Why bias matters more than variance
A team that misses by 12% every quarter and a team that misses by ±2% with a slight positive bias look different on the headline MAPE — but the second team is forecasting well; the first is guessing. Bias (the average direction of miss across multiple periods) is the real diagnostic.
How to read the three results
- Absolute variance — useful for board updates ("we missed by $80k"). Less useful for trend.
- Variance % — useful for comparing across periods of different sizes. Negative = miss.
- MAPE proxy — magnitude only, no direction. Best for benchmarking against industry standards.
What "good" looks like by stage
- Pre-revenue / series A: ±20% is normal. Forecasts are based on pipeline that hasn't matured into a stable distribution.
- Series B–C: ±10% is the bar. Below 10% MAPE indicates the team has built a real model, not a wishful one.
- Public / late stage: ±2–5%. Investors penalize anything outside this band aggressively.
Single-period vs rolling
One quarter's variance can be noise. Rolling 4-quarter MAPE is what you should track on the operating cadence — it smooths out a single bad quarter and reveals whether the forecasting machinery has actually improved.