Mason

Original research · 67 B2B sellers · 320 hypotheses

We tested 67 pitched signals for a buying signal benchmark.
One held as worded.

29 more held once we narrowed them on the evidence. 16 did not hold. Here is what changed.

Read the findings Get the checklist
Research run Sep to Nov 2026By Shreyan Rajesh

Mason / Research · Published 9 November 2026

1held as first worded, of 46 runs whose record gives a verdict
29more held once we narrowed them on the evidence
16did not hold, and another signal carried the page

We researched accounts for 67 B2B sellers, each starting from a buying signal we had pitched in one line. Of the 46 runs where the record says what happened, 1 held as worded, 29 held once we narrowed them on the evidence, and 16 did not hold. The pitch is a hypothesis. The records decide its final shape.

THE QUESTION

Does the signal you pitch
survive the records?

A signal is easy to say in one line. The test is whether real job posts, filings and pages say the same thing, at the right companies, inside the window.

HOW WE CHECKED

Hypotheses first, then the records.
Then the run's own verdict.

For each seller we wrote hypotheses before searching, read real posts and records in full, and kept only companies with dated evidence in their own words. Then we counted what each run's write-up says happened to the pitched signal.

67 sellers320 hypotheses92,315 search results read601 companies shown

01 / The scorecard

The signal you pitch is a first draft

What we counted

Every run started from one line we had pitched to the seller, by email or on a call: "companies hiring X", "firms that just did Y". Each run's write-up says, in its opening, what happened to that line once the records were read.

46 of the 67 write-ups give that verdict. The other 21 open with the result instead, so we left them out rather than guess.

What happened to the pitched line

1 held as first worded.

29 more held once we narrowed them on the evidence: a different seat, the duty in the post body, a second dated record, or a filter.

16 did not hold as pitched, and another signal from the same research carried the page.

02 / How the 29 changed

Five moves, again and again

The narrowing was not random. The same five moves turn up across sectors, and many runs needed more than one.

15 of 29

moved from the job title to the duty written in the post

and
12 of 29

moved to a different seat or a different buyer

How the 29 narrowed signals changed. A run can show more than one move.
The moveRunsAn example, from the run's own write-up
Read the duty in the post, not the title15 of 29A consolidation finance seat passed only when the post gave the hire the group consolidation and named the accounting software. It was found by a phrase in the body, never by title.
A different seat, or a different buyer12 of 29The title the pitch named belonged to the wrong kind of employer in 25 of 25 posts. At the buyer, the same work sat under other titles.
A second dated record, dated by the employer10 of 29Three posts that LinkedIn showed as a week old were 79 to 202 days old on the employer's own board.
Drop companies that already run a tool6 of 29About half of the lease seats read in full already named the software in use. In another run, a one-query check for incumbent tools removed 5 of the first 10.
A size or sector filter4 of 29The posts said the right thing, but mostly at companies over 2,000 staff. At mid-market size the work goes to a managed service provider.

03 / Why 16 did not hold

Four reasons a signal fails

"Did not hold" means the pitched line produced at most one company worth showing. In every case the same research found something else that did.

7 of 16

The buyer never writes it.

0 of 72 full job posts at in-shape contractors used the pain words from the pitch. In another run, 0 of 5 at the pilot.

4 of 16

It fires, but it is not a moment.

A public quality-score file listed the whole market for free. But a low score is not a moment, and the companies rarely dated anything in public.

3 of 16

The wrong employer posts it.

Of 58 employers posting the roles, 37 were vendors or consultancies and 8 were staffing firms.

2 of 16

Too rare in the window.

The right buyer, but 1 live post in 60 days across 38 searches.

04 / The one that held

It was a regulator's record, not a job ad

What held

The only pitched line that held as worded was built on a regulator's own published letters and recalls, for a seller of quality software to regulated manufacturers. The record dated the event and said what went wrong, in the regulator's words.

What still had to be checked

Whether the company already ran a competitor's system.

Four strong companies named one in their own records, once in the regulator's own letter.

One case is a story, not a rule. It is why we read public records before job ads.

05 / Volume

What 67 runs read

Counted from each run's own logs. Free reads of job boards and government files are mostly not logged, so these numbers undercount.

Search

92,315

search results read

Jobs

23,308

job-search listings returned, and 2,262 posts read in full through a paid door

Pages

1,158

pages rendered, and 1,290 people profiles read

Output

601

companies shown, from 320 hypotheses written before any search

67 sellers in 9 sector groups (our grouping; small groups are folded into "Other sectors", so no row can point to one seller). Where a signal names a country, it names the US in at least 23 runs and the UK in at least 17. A companion study checked 89 real purchases against each buyer's own year before: what actually comes before a B2B purchase.

06 / Honest notes

What we got wrong along the way

  • We trusted aggregator dates. Posts that looked a week old were months old on the employer's own board. Date from the source.
  • We searched by title. In 15 of the 29 narrowed runs, the signal lived in the duties, under titles we would not have searched.
  • We missed the recruiters. In one run, about half the matching LinkedIn rows were recruiters hiding the employer.
  • We waited. In one run, 4 of 10 picks closed or expired between two checks two days apart. Re-read every record on the day you use it.

Limits

  • The verdicts are our reading of our own write-ups. Each run's sentence is kept beside its label so the count can be checked.
  • 21 write-ups give no verdict in their opening. We left them out of the 46 rather than guess.
  • These are research results, not sales. A company shown is a company worth a call, not a deal.
  • The sellers are anonymous, and the sector groups are ours.

07 / Takeaway

The signal pilot checklist

Run this before you buy a signal, or before you build a campaign on one. It is the order our 67 runs ended up following. It works the same for an agency or a market-entry consultant testing a client's signal.

Download the 67-run data (CSV)

Seven steps
  1. 01

    Write the signal as one sentence. Then write three to six hypotheses for where it would show up.

  2. 02

    Read five to ten real posts or records in full before you search at scale.

  3. 03

    Search for the duty the buyer would write in the post, not only the title.

  4. 04

    Check who is posting. Drop vendors, consultancies, staffing firms and recruiters.

  5. 05

    Date every record from the employer's own board or the source document.

  6. 06

    Ask for a second dated record from the same company, in its own words.

  7. 07

    Search each company's records for your competitors' names, and drop the ones already using one.

Want your signal tested on real records?

We test the signal on the records before anyone builds a list on it.

Book a call See the case studies