There is a campaign in our own account right now with a status most tools would hide from you: rate_limited. It is our own recruitment campaign, it holds 1,892 prospects, and it is the best argument for AI lead scoring we own. Because the constraint in LinkedIn outbound is not how many prospects you can find. It is how many messages the platform lets you send. Around 150 connection requests per week, per account. That is the budget, and every one you spend on the wrong person is gone.
The scoring model we actually use
Every prospect gets scored 0 to 100 before entering a campaign, across three dimensions:
- Fit, worth 60 points. Does the title own the buying decision (30)? Is the company the right type and size (20)? Right geography (10)? A "Head of Talent" at a 40-person agency scores differently from the same title at a 5,000-person enterprise with a fully staffed internal function, and the difference is whether anyone ever buys.
- Timing, worth 30 points. Visible signals that now is the moment: hiring posts, a new role in the last 90 days, growth language in the company page, founder doing their own outreach.
- Reachability, worth 10 points. Active on LinkedIn in the last month, or a ghost profile that will never see the request.
Then the verdicts: pursue, maybe, skip. The skips are non-negotiable: competitors, job seekers, wrong country, companies too large to buy the service. On most raw lists, between a quarter and a third of prospects get cut.
What the cut looks like on a real list
A recent 2,125-prospect construction-sector list came back from scoring like this: roughly 1,400 pursue, 450 maybe, and 275 skipped outright. The skips included quantity surveyors looking for jobs rather than hiring them, sole traders with no commercial function, and a cluster of profiles that had not posted or engaged in over a year.
That last group is the silent killer. Sending requests to dormant profiles does not just waste the request. It drags your acceptance rate down, and acceptance rate is one of the signals LinkedIn appears to weigh when deciding whether your account is behaving like a human. Low acceptance invites throttling. Throttling kills the whole campaign, not just the bad sends.
The evidence: same system, different lists
We run the same AI, the same messaging engine, the same approval flow across every campaign. The only meaningful variable is the list. Recent results across two niches:
| Campaign | List discipline | Acceptance | Reply rate |
|---|---|---|---|
| Quantity surveyors (construction, UK) | Tightly scored, narrow niche | 57% | 29% |
| Recruitment owners (broad UK list) | Broader, looser scoring | 30% | 19% |
Same engine. Nearly double the acceptance. The narrow, ruthlessly scored list outperforms the broad one on every metric that matters, and it does it while spending fewer connection requests.
Why AI changes this, honestly stated
Scoring lists is not new. Sales teams have always known they should do it, and almost none do, because scoring 2,000 prospects by hand is two days of soul-destroying work that never survives contact with a busy week. The AI does it in minutes, explains every score in plain English, and never gets bored on prospect 1,400.
But the judgement still has to come from somewhere. The scoring model is built in a conversation with the founder: who actually buys, who wastes your time, what a timing signal looks like in your market. AI without that conversation just automates guessing.
Do it yourself, free
We packaged our scoring approach as a free Claude skill: paste any lead list and get ranked verdicts with reasons. Download the ICP Scorer, no catch. If you want the scored list worked automatically every day after that, with every message approved by you, that is AI Sales.