Ground truth beats inferred intent

Most intent data sold to GTM teams is inferred. A model watches content consumption, assigns a topic score, and tells you an account is "in market". The score arrives with no artifact behind it. You are asked to act on a number you cannot check.

We think that is the wrong foundation for outbound. Here is the argument.

You cannot audit a topic score

Ask a vendor of inferred intent one question: show me the evidence behind this score. You will not get it. The inputs are consumption data you cannot see. The model is proprietary. The output is a number with no source attached. Lookalike models have the same shape: an account resembles accounts that bought, so it gets a score. Resemblance is not evidence.

This matters in practice. When a rep burns a morning on a "surging" account that was never in market, nobody can say why the score was wrong. There is nothing to inspect, so there is nothing to learn. The feedback loop is closed to you.

A job posting is a checkable artifact

A public job posting is the opposite kind of signal. It is a document the company chose to publish. It names the role, the location, and the problems the hire will own. The pain is written in the text, by the account itself, in its own words.

It is also committed budget. A company that opens a req has decided to spend money on that problem. That is a stronger statement than any pattern of content consumption.

And because it is public, anyone can verify it. Every result Reqbeat returns is a real posting tagged with the board it was detected on. Each row names its source in the boards field, never an inference. When a webhook fires, the payload carries the URL of the posting. Your rep can open it. Your prospect can open it too, which means the signal survives being quoted in an email.

Freshness and dedup matter more than volume

Vendors of inferred intent compete on volume: more topics, more accounts, more signals per week. We think two other properties decide whether a signal is worth acting on.

Freshness. A hiring signal decays. The team that reaches the account while the req is new has a different conversation than the team that arrives after the hire. The Reqbeat corpus refreshes every 3 hours across 7 regions, and a webhook fires when a new matching posting is detected, typically hours after it goes live.

Deduplication. Raw job data is noisy. The same req gets posted, re-posted and syndicated across boards. Counted naively, one opening looks like five, and your "hiring velocity" number is fiction. Reqbeat collapses repeat and re-posted advertisements to one row per company, job title and country, and keeps the most complete row. The exact partition rule is published on the method page. A deduplicated count you can defend beats a large count you cannot.

Unknown is never no

Every data source has edges, including ours. Some companies have no ATS or board coverage we can observe. An honest system says so instead of guessing. When Reqbeat has no coverage for a company, the answer is coverage_status: "no_ats_signal", and the payload says so explicitly.

That status means "unknown". It never means "not hiring". Do not score the absence of coverage as a negative signal, and do not let a workflow disqualify an account because a watcher could not see it. Inferred systems rarely make this distinction. A low score and a blind spot look identical, and both get treated as "no".

The standard to hold any signal to

Before you route a signal to a rep, ask three questions.

• Can I open the evidence? A link to an artifact, not a score.
• How old is it, and how do I know? A detection time, not a weekly batch.
• What does silence mean? "Unknown" and "no" must be separate answers.

Inferred intent fails all three. A watched job posting passes all three. That is the whole bet behind Reqbeat.

Our sample, dedup rule and known biases are public on the method page. What we hold and what we never publish is on data ethics. If you want to test the claim instead of reading about it, the quickstart gets a first signal in about 5 minutes, and the launch story is in our first post.