“We looked at signal-based prospecting. It’s too complicated for where we are, so we’re just going to send more.”
That was a revenue leader at a B2B SaaS company, already behind on the quarter. There is an intelligent way to run signal-based prospecting.
High volume is not it.
The prevailing story: signals are complicated. And “complicated” becomes the permission slip to do one of two things, buy a one-size-fits-all detection tool and treat it as a ‘silver bullet’, or fall back on volume. I argue it’s nuanced, which is a different problem with a different fix.
‘Just send more’ is worth decomposing. The list ages. The ICP blurs. The conversion rate drops at every step. And the response to the drop is more sending.
I have run that play. Across the engagements I have personally seen, no static purchased list has won. Five years ago you could buy a list and spam it. Not now.
”Complicated” is usually just unmeasured
Complicated implies many interacting parts you cannot reason about. That’s not where the real difficulty sits. Most revenue teams can already reason clearly about why more signal would sharpen their targeting.
The harder part has always been building the infrastructure to capture that signal and act on it.
Before GTM engineering existed as a function, that gap was structural, not intellectual. RevOps and go-to-market teams didn’t have engineering capacity of their own. Any real build meant getting in line behind the software engineering team, and that team’s time went to customer-facing product first, by default. Internal tooling took a back seat.
So teams duct-taped things together without real coding ability, not because the problem was too tangled to reason about, but because building the right solution was out of reach.
The engineering signal-based prospecting requires isn’t deep anymore. AI-assisted and no-code tooling means GTM teams can build a good chunk of this themselves, without waiting in queue behind product engineering.
What used to sit permanently in the “we’re not going to do that” bucket is buildable in-house now. That’s the shift behind the recent interest in signals.
A vendor selling one detection engine to every customer runs into a version of the same problem. It has to assume a shared definition of a buying signal. But the buyer defines that signal — by vertical, by firm size, by how the decision gets made.
That means the vendor calibrates it correctly for a small slice of the company’s customers and roughly wrong for the rest. Now, I can already hear the counter-argument to all this.
“Fine, but our data isn’t ready, so this is theoretical for us.”
My response is that unready data is telling you to build measurement gates. Not buy more volume. In short, what you’re thinking of as complicated is merely a symptom of how much you aren’t measuring.
Characterize your buyer journey using data before you do anything
The right signal depends on inputs about the buyer that most teams haven’t collected. The first of those inputs is how the buyer buys, and it has to come before anything else.
A signal only means what you think it means once you know who’s doing the buying and how. Two factors carry most of it. Firm size and vertical.
Firm size changes the motion end to end. At the smaller end the process is ad hoc, the cycle is short, one person decides.
By the time you reach enterprise, the organization runs a formalized RFP (request for proposal) process, wants a pilot, and puts a buying committee in front of you. Winning one champion moves nothing.
Vertical does its own work. Two companies with identical headcount in different verticals do not buy the same way.
A signal that predicts a single-decision-maker SMB purchase is close to noise for a committee-led situation. The committee evaluates in writing, across functions, and leaves a different paper trail. I argue cohort selection has to precede signal selection.
Cohort selection isn’t just about picking the right factor. It’s about picking the right resolution within it. There are two ways to get this wrong, and they sit at opposite ends of the same spectrum.
Treat the whole prospect base as one group, and you’re back to one-size-fits-all, which is the exact problem a generic detection engine has. Go too far the other way and characterize every single prospect individually, and you lose something just as valuable: there’s no longer a pattern across any group large enough to plan a campaign around.
The right level sits somewhere in the middle, and finding it is a judgment call. The fastest way to make it is to get on customer calls, record them, and listen for how prospects describe their own buying process.
Voice-of-customer input finds the real boundaries faster than any amount of data slicing will.
That’s one axis handled: who you’re segmenting by. The other axis, independent of it, is how many steps you’re measuring between first touch and close. That’s where the gates come in.
Build the measurement gates before you pound pavement
The sales team has been out there for eleven weeks. Marketing has been running campaigns against a cohort nobody wrote down. The CRM has stage names people define in different ways.
And now it’s the last week of the quarter and you’re sitting there trying to reverse-engineer a conversion rate out of activity nobody instrumented. The numbers are a story you’re telling about the quarter, not a measurement of it. This is the worst case and it’s common.
Three categories cover most of what you need:
- Volume from step to step
- Speed from step to step
- Conversion rate from step to step
Illustrative, using a modest cohort: (7 qualified first 3 weeks) / (120 leads engaged first 3 weeks) = 5.8% conversion rate.
How many of these steps you track is easy to get wrong. Treat the whole process as a single step and you won’t have the resolution to know where the funnel is actually breaking.
Go the other way — an 87-step pipeline — and you’ve created a different problem: reps can’t keep up with a process that granular, the admin load swamps the team, and most of those extra stages add nothing you’ll ever act on.
The way through is to balance two directions of reasoning at once.
Top-down: someone in strategy, ops, or leadership proposes a pipeline shape from experience and refines it against historical data where that data exists. Experienced operators’ first guess here is usually already close to what the data would say.
Bottom-up: segment gates around how the team is actually built. Every handoff between functions is a natural place to measure, because that’s where bottlenecks hide: demand gen to BDR, BDR to AE once a meeting’s set, AE to close, close to the account manager or CS team who takes it from there.
One gate holds regardless of how the team is structured: did the deal close or not. That one’s universal.
Below that, one layer of sub-segmentation per stage is typically enough. Demand gen splits into building awareness versus getting the prospect to act. AE work runs meeting set, pain point identified, qualified, and, if the deal is committee-led, committee engaged.
The stages closest to close, proposal and negotiation, end up with more resolution than anything upstream simply because they’re closest to revenue; that’s worth knowing but not worth over-engineering here.
And post-sale deserves the same treatment as the front end, broken into its own steps rather than left as one undifferentiated block. This ensures that part of the funnel can be optimized too, not just assumed to be working.
On the infrastructure side, I argue you want to minimize the amount of collection that depends on a human, because a human is where the error enters. Tools that let you work with APIs get you part of the way, taking the logging and the timestamping out of a rep’s hands entirely.
Past that you need engineering. Before GTM engineering existed as a function, I approached engineers, but this is not a limiter anymore.
Pick a cohort, then trace it backwards
This step is best illustrated with an example. We worked with a mortgage lending automation client whose market was mortgage lenders. That’s not one market.
Retail lenders, wholesale shops, credit unions, and construction lenders all buy differently, and when we started they already knew a couple of verticals converted well. That was their knowledge before it was a signal.
That’s the point.
The mechanism runs backwards. We start at the bottom of the funnel and find the cohorts of leads and deals that close at a higher rate, higher volume, and higher speed.
Then we trace that cohort back up the funnel to infer where those buyers were spending their time, physical or digital first (that split does most of the work), then handshake selling versus a small number of specific conferences inside physical, paid versus organic inside digital.
I can hear the objection: “who cares, there’s never data that clean at a startup.”
Correct. You don’t get to measure every point in the funnel you’d like to. So you sequence on what’s measurable first, then prioritize the build against that.
Separating signal from noise without a stats degree
Most marketers are running finger-in-the-wind measurement, no slight intended. It’s what their tooling and the calendar allow.
My recommendation is light statistical work, not much more complex than an A/B test.
At a startup you’ll often land on a p-value around 0.13. That’s not clean by any academic standard. The move that took me a few expensive quarters to learn: fit the statistical power to the decision in front of you rather than to the textbook.
For a binary go or no-go on a cohort, 0.13 is workable. For a decision that has to be certain, it isn’t. Size the confidence to the cost of being wrong, and keep going.
The real risk when you squeeze for correlation between a signal and pipeline movement is overfitting. Dimensionality reduction and principal component analysis are how you fight it.
Of course, some readers will never run any of this, for reasons beyond their control. For such marketers and situations, I say keep a standing panel of customers, buyers, and prospects who represent your TAM and who’ll give it to you straight.
When the qualitative contradicts the quantitative, I argue you go with the qualitative, though that preference rests on a handful of cases, not a large sample. The stats-strong people inside marketing operations are sharp, and frequently lack the customer context that tells them which correlation is real.
This is why qualitative data and talking to customers beats any statistical measurement.
Finding a signal does not guarantee a great message
I argue the most common mistake in the move from volume to signal is treating the trigger as permission to pitch. Teams find the signal, then send the same email they were already sending, with the signal bolted on as the first line.
Most outbound is just us getting contacted by companies whose entire opening move is proving they were watching us. That’s not targeting. It reads as invasive, and it burns the account.
Here’s the test I use. If the reader can get value from the email without replying, you’re doing signal-based prospecting. If they can’t, you’re doing surveillance-based pitching. I’d rather run volume with a good offer than run signals and be creepy.
The signal is the input that lets you put something of value inside the message.
“Everyone does this and it still books meetings.” Sure. And burning accounts costs you pipeline in the quarters after this one.
Teams fall back on volume because the signal work looks like a build they can’t staff, and the output looks like something no rep will trust. Both of those are measurement problems before they’re tooling problems. Krossings builds the tool and measurement layer underneath.
If you’re interested in seeing how this works live, book a revenue diagnostic. We’ll uncover signals tailored to your business and show you how you can run with them.