the GTM system · The Sales Stack · the signal library

Every signal we score on - and every signal we read for context

One canonical list. Signals do two different jobs, so we keep two layers: a lean Score layer that moves the number and a greedy Context layer the AI reads to write the brief. This page is the source of truth for both - what the signal is, where it comes from, what job it does, and which channel it serves.

The rule that changed everything: buying signals are additive, never a leg that can collapse to zero. Every in-ICP account starts with a real base. Product usage is the loud engine. Buying signals only ever add points - their absence is never a penalty. So a strong-fit account in our product ranks high before a single third-party signal shows up, and nothing scores zero on signals again.

How the score is built now - additive, not a weighted average

ICP qualifies and sets the floor. Product is the loudest mover. Signals are bounded boosts on top.

The old model blended Product, ICP and Signals as three co-equal legs - so an empty signals leg craved 30% of the score and buried good accounts. The new model is layered: fit gets you a base, product does the heavy lifting, and every buying signal is a capped add-on that can only help.

Gate + Base

ICP

Fit qualifies and sets a 40-70 base. Below the floor → park/DQ. Never starts at 0.

Engine (loudest)

Product

Adoption lift up to +30. Sequence activation is the top driver. Absent for non-product leads → +0, no penalty.

Boosts (capped)

Signals

Each buying signal adds points, total capped ~+20. Absence = +0.

Final = min(100, ICP base + Product engine + Σ signal boosts) × email gate · Octave hard-disqualify still caps at 40.

Why product is louder than any buying signal

The single biggest product boost (Power adoption, +30) dwarfs the single biggest buying-signal boost (~+10). An account already using us ranks high on its own behavior - exactly as it should. A third-party signal sharpens the order; it never gatekeeps.

The Score layer - what moves the number

Lean and defensible. These are the only inputs that change the score, so each one has a job, a magnitude, and a source. score = scoring input · both = also feeds context.

Signal	Source	Role	Magnitude	Channel
ICP fit	Octave (0-10)	Gate + base	40-70 base	All	both
Sequence activation - the loudest	Amplitude	Product engine	up to +30	Product	both
AI-generated sequences · breadth · 12-wk trend	Amplitude	Product engine	in engine	Product	both
New active users (14/30d) · seat growth	Amplitude	Boost (timing)	+ boost	Product	both
Pricing / cart views	Amplitude	Boost (intent)	+ boost	Product	score
Inbound hand-raise (demo / form fill)	SFDC / Marketing	Boost (loud for inbound)	+ boost	Inbound	both
New VP Sales / CRO hire	PredictLeads	Trigger boost	~+6	All	both
Funding round	PredictLeads / Crustdata	Trigger boost	~+6	All	both
Competitor tool in stack (displacement)	TheirStack	Boost	~+5	All	both
Hiring sales roles + velocity	TheirStack / Crustdata	Boost	~+4	All	both
CRM in stack (1.64× win-lift)	BuiltWith + Amplitude integrations	Boost	~+3	All	both
Decision-maker found + title	FullEnrich	Boost (actionability)	~+2	All	both

Magnitudes are the design intent; final boost values get tuned against won/lost in the re-run. The cap (~+20 total) keeps any one account from stacking boosts into a false Tier 1.

The Context layer - what the AI reads to write the brief

Greedy on purpose. Context never has to earn its weight in the score - more is always better for a richer narrative. We pull it once, store it, and the AI reads it instead of re-querying six providers every time. context = narrative enrichment.

Signal	Source	What it adds to the narrative
Capability breadth · power users · WAU	Amplitude	How deep and sticky the deployment is; who the champions are
Confirmed integrations (SFDC / HubSpot / Zoom)	Amplitude	First-party technographic - stronger than an inferred guess; integration talking points
Recent news / press	DataForSEO · Serper · Exa	A timely hook for the opener
Funding history · headcount by department	Crustdata	Sizing, budget signal, and which teams are whitespace
Leadership bios · new executives	PredictLeads	Who to reference and who the new buyer is
Competitor mentions · G2 review activity	G2 · Slack	The displacement angle - what they're comparing us against
Full tech-stack detail	BuiltWith · TheirStack	Integration fit and tooling maturity

One score, all of new business - inbound, outbound, product

The additive base is what makes this work across channels. A lead that isn't in the product yet simply scores +0 on the product engine - no penalty - and rides ICP fit plus whatever signal its channel can produce. The same scale ranks all three head-to-head on one list.

	Product lead	Inbound lead	Outbound lead
ICP base	✓	✓	✓
Loudest available signal	Sequence activation	Hand-raise (demo / cart)	Trigger event (funding, new VP, hiring)
Product engine	Big lift	usually +0	+0
Worst-case floor	ICP base	ICP base	ICP base

A sequence-activating product lead and a freshly-funded outbound account compete on the same composite - each scored on what it can actually show.

Where it lives in Salesforce - ~9 fields, not 50

The split maps cleanly to two field types. The score goes in small structured fields you can filter, route, and report on. The context goes in a few long-text blobs the AI reads - pulled once, not re-queried every time. We theme the context into three so a product change rewrites only the product blob (fewer needless writes) and a brief reads only the field it needs (fewer tokens).

Score layer - structured & filterable

Composite_Score__cNumber · the 0-100 score, drives routing & tiering

Score_Tier__cPicklist · Strike / Hot / Warm / Watch / Low / DQ

Score_Product__c · Score_ICP__c · Score_Signals__cNumber · the three components, for transparency & QA

Score_Context__cLong Text · the rep-facing 4-block summary (already live)

Context layer - AI-read, long text

AI_Context_Company__cLong Text Area · firmographics, funding, news, leadership

AI_Context_Product__cLong Text Area · adoption, capabilities, integrations, champions

AI_Context_Signals__cLong Text Area · triggers, hiring, displacement, intent

Field limits worth knowing

A Long Text Area holds up to 131,072 characters - capacity is never the issue; keep each blob to a ~10k budget so the AI read stays cheap. Long-text fields are not filterable or sortable in SOQL and don't show in list views, which is exactly why the score stays in number/picklist fields and the context stays in blobs. Use Long Text Area, not Rich Text (Rich Text wastes characters on markup). The Salesforce MCP is data-plane only - these three AI_Context_* fields are a one-time create for RevOps; everything populates the moment they exist.

What we removed and changed

Decision	Why
Common Room - removed from the stack	It fed one intent input and nothing else we used. Deepline now covers triggers, hiring and technographics; Amplitude covers product. Not worth the spend.
Free signups - not a buying signal	We engage once they're activated, not when they sign up. Sequence activation is the loudest product signal - strongest payment predictor in the funnel.
Signals can never be 0	ICP base + product engine floor every account. A missing signal stops being a penalty and becomes simply "no boost."
Microsoft / Outlook intentionally demoted	The email gate keeps its ×0.55 on Outlook on purpose - we're not investing in that surface as a company.

The takeaway

Two layers, one source of truth. A lean score you can trust and route on; a rich context the AI reads to write the brief. Buying signals only ever add, product is the loud engine, and the same score runs across inbound, outbound, and product leads.