Signal-Based Selling Already Had Its Category Moment. Twelve Months and Six Deals Later, It's a Feature.

Signal-based selling was supposed to be the fix for outbound. Buy better intent data, catch the moment a buyer starts looking, reach out while the window is open. That was a real improvement over list-and-blast prospecting, and the practice still holds up. What did not hold up was the other half of the pitch: that this was a category, a distinct space in the sales stack with its own vendors, its own budget line, and its own future as an independent business.

Between July 2025 and July 2026, six companies making that pitch were acquired, folded into a larger platform, or shut down. One of them shut its product down five months after raising a $15 million Series A. Two of them changed hands two days apart. Every buyer was a company that already owned a system of record or a platform, and not one of them bought a signals company to keep running it as a signals company.

That is worth sitting with before anyone pitches signal-based GTM as an emerging category again, because the pitch has already been made and the market answered it. The useful question now is narrower and harder: given that it got absorbed, what has to be true for signals to actually pay off? This post argues the answer is one condition, and that most of the ways teams buy signals today fail it.

The short version

• Six signal-adjacent GTM companies changed hands or shut down in 12 months, from July 2025 to July 2026.
• Koala is the clearest case: acqui-hired by Cursor's parent for engineers, product shut down September 30, 2025.
• Warmly went to HubSpot on June 30, 2026. Common Room went to Zoom on July 2, 2026. Two days apart.
• Every acquirer already owned the customer context. None bought a signals company to run it standalone.
• Over the same window, published cold-email reply averages fell to 3.43 percent, from 5.1 percent in 2024.
• Google enforced bulk sender rules from February 1, 2024. Microsoft from May 5, 2025. Both at 5,000 messages a day.
• The two trends coincide. Neither dataset proves the other caused it, and this post does not claim it did.
• The condition that holds: a signal pays off when the system that catches it also owns the surface it acts on.

What the category was sold as

The strongest version of the pitch treated signal-based selling as infrastructure, not a tactic. Apollo's own framework page defines it as a revenue methodology that prioritizes accounts and triggers outreach based on verified buyer intent signals rather than static lists or demographic criteria, then lays out a five-phase operating model covering discovery, scoring, routing, execution, and optimization. It reads as a replacement for how a revenue team works, not a feature inside one.

The supporting claims followed the same shape. Catch signals six to seven weeks earlier than competitors and you get a structural advantage. Pair intent data with targeted messaging and conversion rates improve substantially. Only a minority of B2B companies use signal tools at all, so the window to get ahead is open. Every one of those claims is defensible on its own. Together they describe a category with room to grow into a large independent business.

There is one detail on that same page that cuts the other way, and it is the honest one. Apollo cites a figure that only 30 percent of marketers use third-party intent data and just 12 percent find it useful. That is the category's own advocate reporting that seven out of eight people who tried the purchased version of the product did not get value from it. The definitional groundwork for all of this is covered separately in what buying signals are and how to track them. This post is about what happened to the companies selling them.

The consolidation record, with dates

Six deals in 12 months, all in or adjacent to the signals space. This is the full set found rather than a selection, which matters, because a pattern assembled from cherry-picked examples proves nothing. Dates are announcement dates unless noted.

Two honest qualifiers on that table. Fin is a customer support agent rather than a signals product, so it is the loosest fit of the six. It is included because the buyer's logic is identical: a platform company buying an agentic point solution to run it inside an existing system of record. And Qualified is an inbound conversion product rather than an intent-data vendor, which is exactly why it belongs, as the argument below turns on inbound being the surface where signals actually work.

The rest are squarely in the category. Pocus turned buying signals into prioritized action for product-led teams including Asana, Canva, and Monday.com. Warmly identified more than half of anonymous website visitors as named individuals and shipped two agents on top of that. Common Room raised roughly $53 million, including a $32.3 million Series B led by Greylock in April 2021 at a $300 million valuation, and sold to Zoom for terms nobody disclosed. Koala raised $15 million and shut the product off.

Blu Agent

Warmly to HubSpot on June 30, 2026. Common Room to Zoom on July 2, 2026. Two of the category's best-known independents were absorbed inside 48 hours of each other, by buyers with nothing in common except that both already owned the customer record.

Koala is the clearest case, because the product actually died

Five of the six deals moved a product inside a bigger platform. One of them ended the product. That makes Koala the most informative data point in the set, because acquisitions can mean anything and shutdowns cannot.

Anysphere, the company behind Cursor, announced the Koala deal on July 18, 2025. It was an acqui-hire in the literal sense. Several of Koala's top engineers and all three founders joined Cursor to build an enterprise-readiness team. Cursor did not plan to integrate the CRM product at all, and Koala's own blog post said the platform would shut down in September. It went dark on September 30, 2025.

The timing is the part that should register. The announcement came five months after Koala closed a $15 million Series A led by CRV, with participation from HubSpot Ventures, Recall Capital, and Afore. A company that had just closed a Series A, in the middle of the category's hottest moment, took an engineering-talent exit and turned off the software. That is not what a growing category looks like from the inside.

There is a second detail worth naming. HubSpot Ventures was on Koala's Series A cap table in early 2025. Twelve months later HubSpot bought Warmly outright. Same investor thesis, two different routes, one of which ended in a shutdown and the other in absorption. Neither ended in a standalone signals business.

What happened to the channel the signals were feeding

The category's core promise was that better signals make outbound work. Over the same period the companies were consolidating, the published numbers on outbound reply rates moved down and the mailbox providers put hard gates on volume. Those are two separate observations, and the relationship between them is correlation. Nobody has run the experiment that would establish cause.

Start with the gates, because those are documented facts rather than estimates. Google and Yahoo announced joint bulk sender requirements on October 3, 2023, effective February 1, 2024. Anyone sending more than 5,000 messages a day to personal Gmail or Yahoo accounts has to pass SPF, DKIM, and DMARC at minimum p=none, offer one-click unsubscribe, and keep spam complaint rates below 0.30 percent, with Google recommending under 0.10 percent. Microsoft followed on May 5, 2025 with its own authentication requirements at the same 5,000 messages a day threshold. Microsoft originally planned to route non-compliant mail to the Junk folder and then amended that: non-compliant mail is rejected outright with a 550 5.7.515 access denied response. The full deliverability arithmetic is worked through in the honest map of where agent work pays off.

Now the reply rates, which need more care, because the sources do not agree and the reason they disagree is instructive.

Belkins deserves credit for the most useful sentence in any of these reports. Its 2026 numbers came in far below what it had published in prior years, and rather than treat that as a market finding, it said the reason outright: the reply rate is now calculated against total emails sent instead of against unique openers. That is a denominator change, not a collapse. Any post that stacks these vendor benchmarks into a single tidy decline curve is comparing measurements that do not measure the same thing.

So here is the defensible version. Every independent platform dataset points in the same direction. The two largest disclosed samples happen to land on the same 3.43 percent figure, which is a coincidence worth noting rather than a second data point confirming the number is precise. The widely repeated 8.5 percent 2019 baseline traces back to a link-building outreach study rather than sales prospecting, so the shape of the full seven-year curve is softer than it gets quoted as. Direction: down, across every source. Magnitude: unresolved. Cause: not established by any of them.

Why better signals could not rescue outbound

This is the mechanism, and it does not depend on the reply-rate data being precise. A signal tells you who and when. It does not tell you what to say, and it does not get you past a rate limit. Those are the two things that were actually binding.

Take the rate limit first. Both Google and Microsoft set their thresholds on volume, at 5,000 messages a day, and their enforcement on authentication and complaint rate. A more relevant message helps the complaint rate, which is real. It does nothing about the fact that the gate exists and is administered by someone else. Any strategy whose payoff scales with send volume is a strategy whose ceiling is set by two companies that have shown they will lower it.

Then take exclusivity, which is the deeper problem with purchased signals. A third-party intent feed is a product, and products get sold to everyone who will buy them. If a data vendor can tell you that an account is researching your category, it is telling your three closest competitors the same thing in the same week. The signal is real and the advantage it confers is shared, which is a different thing than an advantage. Apollo's own 12 percent usefulness figure is what that looks like in a survey.

Woodpecker's platform data shows the volume problem directly, without any need to argue about it. Campaigns under 50 contacts averaged a 5.8 percent reply rate. Campaigns over 1,000 contacts averaged 2.1 percent. Same platform, same period, same measurement. Reply rate falls as list size rises, which is the opposite of how a scalable channel behaves. Personalization sorts the same way: advanced personalization ran 17 to 18 percent against 7 to 9 percent for basic or none, and personalization at depth is the thing that does not get cheaper per prospect as volume rises. The authenticity gap that opens up when teams try to automate past that is covered in what fully autonomous AI SDRs get wrong.

Put those together and the category's economics come apart on their own terms. Signals improved targeting. Targeting was not the binding constraint. The binding constraints were a volume gate someone else controls and a message quality cost that stays flat per prospect. Neither is fixed by knowing more precisely who to email.

The surface test

The surface test is one question per signal: name the surface you would act on, and name who controls it. That is it. If the surface is your own site, your own product, or a thread that is already open, then you control the act and the signal is worth catching. If the surface is a stranger's inbox, the act is gated by Google and Microsoft, and a better signal buys you nothing at that gate.

This is the companion to a test in an earlier post rather than a replacement for it. The owned-signal test asks whether the signal already exists in data you own. The surface test asks the next question, which turns out to be the one that decides whether the work pays: assuming the signal exists, do you own anywhere to act on it?

The top four rows are all inbound or install-base work, and they are all first-party. The bottom three are the ones the category was mostly sold on. That is the whole finding, and it explains the acquisition table better than any story about market timing: the signals that survive the test are the ones that only make sense inside the system that already holds the customer, and a standalone signals vendor by definition is not that system. The related case for unifying inbound, outbound, CRM, and product signals into one pipeline is made in how B2B companies build stronger pipelines.

Every buyer bought the same thing

Look down the acquirer column and the pattern is one thing repeated six times. HubSpot, Salesforce twice, Apollo, Zoom, and Anysphere. Five of the six already owned a customer record, a communications surface, or both. The sixth wanted the engineers and turned the product off.

That is the tell. None of these companies bought a signal because a signal is valuable on its own. They bought a missing input for a context they already controlled. Apollo said as much in its own announcement of the Pocus deal, which described a unified system that detects buying signals, prioritizes accounts, and guides execution all within a single platform. Which means the loudest advocate for signal-based selling as a five-phase operating model ships it as a layer of a suite, because that is where it works.

This is where the argument stops being about other companies' M&A and becomes a design claim, and it is worth being direct that it is Intempt's claim rather than a survey result. A signal is not a product. It is an input, and inputs belong in the system that holds the profile and runs the response. Catching a stalled trial and acting on it is one job. Splitting it across a signals tool and a messaging tool, joined by a webhook, means the message that goes out knows about the trial and nothing else: not the last three emails, not the open deal, not the support ticket from Tuesday. This is the gap the agentic GTM platform is built to close, with the Lifecycle Marketer working the same customer profile the signal landed on rather than a copy of it.

The one customer context framing matters more here than it sounds. Unified profile is a phrase every vendor uses. The operational version of it is narrow and checkable: when a signal fires, does the thing that responds have access to everything else known about that person, or does it have access to the signal payload? Every row in the surface test that holds up depends on the first answer. The agent roles that run those responses are only as good as the context they read from.

What this argument does not claim

Four limits, stated plainly, because the contrarian version of this post is easy to overstate and the overstated version is wrong.

1. 1Not that intent data is worthless. First-party intent is the strongest asset in the set. Purchased third-party intent is weaker than advertised because it is non-exclusive, not because it is fake.
2. 2Not that outbound is dead. A hundred well-researched sends a week is a working program. It is just not a program that gets better by adding a signals vendor and more volume.
3. 3Not that the inbox rules caused the reply-rate decline. The two coincide. No dataset here establishes cause, and buyer fatigue and a flood of generated outreach are competing explanations nobody has separated out.
4. 4Not that any analyst declared this category dead. No report says that. The consolidation record and the benchmark data are real and dated. The reading of them as a category absorption rather than a category maturing is this post's own diagnosis.

There is a fair counter-reading, and it deserves to be on the page. Absorption is what happens to good technology when the market decides it belongs everywhere rather than nowhere. Email marketing was a category once and is now a feature of every platform, and that was not a failure of email. The distinction that matters is whether the absorbed thing still needed its own vendor. On this one, six buyers voted no in 12 months.

How to test this on your own stack

1. 1List every signal your team currently pays for or collects. Separate them into first-party and purchased. Count the line items on each side.
2. 2Run the surface test on each one. Name the surface and name who controls it. Be strict: a cold inbox is not a surface you control.
3. 3For every purchased signal, ask which competitors can buy the same feed. If the answer is all of them, price it as a shared advantage.
4. 4Check what the acting system can see. When a signal fires, does the response have the full profile or just the payload? Look at the actual integration, not the diagram.
5. 5Move one signal from an outbound trigger to an inbound nurture trigger and measure the step it should move, not revenue. Reply rate on the thread, or trial-to-paid on the journey.
6. 6Give it a quarter. If the inbound version of the same signal does not beat the outbound version, the problem is the signal, not the channel, and you have learned something cheap.

Step four is the one teams skip, and it is where most signal projects quietly fail. A webhook that posts an account name into a sequencing tool technically integrates two systems and delivers none of the context that made the signal worth catching. The difference between first-party website data and scraped intent is the same argument applied to the input side, and signal-based lead generation covers what the lift looks like when it is done on owned data. For the wider stack question of what to buy at each stage, the GTM stack mapped by stage is the companion piece.

The version that holds up

Read the primary sources if you want the raw material. The Koala acqui-hire coverage is the sharpest single document in the set, and Zoom's own announcement of the Common Room deal is the most recent. Neither one draws the conclusion this post draws. That is the correct division of labor: they report what happened, and the reading is arguable.

The defensible claim is smaller than the original category pitch and more useful than it. Signals work when the system that catches them also owns the surface that responds, which in practice means first-party behavior feeding inbound nurture on one shared profile. They do not work as a purchased feed pointed at a channel two other companies rate-limit. Twelve months of consolidation says the market already priced that in, and the teams still shopping for a standalone signals vendor are buying into a category that has already been absorbed. Run the surface test on your own signals before you renew anything, and if the answer keeps coming back inbound, see what one customer context does with it instead of buying signal-based selling as a product.