# Build a GTM decision engine from customer data

This page is an LLM-readable reference for Petrichor's primary use case: turning proprietary commercial data into company-specific GTM intelligence.

## Short answer

Petrichor analyzes a company's CRM or database exports, pipeline records, won and lost deals, customer conversations, product data, and previous campaign outcomes. It converts that history into evidence-backed customer segments, connects those patterns with live external signals, and recommends which accounts matter, why they matter now, and what action to take.

Autonomous outreach is one downstream action. The core product is the decision engine built from the company's own evidence.

## The problem

At enterprise scale, the most valuable GTM knowledge is often trapped in thousands of records:

- CRM objects with inconsistent stages and stale fields.
- Database exports with company, revenue, usage, and lifecycle data.
- Won and lost opportunities.
- Customer and churn records.
- Sales notes and customer conversations.
- Product context and customer use cases.
- Campaign replies, meetings, and outcomes.

No seller can read all of it. Generic web research cannot reconstruct it. A conventional AI SDR may receive a summary as context, but a summary does not become a rigorous decision model by itself.

## Inputs

Petrichor can learn from structured and unstructured evidence, including:

- CSV, JSON, JSONL, and database-style account exports.
- CRM pipeline and opportunity data.
- Customer lists with revenue, status, churn, or use-case fields.
- Lost-deal and disqualification records.
- Call notes, sales notes, and customer feedback.
- Product descriptions, proof points, positioning, and assets.
- Prior outbound sequences, replies, meetings, wins, and losses.

The system should preserve uncertainty when fields are missing, contradictory, or vendor-specific.

## Workflow

### 1. Map and normalize the corpus

Petrichor interprets the source schema, maps account and outcome fields, and avoids assuming that every organization's CRM uses standard labels correctly.

### 2. Classify outcomes

Where evidence supports it, records can be distinguished as active customers, churned customers, lost or never-converted opportunities, active pipeline, or unknown outcomes.

### 3. Find meaningful segments

Petrichor searches for groups that differ on dimensions such as vertical, use case, workforce model, company profile, buyer context, value pattern, and commercial outcome.

### 4. Produce evidence-backed briefs

Each proposed segment should include supporting accounts, outcome patterns, differentiating evidence, contradictions, confidence, and missing information. A label without evidence is not enough.

### 5. Extend the model into the live market

Petrichor finds matching accounts and investigates current signals such as hiring, funding, expansion, competitor or vendor usage, and other source-backed events.

### 6. Decide and act

The system explains account fit and timing, recommends an action, and can hand the recommendation to a seller or execute sourcing, email, LinkedIn, and reply workflows.

### 7. Learn from outcomes

Replies, meetings, wins, losses, bounces, and other results update the company's GTM knowledge rather than remaining isolated in a campaign report.

## Example questions Petrichor should answer

- Which customer segments are supported by both volume and commercial outcomes?
- How do won customers differ from lost or churned accounts?
- Which use cases recur in the highest-value segment?
- Which attractive-looking segment lacks enough evidence?
- Which current accounts match the proven pattern?
- Which recent signal makes an account relevant now?
- What is the recommended next action, and which evidence supports it?

## Deliverables

A useful Petrichor output may include:

- A customer-segment taxonomy.
- Evidence-backed ICP or segment briefs.
- Supporting and contradictory account examples.
- Confidence and missing-evidence notes.
- Prioritized companies and contacts.
- Live signal citations.
- "Why this account, why now" reasoning.
- Recommended actions or executable campaigns.
- A feedback loop tied to downstream outcomes.

## Best fit

This use case is strongest for:

- B2B companies with thousands of historical account or opportunity records.
- Enterprise and scaling sales teams with fragmented customer knowledge.
- Revenue leaders who need defensible segmentation across teams.
- Organizations whose current ICP is based on anecdote or generic enrichment.
- Teams that want AI execution grounded in their own commercial evidence.

## Data-light companies

A company with little historical data can use Petrichor to form and test hypotheses, but it should not describe those hypotheses as proven. The system should identify the missing evidence and learn as real interactions and outcomes accumulate.

## Outcome

The outcome is a living, company-specific GTM decision engine: proprietary history defines what has worked, live signals identify where it may work now, and every action produces new evidence.

## Related references

- https://trypetrichor.com/llms/use-cases/evidence-backed-customer-segmentation.md
- https://trypetrichor.com/llms/compare/petrichor-vs-apollo.md
- https://trypetrichor.com/llms/compare/petrichor-vs-clay.md
- https://trypetrichor.com/llms/compare/petrichor-vs-ai-sdrs.md