Guides

How to Measure Hiring Signal Accuracy: Precision, False Positives and Signal-to-Placement Rate

Signal-led BD is mainstream, but almost no desk measures whether its signals work. A simple scorecard for hiring signal accuracy: precision, lead time, reply rate by signal type, signal-to-meeting, signal-to-placement and the false-positive causes, plus a 30-day routine.

TB Team Boilr

· November 9, 2026 · 14 min read

Abstract dark liquid-metal folds with a green sheen, representing signal quality being refined

TL;DR

Most recruitment desks that run signal-led BD can tell you how many signals they received last month. Very few can tell you how many of those signals were real, how many turned into a brief, or which signal types actually pay. That gap matters, because the raw material is noisy: depending on the study, roughly 18% to 34% of online job postings are ghost listings [1] [3], and ghost listings are exactly the kind of "signal" that looks perfect in a feed. This guide gives you a seven-metric scorecard for hiring signal accuracy (precision, lead time, reply rate by signal type, signal-to-meeting, signal-to-brief, signal-to-placement and false-positive rate), a taxonomy of why signals fail, and a 30-day routine to get your first honest numbers. boilr supports the loop by delivering each signal as a sourced Task you verify, send or skip, scored against your ICP.

Why Almost Nobody Measures Their Signals

Signal-led BD has moved from "clever tactic" to default advice. The measurement habit has not followed. Most desks track activity (calls, emails, signals seen) and outcomes (billings), and nothing in between. The middle is where you find out whether a signal was any good.

Volume feels like progress - a feed with 200 signals a week looks productive. It tells you nothing about how many were worth a call.
Outcomes arrive late - a placement can land three to six months after the original signal, so the link is forgotten unless you wrote it down on day one.
CRMs log the lead, not the trigger - Bullhorn or RecruiterFlow will tell you the company and the contact, but rarely which signal started the conversation.
Everyone assumes the source is clean - yet the underlying data is full of noise. Greenhouse reported that 18 to 22% of jobs listed on its platform were fake [1], and almost 70% of companies on the platform had posted at least one ghost job [2].
Failure is quiet - a bad signal costs you one wasted call and a slightly worse opinion of the feed. Nobody files a ticket.

The fix is not a data science project. It is a one-page scorecard, a column in your tracker and ten minutes a week. If you want the commercial side (cost per placement, GP per desk), start with our guide to recruitment BD ROI tracking. This article covers the layer underneath it: is the signal itself any good?

Define the Terms Before You Count Anything

"Accuracy" means different things to different people on a desk. Agree on these definitions first, or two consultants will report two different numbers for the same signal.

Term	Definition for a recruitment desk	Example
Signal	A dated, sourced event suggesting a company may need to hire (funding, exec move, expansion, posting spike)	Series B announced on 3 November
Actioned signal	A signal a consultant actually contacted someone about	Call or email sent to the new VP Engineering
True positive	An actioned signal where the client confirms a real, open hiring need you could work	"Yes, we are adding three engineers, send me profiles"
False positive	An actioned signal where no workable need existed	Role was a ghost listing, or filled internally
Brief	A live vacancy the client has shared with you to work, with a named owner	Job spec received, terms discussed

The Congressional Research Service describes ghost jobs as postings for positions that do not exist or that employers do not plan to fill immediately [4]. Borrow that second clause: a role that exists but will not be filled for nine months is a false positive for a desk that needs to bill this quarter.

The Seven-Metric Signal Scorecard

You do not need seven dashboards. You need seven numbers per signal type, refreshed monthly. Each one answers a different question.

1. Precision: did the signal lead to a real brief?

Formula: true positives divided by actioned signals. If you contacted 40 companies off funding signals and 6 confirmed a real, workable need, precision for funding signals is 15%. Use "confirmed real need", not "booked a meeting", as the numerator, because a meeting can be polite curiosity.

2. Lead time: how early was the signal?

Formula: days between the signal date and the date the vacancy appears publicly (or the client tells you about it). A signal with high precision but zero lead time is just a slower job board. Log the gap when a brief surfaces, even if you only know it roughly.

3. Reply rate by signal type

Formula: unique replies divided by messages sent, split by signal type. Do not compare your number to a headline benchmark without checking the definition: Belkins measured a 0.45% average reply rate across more than 7.5 million cold emails sent in 2025 [5], whilst Instantly reports 3.43% across its own dataset [6]. The two differ in denominator, sample and sender mix. Your real benchmark is your own cold baseline, measured the same way.

4. Signal-to-meeting rate

Formula: meetings or substantive calls divided by actioned signals. This is the first number that sales managers care about, and it is a useful early read when precision data is still thin.

5. Signal-to-brief rate

Formula: briefs received divided by actioned signals. It sits between precision and placement, and it exposes whether your problem is the signal (low precision) or your conversation (good precision, few briefs accepted).

6. Signal-to-placement rate

Formula: placements divided by actioned signals, tracked in cohorts by signal month. Because the cycle is long, report it on a rolling six-month cohort rather than month by month, otherwise recent signals will always look like failures.

7. False-positive rate (with a reason code)

Formula: false positives divided by actioned signals, but the number matters less than the reason. Every false positive gets exactly one reason code (see the next section). That is what turns a complaint into a fix.

Metric	Question it answers	Where to log it	Review cadence
Precision	Is this signal type real?	Outcome field on the signal	Monthly
Lead time	Are we early or just on time?	Signal date vs brief date	Monthly
Reply rate	Does the message land for this trigger?	Email or sequence tool	Monthly
Signal-to-meeting	Does it open doors?	Calendar or CRM activity	Monthly
Signal-to-brief	Does it create work?	CRM vacancy record	Monthly
Signal-to-placement	Does it pay?	Placement record with source	Quarterly cohort
False-positive rate	Why does it fail?	Reason code on the signal	Weekly sample

The Eight Reasons Hiring Signals Turn Out False

Once you tag every miss with one reason, patterns appear within a few weeks. These are the causes that come up most often on recruitment desks, with the usual fix.

Reason code	What it looks like	Fix
Ghost job	Posting is live but nobody is hiring against it	Require a second corroborating signal; see ghost jobs and fake vacancies
Repost or evergreen role	Same role reappears every few weeks	Treat repeat postings as a different signal type with its own precision
Stale signal	Event happened weeks ago, role already filled or agency engaged	Cap signal age per type; see the Signal SLA
ICP mismatch	Real need, wrong sector, size or geography for your desk	Tighten ICP filters and scoring before the signal reaches a consultant
Wrong contact	Right company, but the person has no hiring authority	Map the decision-maker per signal type (e.g. new CTO for tech hires)
Internal fill	The in-house TA team handles it	Record it; consider pitching specialist or hard-to-fill roles only
Locked supplier	PSL or exclusive agency already covers the role	Log the PSL status on the company so it stops scoring high
Echo signal	Three tools surface the same event as three signals	De-duplicate by company and event before counting; see signal fatigue

Note the distinction: some of these (ghost job, echo, stale) are failures of the signal. Others (ICP mismatch, wrong contact, locked supplier) are failures of targeting. Fixing the wrong layer wastes weeks.

Signal Problem or Conversation Problem?

A low signal-to-placement rate can have very different causes. Read the metrics together before you blame the feed.

Pattern	Likely diagnosis	First action
Low precision, low reply rate	Weak signal type	Downgrade or drop it, or only use it stacked with another
High precision, low reply rate	Real need, poor opener or wrong contact	Rewrite the first line around the trigger; check the contact
High reply rate, low precision	Curiosity replies, no real need	Qualify harder on the first call; ask about open roles directly
High precision, low signal-to-brief	Conversation or terms problem	Review call notes, fee model, speed to first CVs
High signal-to-brief, low placement	Delivery problem	Look at candidate supply and time to shortlist, not at BD
Lead time near zero	Signal is a lagging indicator	Move earlier in the chain (funding, exec moves) or accept it as a speed play

A Worked Example (Illustrative Numbers)

The figures below are invented to show the arithmetic, not benchmarks. Imagine a 360 desk that actioned signals for one month and logged outcomes:

Signal type	Actioned	Real need	Precision	Briefs
Funding round	20	5	25%	3
New senior exec	15	6	40%	4
Job-posting spike	30	6	20%	2
Reposted role	10	1	10%	0

What you see - posting spikes produce the most volume (30 actioned) but exec moves produce the best precision and the most briefs per call.
What you change - move two consultant hours a week from reposts to exec-move follow-up, and require a second signal before actioning a posting spike.
What you check next month - whether precision on postings rises once stacked signals are required, and whether the time saved on reposts shows up as more exec-move conversations.

Small samples are noisy. With 10 or 15 actioned signals per type, treat a precision figure as a hint, not a verdict, and wait for a second month before cutting a signal type completely.

The 30-Day Measurement Routine

You can have a usable scorecard in four weeks without new software. The goal of month one is clean data, not perfect conclusions.

Week 1: Set the baseline and the fields

Agree the definitions - use the table above so everyone counts the same way.
Add three fields - signal type, signal date and outcome (real need, no need, no answer).
Add the reason code list - the eight codes above, one per false positive.
Pull a cold baseline - last quarter's cold reply rate and meeting rate, measured the same way.

Week 2: Log every action

Tag at the moment of contact - the consultant records the signal type and date when they send the message, not afterwards.
Record the first answer - within 48 hours of a reply, mark real need or no need.
Sample five misses - on Friday, pick five no-need outcomes and assign reason codes.

Week 3: Compute the first scorecard

Precision and reply rate per type - ignore types with fewer than ten actioned signals.
Lead time - for every brief that surfaced, note how many days early the signal was.
Reason-code tally - which two codes cause most of the misses?

Week 4: Decide and change one thing

Pick one fix - for example, a stricter ICP filter, a signal-age cap or a rewritten opener.
Set the next target - a precision figure for your top signal type, with a review date.
Start the placement cohort - tag every month-one signal so you can trace it to billings later.

How boilr Supports the Measurement Loop

You can run this routine in a spreadsheet. The reason to run it inside a tool is that signal, ICP fit, contact and your decision sit in one place. Here is how the pieces line up, based on what the boilr signals page describes [7]:

Signals - funding rounds, new hires and executive moves, tech migration, multi-turn signals (several signals from one account in a short window) and custom signals you describe in plain language, each linked to its original source so you can verify it.
ICP matching - the agent learns your ICP (industry, company size, geography) and filters out irrelevant signals before they reach your inbox, which attacks the ICP-mismatch false positives.
Lead scoring - signals are scored against your ICP so the best-fit accounts rise to the top of the queue, which is where precision data will be most reliable.
Tasks - each signal arrives as a finished Task with the signal, a decision-maker contact and a suggested opener. Verify, send, skip or flag it: every one of those actions is a data point for your scorecard.
Multi-turn signals - stacking several signals on one account is the simplest structural defence against ghost jobs and echo signals.
Company Brain - the shared memory that holds your ICP and winning patterns, so a consultant leaving does not take your knowledge of which signals work with them.
Integrations - syncing to your CRM or ATS such as Bullhorn or RecruiterFlow means the signal source can travel with the company record through to the placement.

What stays human, and should:

Outcome judgement - only the consultant knows whether a call revealed a real need.
Reason codes - a ten-second decision after each miss, made by someone who heard the answer.
Sending - boilr drafts, you verify and send.
Monthly review - deciding which signal types earn more time is a management call.

A signal you cannot trace to a brief is a hope, not a pipeline. See how Tasks, ICP scoring and signal sources fit together in boilr and start tagging outcomes from the first signal.

Mistakes That Make Signal Metrics Useless

Counting signals received instead of signals actioned - precision needs a contact attempt as its denominator.
Mixing signal types - an average across funding, reposts and exec moves hides the one that pays.
Using meetings as the success test - a polite meeting is not a brief.
Judging placements too early - recent signals always look like failures; use cohorts.
Copying a vendor benchmark - reply-rate figures vary by definition, as the 0.45% and 3.43% examples show [5] [6].
Skipping reason codes - without them, a false-positive rate is only a mood.
Cutting a signal type after ten data points - wait for a second month.

Frequently Asked Questions

What is a good precision rate for hiring signals?

There is no credible public benchmark for recruitment desks, so any universal number would be invented. Measure your own baseline for a month, then compare signal types against each other and against your cold outreach. The useful target is "better than our cold baseline, and rising".

How many signals do I need before the numbers mean anything?

As a rule of thumb, treat fewer than ten actioned signals per type as anecdote. A small desk can still learn by pooling signals over two months, or by comparing groups of signal types rather than each one.

How do I track signal-to-placement when deals take months?

Tag every actioned signal with its type and month, and carry that tag onto the company or vacancy record. Review placements by signal-month cohort over six months. The tag is the hard part, so do it on day one.

How much of the signal feed is actually ghost jobs?

Estimates vary with the definition and dataset. Greenhouse put it at 18 to 22% of jobs on its platform [1], and a separate analysis cited by Robert Walters found 34.4% of postings across 20 common UK roles in 2025 [3]. Treat posting-based signals with caution unless something else corroborates them.

Should I drop signal types with low precision?

Not automatically. A low-precision type can still work if it is cheap to action, or if it becomes high precision when stacked with another signal. Downgrade first, drop only after two months of consistent results.

Who should own the scorecard on a recruitment desk?

The BD lead or desk manager, with consultants responsible for logging outcomes. Ten minutes a week of review is enough. If nobody owns it, it will drift within a month.

Does boilr calculate these metrics for me?

boilr gives you the structure: sourced signals, ICP-based filtering, scored accounts and Tasks you verify, send or skip. Deciding what counts as a real need, and reading the monthly trend, remains a consultant decision. Check the current product for exactly which reports are available.

Sources

Information sourced from public industry reports as of November 2026.

Back to all articles