Signal layer before more sequences (L3)
World-class sending machinery in Outreach or Salesloft pointed at an unscored list produces volume, not meetings, because AI makes bad targeting faster rather than better. This converts the ICP into a machine-checkable filter, adds exactly two signals, and caps the queue at 25 accounts per rep per week, the cap being the entire mechanism. Strongest at 100 to 2,000 employees running outbound with a defined ICP, best where the addressable market is under roughly 20,000 accounts.
The steps
- 01
Rewrite the ICP as a machine-checkable filter
Tool: Clay
Convert the ICP paragraph into conditions a machine can evaluate: employee range, industry codes, geography, funding stage, and the two or three technologies whose presence signals readiness. Every condition needs a threshold, not an adjective; "mid-market" is not a filter, "201 to 1,000 employees" is. Then validate it: run the filter and check whether your best 20 current customers pass. If a third of them fail your own ICP, the filter is wrong and you just found out cheaply. • Owner: Marketing plus sales leadership • Tool options: a doc first, then your enrichment provider's filter UI • Pitfall or what breaks: leaving conditions as adjectives instead of thresholds, so the filter cannot actually be evaluated by a machine. • Definition of done: the filter runs against your provider, returns a count you believe, and at least 80% of your best customers pass it.
- 02
Score the database you already have before buying new data
Tool: Clay
Apply the filter to accounts already in the CRM. Report the percentage that fail. Most teams find a third to a half of their own database misses their own ICP, and that single number ends the "we need more leads" conversation permanently. • Owner: RevOps • Tool options: Clay, Clearbit, Cognism, Lusha or ZoomInfo against your CRM • Pitfall or what breaks: buying new lead data before checking whether the existing database even passes the filter. • Definition of done: every account record carries a fit score and a pass or fail flag visible to reps.
- 03
Add exactly two signal sources and define them numerically
Tool: Google Analytics 4
One first-party, one third-party. Define what counts, in writing, with numbers, before you build anything. First-party example: two or more pricing-page views, or three or more docs pages, from a company-matched visitor, within 14 days. Third-party example: category research activity above the provider's own baseline threshold, within 14 days. Both need an expiry; a signal from six weeks ago is history, not intent, and fourteen days is a sane default. • Owner: Marketing ops • Tool options: GA4 or Adobe Analytics for first-party, plus one of G2, TrustRadius or Bombora for third-party • Pitfall or what breaks: adding six intent sources instead of two, so everything scores high and the queue becomes a list with extra steps. • Definition of done: both sources land in the CRM as timestamped events on the account record, with the definition documented and the expiry enforced.
- 04
Build the queue and cap it at 25 accounts per rep per week
Tool: Outreach
Queue membership equals ICP pass AND at least one unexpired signal. Rank by signal recency, then fit score. Cap at 25 accounts per rep per week and let the rest wait. The cap is the mechanism; every rep will ask for more, and the answer is no, because a small queue gets real research per account while an uncapped one gets the same template as the old list. Show the triggering signal on the row, which page, which date; a rep who cannot see why an account surfaced will not trust the queue. • Owner: RevOps plus a GTM engineer • Tool options: Clay or your CRM scoring engine, pushing into Outreach, Salesloft, HubSpot or Apollo • Pitfall or what breaks: lifting the cap in month two because a rep complained, collapsing personalization quality the same week. • Definition of done: each rep sees 25 or fewer ranked accounts, refreshed weekly, with the triggering signal visible.
- 05
Let AI write the middle paragraph only
Tool: Outreach
Three-part structure. Opening line and CTA stay human-written and standardized across the team. AI writes only the middle paragraph, and it gets exactly two inputs: the triggering signal, and one verifiable fact from the account's own website or recent announcement. One hard rule: any output containing no verifiable specific gets dropped rather than sent; "I saw you're scaling your team" is not a specific. Have a manager spot-check 20 before you scale it. Never open by claiming the prospect lacks a technology, since enrichment providers under-detect badly and being wrong in the first line ends the conversation. • Owner: Enablement • Tool options: the native LLM in Outreach or Salesloft, or Clay's enrichment-to-copy step • Pitfall or what breaks: opening with "I noticed you don't use X," invented by the provider's blind spots. • Definition of done: 100 messages have gone out under this rule and a manager has verified 20 contain a real, checkable specific.
- 06
Compare signaled versus unsignaled, then kill the losers
Tool: Looker
After 30 days compare reply rate and meetings booked per 100 touches for signal-triggered versus list-based outreach. Then actually turn off the list-based sequences that lose, and redeploy the reclaimed sending capacity into the queue rather than adding new volume on top. Reclaiming capacity is what makes this free. • Owner: VP Marketing • Tool options: Outreach or Salesloft reporting, charted in Looker, Tableau or Power BI • Pitfall or what breaks: adding new volume on top instead of redeploying reclaimed capacity into the queue. • Definition of done: at least one legacy sequence is off and its volume is redeployed.
- 07
Instrument the result
Track meetings booked per 100 touches, split signaled versus unsignaled. That one comparison funds or kills the program. • Where it breaks: six intent sources get added and the queue reaches 400 accounts, becoming a list with extra steps; the cap gets lifted in month two and personalization quality collapses; or an email opens with an invented claim from a provider's blind spot. • Visual guidance: two funnels side by side with identical top volume and different conversion at every step. Left is "list," right is "signal." One chart, one argument, no explanation needed.
Tools in this playbook
Next playbooks
Unfamiliar terms are defined in the AI and Revenue Dictionary. Related frameworks live in the framework library.