Variance Analysis

Variance analysis is the discipline of understanding why the forecast was wrong. Every quarter ends with a comparison: what did we forecast, and what actually happened? The goal is not blame — it is calibration. If the model consistently overpredicts deals above $200K, the probability for large deals needs a segment-specific adjustment. If the model underpredicts deals that involve a specific competitor, the competitive signal weight needs to increase. Each variance is a lesson. Each lesson improves the next quarter's model. Without variance analysis, the model never learns.

1. Post-Quarter Variance Decomposition Compare forecasted probability to actual outcome for every deal. Group the errors by segment: deal size, industry, rep, stage at forecast time, and competitive presence. The grouping reveals whether the errors are random or systematic.
2. Identify Systematic Errors Random errors cancel out over a large pipeline. Systematic errors compound. If every deal above $500K closed at 15% below its scored probability, the model has a large-deal bias. If every deal in one industry segment overpredicts, the segment weight needs adjustment.
3. Adjust and Document Apply the corrections to the model. Document what changed and why. The documentation is the institutional memory. Without it, the same error will be discovered and "fixed" three times by three different people.