Review the ICP using wins, losses and exclusions

When to use it

An ICP review or a repeated pattern of qualification errors starts this workflow. Challenge company-fit rules against mature outcomes and sales context before expanding the market model. The intended output is an icp change proposal with examples, limitations and tier movements. A useful receipt names the accountable person and the next action; it should make the commercial decision easier to inspect.

This is an original GTMhub reference design. It explains how we would run the job, with tools selected for the company's existing stack. It is not presented as completed client delivery or measured commercial performance.

Inputs and decision boundaries

Confirm historical account facts, mature outcomes, exclusions and sales coverage. Keep the original source and observed date alongside each fact. Missing context stays unknown until it is resolved. Facts, interpretations and proposed actions should remain distinguishable in the handoff.

Your company brain holds the approved rules and proof; the live system holds current account state and ownership. Read both before making an operational change. Assign an exception owner before launch so an uncertain case has somewhere useful to go.

Run the job

1. Recover pre-outcome company features where possible.
2. Compare won, lost and disqualified accounts by segment.
3. Investigate selection bias and missed-fit examples with sales.
4. Propose a readable versioned rule with a holdout or prospective test.

At the decision gate, ask: Is the pattern mature and useful enough to change selection? If the answer is no: keep it as a hypothesis and gather the missing evidence. Only the supported path proceeds to its output. A stopped job should preserve the reason and what would let an owner resolve it.

The output to inspect

Failure and recovery

The material failure to watch is this: a model learns from post-sale facts or only from companies the team already favored. Stop that path and preserve the underlying evidence. Repairing a field or a source is different from changing the commercial rule; record which decision was made.

If a connected system fails, keep the original request or event and expose its unresolved state. Retry only the incomplete action, with a stable event key where writes are involved. Do not let a retry create a second owner task or bypass a previous exclusion.

Test before making it live

Replay a supported case, an incomplete case and the failure described above. Where the job writes a record or creates a task, replay the same event twice and inspect the result. Check that the exception owner can understand the hold reason without reconstructing every step.

Judge the result

Track qualified-account selection and missed opportunities, with mature sample counts. Record the sample, dates and definitions before evaluating a change. Operational success shows that the job behaved as designed; commercial outcomes need their own mature evidence.

Return reviewed corrections to the company brain with the decision date and rule version. Keep the original observation available. The decision register records what changed and why, while the GTM brain guide explains how the job fits into one company system.