded, the people you've marked as good and poor fits, and the person: headline, title, company, location, "about", and how FOUND found them.
It answers two questions. Is this person a strong, plausible, weak or no fit for what you sell? And can they decide to buy, or only influence it? A weak fit is dropped in the default, high-precision mode. A decision the model isn't sure of, under 35%, drops one band. Everyone who stays becomes a lead with a line saying why.
Reading the score
- The number is how sure the decision is, from 0 to 1. It's how well they fit your buyer, not a forecast that they'll buy: timing is what the signal is for.
- The bar has a tick where the band changes. A score within a few hundredths of it is marked borderline, because two people either side of that line are the same person.
- Order, not a cut-off. The signal's strength and the fit together decide who is written to first. A lower-ranked lead is still contacted, just after the others.
- The reason travels with them. "Why they are here" and "Found by" sit on every row, and the first message is written from them.
A real run: two signals, one buyer
On 22 September we pointed two signals at FOUND's own buyer, heads of sales and revenue operations at software companies, and ran each once. One read the audience of a large sales-software company's LinkedIn posts. The other searched for people similar to a profile we picked as a model buyer.
| Stage | A competitor's audience | Lookalike search |
|---|---|---|
| People the signal surfaced | 200 | 30 |
| Passed the headline check | 41 | 29 |
| Worth a profile read | 13 | 24 |
| Kept as leads | 0 | 21 |
FOUND's own account, 22 Sep 2026, one run of each signal against the same buyer. From the run's own counts.
The competitor's audience was mostly the industry, not buyers: 66 of the 200 had the company's name in their own headline, its staff and partners. 126 had no buying title, 23 were consultants, service providers, freelancers or students, and 8 were company pages. The lookalike search asked for the buyer directly, and 21 of 30 fit. A score you can't trace would have hidden that difference. Here each drop has a name, so the signal is what you change.
Which signals to start with is in our guide to the seven families.
Teaching it your taste
Every lead has three buttons. More like this marks a perfect fit. Fewer like this marks a poor fit, and also stops the agent contacting that person and cancels what was queued. Not this person takes one lead off the list.
Your latest perfect and poor examples, up to three of each, go into every decision that agent makes from then on, so it learns what you mean by your buyer without you rewriting it. Teaching takes effect on the next scan.
If the scoring model can't be reached, nobody is rejected. The person waits and is scored again within seven days, from the profile already read, at no second read.
The better your buyer is described, the less teaching it needs: how to write it from your own website.
Questions
Is the fit score the chance someone will buy?
No. It's how sure the check is that the person fits your buyer. Whether this is a good week to write to them is what the signal that found them tells you.
Does scoring use my LinkedIn account?
The headline check costs nothing. Each person who passes it costs one profile read from your account's daily allowance, the same allowance your agent reads from, and scoring stops for the day when it's used up.
Which AI model does the scoring?
A scoring model built for making choices like this one, reached through our AI provider. Our privacy policy names it, with every other model FOUND uses.
Why do nearly all my leads rank the same?
Because they passed the same strict checks. That's why the screen leads with the number rather than the band: the score is what separates a clear fit from a borderline one.
How the model is named and what it sees: our privacy policy.