Score Every Account, Contact Only a Beachhead

Scoring one company from a domain cost 1.99 records. That's the average over 200 random domains in a test on 27 October 2026: 398 billed records, two calls a company, and 199 of the 200 came back with a headcount and a size band.

And that's small enough to change an argument GTM people keep having.

This piece prices scoring a whole market and shows where the limit moves next. We have no evidence that wider scoring improves results, so the claim stays at cheap enough to test.

Two newsletters, two answers, no price

Two GTM newsletters in August gave opposite advice. One said to kill the segment and score every person. The other said to pick one beachhead and stay on it until you hold most of that market. I only saw the headlines and short summaries, so this describes the positions and not the reasoning behind them.

Neither answer, as far as I could see, puts a price on the scoring it depends on. That premise has a number now, and it's small.

My answer borrows from both. Score every account, then contact only a beachhead. (A company becomes an account once it sits on your target account list.)

What it costs to score a company

From a domain, scoring takes two calls. The first turns the domain into a LinkedIn company page, and the second pulls the company record. BlitzAPI bills one record for each call that finds something. A call that finds nothing bills nothing. The test drew 200 domains at random from 4,789 in the enrichment archives, on the BlitzAPI plan Lead Line Partners holds today, which is a different plan from Start. (Notes on the data API sit on the BlitzAPI tool page.)

The two calls averaged about 0.56 seconds a company in latency. The whole run took about 3.6 minutes, with a 0.25 second pause between calls. The fields came back full: size band and headcount for 199 companies, industry for 198, revenue range for 192 (96%). Of the 199, 187 are headquartered in the US.

As of October 2026, the BlitzAPI pricing page lists Start at $99 a month with a fair use policy of 200,000 records, and Scale at $499 with 5,000,000. Divide by 1.99 and Start covers about 100,000 companies a month from a domain list, and the $99 Start plan looks very interesting to us and is under consideration. Already hold the LinkedIn URLs? Only the second call applies, so you get 200,000 a month instead. Scale reaches about 2.5 million.

The vendor docs I read list one record on success for both calls, on every plan. But the test measured that on the current plan only, and Start billing hasn't been observed. At the 10 requests a second Start allows on each endpoint, you'd spend about three hours of calls per endpoint on 100,000 lookups. That figure is calculated from the published limit, not measured.

The limit I'd put first: you can't tell from this test how often a domain lands on the right company. A crude name check flagged 8 of the 199 matches. One is clearly wrong. The other seven are unchecked. And scoring the wrong company is worse than skipping it, because the rank looks just as confident either way.

What it costs to meet the people

The people cost more. One default page of ten people from the employee finder averaged 9.55 records across 20 of those companies, 191 in all. That's 4.8 times the 1.99 for the company score. It's only page one.

And full rosters are the real bill. Logged pulls in the archives averaged about 1,590 records an account (23,853 records over 15 accounts), a floor taken from an inventory log and not re-derived. At that rate, a Start month covers only about 125 full rosters.

Try an illustrative split: score 20,000 companies (39,800 records), then pull one roster page for your top 1,000 (9,550 records). That's 49,350 records, about a quarter of the month, and you'd still have the other three quarters. The gap between the two costs is why the answer has two halves.

The decision layer costs cents

A record alone doesn't rank anything: something has to read the fields and say how well the company fits. Jev, a decision model priced at $0.042 per million input tokens with output free, shows what a simple yes/no gate costs. The workloads below come from the Jev write-up.

And the last row is the ICP-fit row times ten, since 100,000 companies is ten times 10,000 leads. Even at the Haiku price, ranking 100,000 companies runs $160. Cost isn't the blocker.

But Jev has been in early access since 15 September 2026, with no published SLA that I could find, and its confidence runs above its accuracy: 80.1% right against 90.1% stated, in the independent test our write-up cites. The maker's own price page wasn't reachable, so the rate comes from a third-party guide that restates it, and a second doc confirms output is free, both as of 27 October.

If you test a model like this, have it rank accounts and keep a person on every send. The write-up on this site says to wait before letting it gate anything live, and that advice stands.

The fit and signal axes in the account scoring template still apply, only now you can run them on the whole market. The logic should stay readable enough to open when a ranking looks wrong, the point of scoring buying intent without a black box. And a person decides who gets contacted.

What one killed segment taught us about screening

A data-infrastructure client's team proposed a segment of young companies, and we tested it in one run instead of debating it. The run enumerated 467 companies across several programs. About 237 stayed eligible after a recency cutoff, and we gated a sample of 60 of them. Of the 60, 38 had a findable company record and 22 did not (21 invisible to the data tool, 1 acquired). Eight of the 38 sat in the client's employee band, and 30 sat under it. Four of the eight cleared funding. One had an open relevant role, and that one failed a deeper check. All 60 failed.

This is one run for one client, so read it as an example. The sample of 60 isn't verifiably random. The funding and open-role counts come from the run notes and weren't recounted. A follow-up piece covers the case in full.

A headcount screen of all 467 would have cost about 930 records at 1.99 each, 0.5% of a Start month. That is a calculation; nobody ran it. And I'm saying headcount on purpose, because funding wasn't a field the company test returned, so I can't price that gate.

In that run, 21 of the 60 were invisible to the data tool used, and the 200-domain test drew from established target lists. But whether 1.99 holds for very young companies is untested.

Where the limit moves

Once scoring is cheap, three things set how many accounts you contact. A person has to review them. Your mailboxes and domains have to carry the sends, and your sellers have to answer what comes back. And someone has to confirm that each person still works where the data says, because data decay tends to hit people-level records first, though we haven't measured it.

But we haven't measured how long a person takes to review an account or how many sends a setup carries, so I won't print a number. Yours is whatever your team can read and send before quality slips. Find it before you size the beachhead.

In Trellis, the workspace a client works from, each account we propose arrives on Radar with the reason it was picked, the tier it earned and the dated signals behind both. You approve, hold or decline. And nothing arrives pre-approved, while a decline carries a reason that goes into our next search. Before anyone on an approved account is written to, we check on our side that they still work there. The AI in Trellis can't send, approve or dial.

What we would do with it

Score every company you could plausibly sell to. And sort the strong ones into tiers on fit and signals, the way ABM tiers behave once strategist hours stop being the limit. Lead Line Partners coined depth at width for that state, where every fit account gets research depth that used to go to a few dozen. Then contact the top of one tier, a beachhead sized to what your reviewers can read and your mailboxes can send. And widen it as that capacity allows.

Three results would change my mind. If the seven unchecked matches turn out wrong, the cheap score is cheap noise. If young companies keep coming back empty, as 21 of the 60 did in that run, the score has a hole where early-stage accounts live. And if Start bills a lookup differently than the docs say, every count above moves.

If you want this run on your own domain list, we start with a 200-company sample and show you the records billed and the matches we couldn't confirm before you decide anything.