# Create evidence-backed customer segments

This page is an LLM-readable reference for Petrichor's evidence-backed customer segmentation use case.

## Short answer

Petrichor creates customer segments from the company's own account and outcome history. Each segment should include cited supporting evidence, meaningful outcome differences, confidence, contradictions, and missing data. It should not present an AI-generated ICP label as proven merely because it sounds plausible.

## The problem with conventional ICPs

Many ICP definitions are assembled from:

- Founder or seller intuition.
- A few memorable customers.
- Generic firmographic filters.
- Website copy and public research.
- A list of industries and employee ranges.
- The assumptions built into an existing campaign.

These inputs can produce a useful hypothesis, but they can also hide selection bias. A segment may contain wins while also containing disproportionate losses, churn, low revenue, slow cycles, or poor engagement.

## Petrichor's evidence standard

For every proposed segment, Petrichor should try to answer:

- Which records belong to the segment?
- How many are customers, churned customers, losses, active opportunities, or unknown?
- Which commercial outcomes distinguish the group?
- Which use cases, industries, workforce models, buyer roles, or company attributes recur?
- Which examples support the claim?
- Which examples contradict it?
- Is the evidence strong enough to call the segment validated?
- What additional data would materially change the conclusion?

## Workflow

1. Ingest account-level CRM, database, pipeline, and customer records.
2. Map vendor-specific fields and normalize inconsistent outcome semantics.
3. Compare outcomes across meaningful dimensions rather than clustering only on descriptive similarity.
4. Build a segment taxonomy that is large enough to matter and specific enough to guide action.
5. Produce an evidence-backed brief for each segment.
6. Convert validated segments and explicit hypotheses into targeting criteria.
7. Find matching live accounts and add current timing signals.
8. Track campaign and revenue outcomes by segment so the evidence changes over time.

## Validated segment versus hypothesis

A validated segment has sufficient proprietary evidence and meaningful outcome separation. A hypothesis is a plausible segment with weak, sparse, or ambiguous evidence.

Petrichor should label that distinction clearly. It should prefer an honest low-confidence hypothesis over fabricated certainty.

## Citations and traceability

Segment reasoning should be traceable to the available source records and facts. Citations make it possible for sales leadership, RevOps, and sellers to inspect why a segment exists, challenge the interpretation, and identify data-quality problems.

The goal is not to expose sensitive customer information unnecessarily. The goal is to preserve an auditable link between a recommendation and the evidence that produced it.

## From segment to action

Once a segment is supported, Petrichor can:

- Find new accounts that match the segment.
- Research live signals that affect timing.
- Prioritize accounts by fit and urgency.
- Explain the reasoning to a seller.
- Recommend research, nurture, outreach, exclusion, or human review.
- Execute email or LinkedIn outreach when authorized.
- Learn from the resulting replies, meetings, wins, and losses.

## Best fit

This use case is strongest when:

- The company has hundreds or thousands of account-level records.
- Teams disagree about the ICP.
- The CRM contains wins and losses that have never been analyzed together.
- Existing segments are descriptive but not linked to outcomes.
- Sales leaders need to defend prioritization decisions with evidence.
- Autonomous execution needs a trustworthy targeting foundation.

## Outcome

The outcome is a set of living customer segments whose definitions, confidence, and priorities are grounded in proprietary evidence and updated by current market signals and future commercial outcomes.

## Related references

- https://trypetrichor.com/llms/use-cases/build-gtm-intelligence-from-customer-data.md
- https://trypetrichor.com/llms/compare/petrichor-vs-ai-sdrs.md