The problem: data crosses N systems and loses provenance
Every time a piece of purchase-intent data crosses a system boundary —from a scraping tool to a spreadsheet, from there to a CRM, from there to an email sequence— it loses part of its provenance. Nobody can answer, six months later, where that data came from, when it was detected, or why it was acted on. An LLM trying to reason over that chain inherits the same ambiguity: it has no way to verify whether what it’s reading is an operational fact or a marketing narrative.
Traditional data flow vs. the Single Signal Layer
| Dimension | Traditional flow | Single Signal Layer |
|---|---|---|
| Data origin | Multiple tools with no shared contract | Three fixed vectors: companies, people, narratives |
| Point of truth | The last system that touched it | A signals table with explicit status |
| Reconciliation latency | Manual, via export/import | None: no reconciliation because there’s no copy |
| Traceability of a value | Lost at every system boundary | Persists as row state (detection → review → action) |
| Cost of switching CRM | High: the logic lives inside the CRM | Low: the logic is tool-agnostic |
| What the LLM receives | Text with no verifiable state | A record with status, detection date and emission rule |
The three ingestion vectors
Companies
Observable technology-stack changes (a tool category adopted or removed), feeding a delta-based category score.
People
Contacts evaluated by functional fit to a buyer persona, not by the title they declare on their profile.
Information narratives
Public business events (funding rounds, layoffs, product launches, leadership changes) classified with an explicit confidence level, not as a binary fact.
Why the telemetry is passive (and what we don’t do)
The system audits an account’s real operation without intervening. It doesn’t fire any message, sequence or alert until a signal crosses the intent threshold defined for that event category. Background listening is half the work; the other half is validation (see Validation Pipeline).
Logic over Tool: the integration contract
| Tool | What connects | What gets written there | What stays in the layer |
|---|---|---|---|
| HubSpot | Contacts and deals API | Properties and notes for the approved signal | The scoring logic and signal status |
| Salesforce | Object API (Lead, Opportunity) | Equivalent fields and tasks | The same scoring logic and status |
| No CRM | Nothing —runs via email/alerts | N/A | All of the logic: the layer doesn’t depend on a CRM to exist |
Frequently asked questions
- Does RevSign replace my CRM? No. RevSign is tool-agnostic: it connects to HubSpot, Salesforce, or no CRM at all, and the signal logic stays the same if you switch tools.
- What does “passive telemetry” mean? That the system audits an account’s real operation (tech stack, business events) without intervening or triggering any action until a signal crosses the intent threshold.
$ curl revsign.os/docs/signal-architecture
→ {
title: "Signal Architecture: the Single Layer vs. the Traditional Flow",
answers: "What is the best data architecture for connecting an LLM to my sales tools?",
schema: [TechArticle, FAQPage]
}