Only 3% of Your Market Is In-Market. Stop Building Outbound for 100%.
At any given moment, about 3% of your market is ready to buy. Another 7% is open to it and warming up. The remaining 90% is not in the market at all. That split is Chet Holmes' buyer pyramid, a rule of thumb sales teams have used for almost twenty years.
Almost nobody builds their outbound around that split. The list is one list. The sequence is one sequence. The volume is one number in a dashboard. Everyone gets the same message in the same week, because the tooling made it cost the same to send to 5,000 people as to 150.
That is the whole problem. Not the copy, not the tool, not the send volume. The problem is treating three completely different audiences as one.
The math nobody sizes their list for
Take one ICP and a clean list of 5,000 decision makers. Apply the split.
Roughly 150 of them can buy this quarter.
Roughly 350 have the problem and no timeline yet.
Roughly 4,500 have no reason to answer you at all this month.
The 4,500 are not a mistake on your list. In-market is a state, not a property. Anyone in that 90% moves into the 3% the week their budget clears, their competitor ships something, or a VP of Sales walks out. Keeping them is correct.
Messaging them like the 150 is not. That is where most outbound quietly breaks: the list is right and the treatment is identical.
Spray does not just waste the 90%. It burns the 3%.
Send one templated message to all 5,000 and the 4,500 who are not in the market are your cheapest loss. They ignore it. Nothing happens.
The expensive loss is the 150.
Those people have a live problem, a forming budget and a shortlist they are building this month. They are the one group reading vendor messages carefully, because they actually need to buy something. They open yours and see the same AI-assembled opener they got from four other vendors before lunch. You do not get a second read from that group. You spent your one shot on a message written for an average of 5,000 people, none of whom exist.
There is a channel cost on top of the trust cost. When we ran a raw batch of 4,774 signals through a real ICP filter, 341 survived. That is 7.1%. The other 93% would have been sent to, counted as activity, and would have degraded the exact accounts and domains that had to carry the 7.1% that mattered. More signals will not fix that. A filter will.
The gap between the two approaches is not subtle. Between June and September 2026 we sent 836 LinkedIn invitations to people picked by signals, with no note: 40% accepted, and 25% of the people we then messaged replied (HeyReach data, pulled 1 October 2026). Expandi's 2026 benchmark across 13.2 million connection requests puts the average at 28.5% acceptance and 10.4% replies. One inbox we rebuilt this way moved from under 1% replies to 42%. Same product, same sender, same market. Different treatment per tier.
One ICP, one list of 5,000, one account
Cap the list at about 5,000 decision makers per ICP, with their companies attached.
Not 50,000. A 50,000-row list is almost never one ICP. It is five ICPs stacked into one file, and no single message is true for all five. The moment you need one sentence to work for a 40-person agency and a 4,000-person manufacturer, you write the sentence that offends neither and interests nobody. If you cannot write the ICP down precisely, the list is not ready.
Then give each ICP its own LinkedIn account.
That sounds like overhead. It removes more work than it adds. The headline, the profile, the posts and the sender identity all get built for one reader instead of being averaged across five. A profile that obviously belongs to someone who works with people like the reader does part of the persuading before the first message lands.
It also gives the 90% a place to see you without being messaged. They read a post, they see the profile, they do not need to be in your sequence for you to be in their week.
Capacity is not a problem here, because you are not trying to work 5,000 people. One account only ever works the 150 that are live, plus a steady rhythm of light touches on the rest.
The watcher layer tells you who is in the 3%
An ICP filter tells you who could buy. It cannot tell you who is buying. Only behavior does that, and behavior changes weekly.
Signals that move a name from the 90% into the 3%:
A role opened for the problem you solve. Roughly 73% of roles go live within 30 days of budget approval, and vendor research usually starts 60 to 90 days after the post. That is the earliest legible budget event you get. Job posts are pipeline if you read them that way.
They engaged with a competitor. Followed the page, liked a post, argued in the comments. Your competitor's audience is doing category research in public.
They engaged with an industry voice on the exact topic you sell into, which is the same research behavior with a different host.
They moved into the buying seat. New role, new mandate, 90 days to show something.
The company raised. New money, new plans, a deadline attached to both.
None of these is a permission slip to pitch. Each one is a reason to be relevant this week rather than next quarter.
Three tiers, three plays
| Tier | Size on a 5,000 list | What it means | The play | What kills it |
|---|---|---|---|---|
| The 3% | ~150 | Live problem, budget forming, shortlist building | Reach out now, by hand. Name the consequence the signal points at, ask for a reply, not a meeting | A generic opener, or a calendar link in message one |
| The 7% | ~350 | Problem is real, timeline is not | Stay in the feed. Comment, react, follow up when the next signal fires. One new reason per touch | Re-pitching the same message with a new subject line |
| The 90% | ~4,500 | Not in the market | Be visible, not present in their inbox. Content from the ICP account, the occasional genuine comment | Adding them to a sequence because the seat was free |
The 3% is where humans belong. On our side a person writes those messages, off the signal and the person's actual activity, because that is the group where a mediocre message costs the most. AI does the finding, the watching and the drafting. It does not do the conversation. That distinction is the difference between a warm pipeline and an AI SDR that quietly burns the list.
For the messages themselves, the rule is simple: name the consequence, not the signal. Nobody wants to hear that you noticed they are hiring. They want to hear what you think happens next because of it.
What happens after they ignore you
Most teams answer this with a number. Six touches, then the sequence ends and the record goes cold.
That is the single most expensive habit in outbound, because ignoring you is not a verdict on your product. It is usually a verdict on your timing, and timing is the one variable that changes on its own.
So the sequence ends and the watching does not. The signals keep firing. Comments, reactions, a follow-up when something real happens at their company. When they finally enter the 3%, you are not a cold name reappearing after four months. You are the person who has been around, visibly, without asking for anything.
This is what warm outbound actually means in practice. Not a friendlier opener. A longer memory.
Where this goes wrong
Buying more data instead of filtering. More rows do not raise the 3%. They raise the cost of finding it. Fix the filter first.
Counting activity instead of replies. 5,000 sends is not work, it is a number. Split the list and hold volume flat before you buy more capacity.
Treating a signal as a pitch trigger. A signal earns you relevance for one message. It does not earn you the meeting.
Automating the conversation instead of the finding. The finding scales. The conversation does not, and buyers can tell within one sentence which one you automated.
Stopping at the end of the sequence. The 90% is a pipeline you have already paid for. Abandoning it every quarter and re-buying the same names later is the most common unforced error in B2B outbound.
It is not a fancy system. It is the old way of selling with a machine doing the watching, which is the part humans were always bad at anyway. The full version of this, stage by stage, is in signal-based selling: the complete system.
Frequently asked questions
Where does the 3% number come from? From Chet Holmes' buyer pyramid in The Ultimate Sales Machine (2007): about 3% of a market is buying now, about 7% is open to it, and the other 90% is split between people not thinking about it, people who believe they are not interested and people who are sure they are not. It is a rule of thumb, not a measurement of your market. The 95:5 rule from John Dawes at the Ehrenberg-Bass Institute makes the same point with a 5% in-market share. Your exact split moves with deal size and contract cycle. The point is not the decimal, it is that the three groups need three different treatments.
If only 3% are in-market, why keep the other 4,850 on the list? Because membership rotates constantly. The 90% is next quarter's 3%, and you have already paid to build and enrich it. Deleting it means buying the same names again later, colder.
Does one LinkedIn account per ICP not multiply the work? It reduces it. One account per ICP means one voice, one set of proof points, and content that fits every person it reaches. The alternative is one account writing averaged messages for audiences that share nothing.
How many signals should one rep work per day? Fewer than you think. If a rep is working 150 live names, the constraint is research depth per message, not volume. When the qualified queue runs dry, that is information about your real in-market size this month, not a reason to widen the filter.
Can this run on email instead of LinkedIn? The tiering works on any channel. The 90% play does not. Being visible without being in someone's inbox is a social-feed behavior, so email-only teams end up with two tiers and a much shorter memory.