Buying Signal Conversion Rate: We Filtered 4,774 Signals to 341 Leads
Most buying-signal statistics you will read are laundered. "Stacked signals convert 5 to 10x." "Signal-based outreach replies at 5 to 18 percent." The same ranges appear word for word across a dozen vendor blogs with no source behind any of them. So instead of quoting someone else's uncited number, here is a real one from our own pipeline, with the methodology attached.
We ran buying-signal detection across a market and watched 4,774 raw signals become 341 qualified leads. That is a 7.1 percent conversion from raw signal to actionable lead. This is what happened at each stage, what got cut and why, and what a number like that should actually tell you.
The scope, up front
One line of honesty before the numbers, because scope is what separates a real figure from a marketing stat. This was one detection run, for one customer's ICP, across a defined window, pulling five signal types: hiring, funding, job changes, competitor engagement, and tech stack changes. It is not a multi-company scientific study, and the exact percentages depend on the ICP and the signal mix. What generalizes is not the number, it is the shape of the funnel.
The funnel: 4,774 to 341
Raw signals in: 4,774. Qualified leads out: 341. Between those two numbers are three cuts, and the size of each cut is the actual lesson.
The instinct when you see a 93 percent drop is that the detection was bad. It was not. A signal firehose is supposed to be mostly noise, because you are casting across an entire market and most of that market is not your customer and not in motion. The job of a signal system is not to find a lot, it is to throw away the right things. Here is where the other 4,433 went.
Where most signals died: fit, not volume
The single biggest cut was ICP fit.
Most raw signal activity fires on accounts that were never going to buy from this customer: wrong industry, wrong size, wrong geography, or a role outside the buying committee. A funding round is a real event, but a funding round at a company you cannot sell to is trivia. The same is true of every signal type.
This is why filtering matters more than detecting. A team that monitors more sources without a hard ICP filter does not get more pipeline, it gets a bigger firehose and more burned rep hours. We wrote about that failure mode in more signals will not fix your pipeline. Fit cut the largest share of the 4,433, and that is the healthy result, not the disappointing one.
The second cut: recency
The second cut was recency. A real signal at a real ICP account is still worthless if it is too old to act on.
Buying signals decay, and different types decay at different speeds. A behavioral intent signal can be stale in days, a funding round is strongest for a few weeks, a new executive is workable for a couple of months. By the time a batch is processed, a meaningful slice of technically-valid signals have already aged out of their window. Cutting them is correct, because a signal worked past its decay window is just a cold pitch with a worse hit rate.
Dedup: the noise inside the signal
The third cut was duplication. The same account can trigger several raw events, and the same event can surface through more than one source. Left in, duplicates inflate the count and send two reps at the same company. Collapsing them removes volume that looks like signal but is really the same signal counted twice. It is a smaller cut than fit or recency, but it is the difference between a clean list and a messy one.
What survived, and what it looked like
The 341 that made it through were not a random seven percent. They skewed hard toward one thing: stacked signals.
The accounts that scored highest were the ones with two or more fresh signals pointing the same direction. A raise plus hiring. A new leader plus a growing team. They scored highest partly by construction, since our model explicitly weights signal strength and recency, which favors accounts with more than one fresh signal, and partly because a multi-signal account is genuinely likelier to be in a buying moment. That is not proof that stacking beats single triggers in-market, it is directional, single-run evidence, and we would rather show you that than repeat the uncited "5 to 10x" everyone else quotes. The scoring method is what turned the 341 into a ranked call list.
What the published benchmarks get wrong
While researching this, we tried to verify the buying-signal stats that circulate as fact. Most did not survive a check.
Several widely-repeated case-study numbers could not be confirmed on the vendor's own site. A "Forrester 11-day average signal-to-outreach" figure appears on marketing blogs but on no Forrester property we could find. The "stacked signals convert 5 to 10x" range shows up near-verbatim across unrelated domains with no shared citation. That is what an uncited number copied between blogs looks like, and we could not trace it to any original research.
The one category with genuinely traceable data is response speed. The Harvard Business Review study "The Short Life of Online Sales Leads" (Oldroyd, McElheran, and Elkington, 2011) audited 2,241 companies and found the average first response took 42 hours, with only 37 percent answering inside the first hour. The same article cites a related study of 1.25 million leads showing that firms responding within an hour were nearly seven times as likely to qualify the lead as those who waited even one more hour, and more than 60 times as likely as those who waited a full day. That is real, sourced data, and it reinforces the whole point: with signals, speed is most of the edge.
What a seven percent conversion should tell you
If your raw-signal-to-lead rate is in the single digits, that is usually not a broken system, it is a working filter. The takeaways from the funnel:
→ Judge a signal system by what it throws away, not what it surfaces. Precision beats volume. → Fit is the biggest cut, so the ICP filter is the highest-leverage part of the stack. → Recency is the second biggest, so speed and decay-awareness are not optional. → The survivors cluster around stacked signals, so weight combinations heavily. → Distrust any round, uncited conversion stat, including the ones that flatter the category.
The number that matters is not 7.1 percent. It is the shape: a lot in, most correctly discarded, a small ranked set worth real effort. Build for that, and signals become an edge instead of a firehose.
Frequently asked questions
What is a good buying signal conversion rate? There is no universal figure, because it depends entirely on ICP breadth and signal mix. In one of our runs, 4,774 raw signals became 341 qualified leads, about seven percent. A low raw-to-qualified rate is usually a sign the ICP filter is working, not failing.
What percentage of raw signals become qualified leads? In our data, roughly seven percent, after cutting for ICP fit, recency, and duplication. Expect most raw signal activity to be filtered out; that is the system working as intended.
Why are so many signals filtered out? Because detection casts across a whole market, and most of that market is not your ICP or not in motion. Fit is the largest cut, followed by recency (aged-out signals) and deduplication.
How do you calculate signal-to-lead conversion? Divide qualified, ICP-filtered leads by the raw signals detected in the same window. The honest version reports the stages in between, fit, recency, and dedup, not just the endpoints.
Did stacked signals really score higher in your run? Yes. The surviving high-score accounts skewed heavily toward those with two or more fresh signals. That is directional, single-run evidence about scoring and survival, not a measured conversion lift, but it is real evidence, which is more than the uncited "5 to 10x" claims offer.
The honest number
We would rather publish one real funnel with its methodology attached than repeat a flattering statistic we cannot source. 4,774 signals in, 341 leads out, most of the difference cut for fit and timing, and the survivors clustered on stacked accounts. If your own numbers look similar, your filter is working. If they look too clean, check whether you are filtering at all.
That 4,774 to 341 cut is the filter and score stages in action. See the full system.