How scoring works

Every scored account gets a fit score for one product and one purpose: a number from 0 to 10, a fit band and a few typed reasons. This page explains how that score is made. What the fit criteria are, and how their versions mark a score out of date, is on Fit criteria and fit score.

Scoring is the bridge between your raw account database and the accounts you actually work. Done well, it turns “here are 20,000 accounts” into “here are the 300 to look at this week.”

Where scoring runs

Rank on the Accounts list scores any filtered list, or a sample of it, with no campaign. See Rank a list without a campaign.
Score N remaining on a campaign’s Review tab scores its accounts that have no score yet; Score remaining on the Campaigns list does it for every campaign of the project.
Give feedback & rescore on the Review tab rescores every account of a campaign with a new scoring brief.
Score in a campaign, on an account page’s Fit scores card, scores that one account against one campaign’s fit criteria. It uses one AI call, and adds the account to the campaign if it isn’t in it yet.

All of them use the same scoring, and every run that spends money (ranking, scoring a campaign) shows an estimate first and starts only after you confirm. A ranking run also has a budget cap.

The purpose decides the question

Scoring asks “how well does this account fit?”, and the purpose decides what fit means. For sales it’s “how well do they match our buyers”. For press it’s “how well does this outlet fit our story”. For recruiting it’s “how well does this person fit the role”. For partnership it’s “how strong a partner would this be”.

The 0–10 scale and the typed reasons stay the same for every purpose, so a 9 always means a very strong fit and a 3 a poor one. What changes is the axis being rated, and the reasons follow it: “covers our space editorially” for press, “complementary product” for partnership. See Outreach reasons and targets for every purpose.

What the model sees

Every input that goes into a score, in the order the model reads it. A blank section is left out.

About what you’re looking for, once per run:

Language to write the reasons in: the campaign’s language, else the product’s, else yours, else English.
Purpose: sales, partnership, press, influencer, investor, recruiting, customer expansion, win-back, event, research or renewal.
Fit criteria for that product and purpose: name, summary, industries, roles, pain points, company size.
Product name and description (up to 2,000 characters). Editing the description under Settings → Products changes the next scores without regenerating the fit criteria.
The campaign’s scoring brief, when you score inside a campaign that has one (up to 2,000 characters).
Good-fit and bad-fit example websites set on the product, with what each site says (up to 2,000 characters per site). After the fit criteria, these do more for scoring quality than anything else.
Calibration examples: your latest Shortlist and Not a fit decisions for that product and purpose, once there are enough of them (see the feedback loop).

About each account, repeated for every account in the batch: name, location, business type and specialization, website, rating, language, email and phone, the account’s notes (first 500 characters), any extra data kept on it, and the text of its website (first 3,000 characters, marked as untrusted content so a page can’t give the model instructions).

The notes go to the model. Use them for short context worth weighing (“Referred by an existing client”, “Opened a second shop in 2025”), not for private remarks.

What it produces

A fit score from 0 to 10.
A fit band: Strong fit (8–10), Good fit (6–8), Weak fit (4–6), Poor fit (below 4).
Two to four reasons, each typed so you can read the score at a glance:

Type	Shown as	Example
Positive	green check	”FM rock station, matches the industries in the fit criteria.”
Negative	red cross	”Single-location operator, below the size range.”
Neutral	amber mark	”Spanish-language content.”

Strong fits lead with positives, poor fits with negatives, and when both pull the AI includes at least one of each.

An account the model couldn’t judge, or that hasn’t been scored yet, reads Not scored yet in grey. It is never shown as a poor fit or a 0.

What scoring doesn’t see

Your conversations with the account. Past messages feed message drafting, not scoring.
Other projects. Scoring only reads the active project’s product and account.
Custom field values. Use them to filter what you rank, not to sway the score.
Paid data providers you haven’t brought in.

One score, read everywhere

A fit score belongs to an account, a product and a purpose. The same score shows in the Fit column on Accounts, the account page’s Fit scores card, every campaign of that product and purpose, and Today’s Best accounts to work next. So:

An account in three sales campaigns of a product is scored once; adding it to a fourth costs nothing.
The same account in a sales campaign and a press campaign of that product gets two scores, because the questions differ.
Changing a campaign’s purpose leaves the old scores in place (the campaign’s Overview flags them); rescore to get scores for the new purpose.

A campaign with its own scoring brief keeps its own scores. Rescoring it never changes another campaign, the Accounts list or Today. Campaigns without a brief share the scores, so rescoring one updates what the others show; the Give feedback & rescore dialog says how many other campaigns that affects before you start. More on a campaign’s own scoring brief.

A rescore keeps every Shortlist, Defer and Not a fit decision, every score set by hand and every conversation. Only the AI’s score and reasons change. Set score by hand (in a review row’s menu) changes the score for that campaign only; the shared score stays as it was.

Give feedback & rescore

When you see the AI missing the same thing across many scores (“chains score too high, we want owner-operated shops”), click Give feedback & rescore on the campaign’s Review tab:

The dialog opens with the campaign’s current scoring brief, so you can revise it.
What you type is saved as the campaign’s scoring brief, even if the rescore doesn’t start.
After the cost estimate and your confirmation, every account of the campaign is scored again with the brief.

The brief stays on the campaign for every later scoring run until you change or clear it. Giving a campaign a brief gives it its own scores (above).

The feedback loop — your reviews calibrate the model

The best way to improve scoring isn’t rewording the fit criteria: it’s deciding accounts one at a time, Shortlist or Not a fit, in Review. Once a product and purpose have about ten such decisions across their campaigns, LeadHunter adds the latest ones to the scoring prompt as examples of “what good looks like here” and “what to skip even when it looks right”. Below ten they’re held back: a small sample is noisier than none.

Only single decisions count. Bulk decisions (a whole band, the top N, everything matching) are recorded as bulk and never teach the scoring, because shortlisting 500 accounts at once says nothing about each of them. Defer isn’t a fit judgement, so it never counts either.

The examples stay within one product and purpose: press decisions never teach sales scoring. An account being rescored is never shown its own past decision.

That’s why the first campaign on a new product and purpose is the one to review carefully, one account at a time.

What if the account has no website?

Accounts that arrive with a name and a city only can still be scored, but the model leans on the few fields it has, so scores bunch in the middle and the reasons are vague. Before scoring, LeadHunter looks for missing websites and reads them; you can also look an account up again with Deep research (Accounts → Add account → From URL → Deep research). Rescore once the content is there.

After scoring: work the bands

The Review tab’s Fit score bands (8–10 Strong fit, 6–8 Good fit, 4–6 Weak fit, 0–4 Poor fit, Not scored yet) show the whole campaign at once. Shortlist the top band in bulk, mark the bottom one Not a fit, and go through the middle by hand: see Review thousands of accounts. Remove fits under 5 on the Campaign accounts panel of the campaign’s Overview takes accounts scored below 5 out of the campaign while keeping shortlisted and contacted accounts; see Campaign → Pruning the low-fit tail.

A healthy campaign is usually two piles, strong fits and poor fits, with a modest middle. A flat or all-middle shape usually means the fit criteria or the example websites need work.

When to rescore

Rescore when:

The fit criteria change. Scores made with an older version show an old criteria badge, and the Review tab offers Rescore.
The example websites or the product description change.
Accounts gained data: websites found, descriptions or notes filled in.
You’ve just passed about ten single decisions for the product and purpose: new scores benefit from the examples, old ones only when rescored.

A rescore overwrites the score in place; the previous score isn’t kept.

Cost and speed

Scoring runs on a fast, low-cost AI model, in batches of about ten accounts. Each account costs a fraction of a cent, and a few hundred accounts take 5 to 15 minutes. The estimate before each run shows the accounts, the batches, the model and the expected cost; what each run actually spent shows on Runs, the campaign’s costs and Insights → Usage (see Usage and costs).

If a campaign scoring run fails partway, the scores it made are kept; Score N remaining scores only what’s missing.