Signals
A Practical Account Scoring Model for GTM Engineers
A transparent scoring framework that separates fit, timing, evidence confidence, and actionability so teams can prioritize accounts without creating a black box.
Most account scores fail for one of two reasons: they combine unrelated ideas into a number nobody can explain, or they assign points to every available field until almost every account looks important. A useful model should help a team make a specific decision and make disagreement easy to diagnose.
Use four components
- Fit: the structural likelihood that the account can receive meaningful value.
- Timing: evidence that a relevant condition changed recently.
- Confidence: reliability of the identity, source, and interpretation.
- Actionability: whether the team has a sensible action, owner, and channel now.
Fit should change slowly
Fit usually comes from durable company or customer characteristics: segment, business model, geography, use case, technical environment, operating complexity, or known disqualifiers. Keep the set small enough to defend. A fit score should not jump because someone visited a webpage yesterday.
Timing should decay
Timing is event-driven and temporary. A leadership change, new initiative, product milestone, hiring pattern, integration event, or relevant engagement may change priority for a defined window. Every timing feature needs a half-life or expiration rule; otherwise the score accumulates stale urgency.
Confidence should be multiplicative
Treat confidence as a modifier rather than free points. A strong event connected to the wrong account should not outrank a modest verified event. One practical approach is to calculate a base priority from fit and timing, then multiply it by source and identity confidence.
Priority = (fit × fit weight + timing × timing weight) × confidence.
The exact formula matters less than the behavior: uncertain evidence should reduce priority, and the reason should remain inspectable.
Actionability should gate routing
An account can be high-fit and well-timed while still lacking a useful action. Perhaps ownership is unclear, the relevant persona is missing, an opportunity is already active, or policy suppresses outreach. Preserve the insight for analysis, but do not force it into an execution queue.
Return reason codes with the score
- High fit: matches the target operating profile.
- Timing event: verified relevant change within the last defined window.
- Confidence: direct source and verified account match.
- Constraint: no open opportunity, active customer motion, or recent contact collision.
- Recommended action: owner and response pattern defined.
Calibrate before automating
Sample accounts across score bands and ask experienced operators to grade both priority and explanation quality. Compare model rank with their judgment, then observe actual acceptance and outcomes. Tune thresholds and features; do not repeatedly add points to exceptions. If the model needs dozens of special cases, the decision definition is probably unstable.
Monitor the distribution
Watch score distribution, route volume, acceptance rate, false-positive reasons, time to action, and outcome rate by band. A scoring model is not done when the formula ships. It is production software whose inputs and operating environment will change.