Signal Activation: Tracking Signals Was Never the Hard Part

Signal based outreach is the loudest category in GTM right now. New platforms every month. Enterprise suites priced like infrastructure. Long threads about which feed catches job changes fastest.

Almost all of that conversation is about detection. Detection is the part that already works.

Here is the uncomfortable version, learned building our own outbound before we sold it to anyone: knowing that an account lit up is easy. Knowing what to do about it, and actually doing it that day, is the whole job. That gap is where most signal investments quietly die.

Detection is a commodity now

A few years ago, knowing that a company had just posted the job that means budget was approved was an edge. Today it is a purchase. Job posts, funding rounds, exec moves, tech stack changes, LinkedIn engagement: every one of those has multiple vendors, most of them cheap, several of them free if you are willing to maintain a scraper.

What follows is predictable. A team buys a feed. The dashboard fills up. Nothing else changes.

We have run more than a hundred onboarding calls this year and the same conversation happens on most of them. The signals are real, the dashboard is busy, the pipeline looks like last quarter. When we ask what happens after a signal appears, the answer is always a person. Someone will look at it. Someone will decide. Someone will write something. Eventually.

Signals do not wait politely. A hiring signal is worth the most in its first two weeks, because 73% of those roles go live within 30 days of budget approval and formal vendor research usually does not start for another 60 to 90 days. That is the window where you are the only vendor in the room. A signal sitting in a dashboard for three weeks is not a signal, it is a record.

Where the activation gap actually opens

Between "signal appears" and "message sent" there are four decisions. Every one of them is a place teams stall.

→ Is this even our buyer? Raw feeds fire on the whole market. In one recent run of ours, 4,774 raw signal records came down to 341 worth touching, which is 6.7%. If a human does that filtering by eye, the filtering becomes the job and nothing else gets done.

→ Which of these do we work first? Twenty accounts lit up today. A rep has capacity for eight good messages. Nobody has told them which eight.

→ What do we actually say? The signal names an event. The message has to name what the event costs. That means reading the account, and reading takes minutes per lead that nobody has at volume.

→ Who sends it, from where, and what happens on reply? LinkedIn or email, which sender, which sequence, and what stops the moment they answer.

Miss any one of those and the feed becomes decoration. Most teams miss three.

The real price of stitching it together yourself

The common answer is to build it. Clay in the middle, a scraper on one side, an enrichment provider, a verifier, a sequencer, a sheet holding it together, and a Slack channel where it breaks.

It works. It also carries a price that never makes it into the business case.

→ Maintenance hours every week, spent by your most expensive GTM person. Credits move, actors change their output shape, a column silently returns null and 400 messages go out with a blank first line.

→ Per signal costs that only reveal themselves at volume. Enrichment and scraping are priced per call, so a workflow that looks cheap across 200 leads changes character across 4,000. Batching and caching stop being optimizations at that point and become the difference between a working margin and none.

→ No memory. The stack does not know it messaged this person six weeks ago on a different signal.

None of that is an argument against Clay, which is a genuinely good tool. It is an argument against the assumption that the hard part is plumbing. The hard part is the decision layer, and plumbing does not supply one.

Why the expensive tools solve the wrong half

At the other end sit the enterprise platforms. Six figures, a quarter to implement, an intent score per account.

What you get is a better dashboard. A number between 0 and 100 telling you an account is "surging," with no named person, no behavioral event, and no next action attached. A rep cannot open a conversation with an account level score. They can open one with "your Head of RevOps commented on a competitor's post about attribution three days ago."

The pattern at both ends of the market is the same. Detection has been solved several times over at several price points. Activation has barely been attempted, because activation requires opinions: who to skip, what to say, when to stop. Vendors are reluctant to have opinions.

The activation loop: signal to sent in five steps

This is the loop we run on our own pipeline and set up for customers. Five steps, and the order is the point.

1. Detect against a defined ICP, not against the market. The filter belongs at ingestion, not in a rep's inbox. Titles, company size, geography, and the exclusions that matter, applied before anything reaches a human.

2. Attach the consequence to the record. Every signal arrives with the context needed to write to it: what the company does, what the event implies, what it usually costs them next quarter. Without this, step 4 turns back into research.

3. Rank by freshness first. Not by score. A two day old commenter beats a three week old funding round, because the funding round has already been called by nine vendors.

4. Draft per lead, approve in batch. The per lead writing happens automatically, a human reads a queue and approves. This is the only version of "personalized at volume" that has held up for us.

5. Send from the right channel and stop on reply. A LinkedIn connection plus a first message for engagement signals, email for hiring and funding at scale, and a hard stop on the sequence the second a human answers.

Run against ICP filtered signals with consequences attached, that loop gives us roughly 55% acceptance and 30% replies on LinkedIn, against 20 to 30% acceptance and 5 to 8% replies on cold lists. One of our own inboxes went from under 1% replies to 42% once the loop was closed. The audience never changed. The activation did.

Activation capacity: the number nobody calculates

Before buying another feed, work this out: how many signal driven messages can your team actually send per day, at a quality you would be willing to receive?

For one rep writing properly, it lands somewhere between 10 and 30. Call it 20. That is 100 a week, roughly 400 a month.

Now look at how many signals your feed produces a month. If it produces 4,000 and your capacity is 400, then 90% of what you are paying for never gets touched. You do not have a detection problem. You have an activation ceiling. Buying a second feed moves 4,000 to 7,000 and your pipeline stays exactly where it was.

There are two honest ways to raise that ceiling. Tighten the filter so the 400 you can send are the right 400. Or automate the per lead work so the same person approves 200 instead of writing 20. Both beat adding coverage, and neither shows up in a vendor demo.

Where agents change the math

This is the one place the AI story is not hype, and it is narrower than the marketing suggests.

An agent is bad at deciding who your buyer is. It is very good at the reading. Pulling account context, working out what the event implies, drafting four lines in your voice, and doing it for every record instead of the ten a human gets to before lunch.

That is why B2B Signals ships through MCP and not only a UI. Claude Code or any MCP aware agent can pull today's ICP filtered signals directly, draft per lead, push into the sequencer, and leave a human approving the queue. The agent does the reading. The operator keeps the judgment and the send button.

We built this for ourselves first, because we had the exact problem everyone on those onboarding calls has. The signals were never the bottleneck. Monday morning was.

Frequently asked questions

What is signal activation? Everything that happens between a buying signal appearing and a message landing with the right person: fit filtering, prioritizing, writing to the signal, choosing the channel, and stopping cleanly on reply. Detection tools deliver the event. Activation is what turns the event into a conversation.

How do I know whether I have an activation gap? Compare signals produced per month against messages actually sent off those signals. If under half get touched, activation is your bottleneck and more coverage will not help. Second tell: ask how old the average signal is when someone finally works it. Over two weeks means you are arriving after the moment has passed.

Is building it in Clay a mistake? No. For a team with a dedicated GTM engineer it can work well. Just price it honestly: maintenance hours from an expensive person, per call costs that scale badly, and no shared memory of prior touches. Most small teams discover that the maintenance is the product.

Should I automate the whole loop and step out of it? No. Automate the reading and the drafting, keep a human on approval and on replies. Fully autonomous sending is where the AI SDR category earned its churn rate. The approval queue is what protects the quality that made the signal worth acting on.

Which signal should a small team activate first? Hiring signals for the role that implies budget for what you sell, then competitor engagement. Hiring gives you the widest window, since vendor research starts 60 to 90 days after the post goes live. Competitor engagement gives you the warmest person, because they just raised their hand on your exact problem.