Customer Signals by WarmStart

Evidence-backed customer development

Your best product conversations are already hiding in your usage data.

Customer Signals reads the behavior your users have already shown, then tells you who may be worth interviewing, cultivating as an evangelist, or inviting closer to the product.

  • Read-only connections
  • Unknown never means zero
  • No outreach without approval
Product eventsRepeat behaviorPurchasesReferralsSupportConsent

A diagnosis before a score

First, find out what your data can honestly support.

Most tools rank people as soon as they find a row. Customer Signals first checks identity coverage, event meaning, exclusions, freshness, and contact permission. If the evidence is not ready, it says so.

This worked example runs locally in your browser. It does not inspect your device, URL, or accounts.

customer-signals / example-project ready to scan
Data readinessRun the sample to inspect 19 evidence checks.
Signals measuredmissing remains unknown
Identity coveragepseudonymous first
  • Durable identityEvents resolve to a stable, pseudonymous subject.
    verified
  • Product scopeStaff, tests, and unrelated product rows are excluded.
    verified
  • Event semanticsActivation and success events have reviewed meanings.
    verified
  • ?
    Support evidenceNo scoped support source is connected yet.
    unknown

From events to a useful conversation

Three layers. Each one earns the next.

01

Diagnose the data

Check freshness, identity coverage, event meaning, exclusions, consent, and missing sources before interpreting behavior.

Output · readiness brief
02

Find product pull

Compare activation, repeat use, depth, purchases, referrals, and support evidence without pretending correlation is causation.

Output · opportunity profile
03

Review people safely

Score each role independently, show counterevidence, and keep identity and outreach behind explicit human approval.

Output · candidate review set

One user can mean different things

Role-specific evidence, not one magic score.

An evangelist is not automatically an influencer. A deeply engaged user is not automatically a product owner. Customer Signals evaluates each path separately and keeps the evidence attached.

Evangelist

Repeat value plus voluntary advocacy.

Influencer

Credible reach plus authentic product use.

Customer interview

Recent, relevant behavior and permission to learn.

Potential product owner

Broad, sustained use and product-shaped judgment.

Built to stop before it guesses

Useful customer intelligence without turning people into targets.

Customer Signals separates behavioral evidence from identity and contact actions. The analysis can tell you there is a promising cohort before anyone is named or messaged.

  • Read only

    Scoped database roles and aggregate endpoints. No product writes.

  • Pseudonymous first

    Candidate reports use project-scoped subject references, not emails or names.

  • Fail closed

    Missing, stale, or semantically unclear data blocks conclusions instead of becoming zero.

  • Human gate

    Identity review and outreach remain separate, explicit actions.

agent instructions
# Install the interpretation contract
https://brain.warmstart.io/skills/
customer-signals/SKILL.md

# Discover the read-only MCP server
https://brain.warmstart.io/.well-known/
mcp.json

The same guardrails for agents

Give your agent the diagnosis, not your customer database.

The public skill teaches agents how to preserve unknowns, compare roles independently, and stop at identity or outreach boundaries. The authenticated MCP server is read-only and exposes published reports, never provider credentials.