The honest answer to how accurate a sales forecast should be: the plus or minus 10 percent that is often quoted depends on when in the period the forecast was made, whether the miss is random or always in one direction, and whether accuracy is measured on the total or per rep. This page gives the ranges by horizon, the three measurable things that set the right figure for one team, the horizon, the bias, and the per-rep spread, and the table to compute before anyone quotes a percentage.
Forecast accuracy is one minus the absolute error over the actual, and the question is what it should be. The answer depends on when, which direction, and for whom.
| Forecast made | Typical error, absolute | Good |
|---|---|---|
| Start of quarter | 15 to 25 percent | Under 15 |
| Mid-quarter | 8 to 15 percent | Under 10 |
| Two weeks before end | 3 to 8 percent | Under 5 |
| Final week | 1 to 4 percent | Under 3 |
Desks with long cycles and few large deals sit at the wide end. A lumpy enterprise team within 15 at the start of the quarter is doing well; a transactional team should be within 8.
Error at horizon h = (forecast made h weeks before period end − actual) ÷ actual
This needs the forecast saved every week, not overwritten. Most CRMs keep only the current number, so the first job is a weekly snapshot of the forecast per rep.
| Weeks before end | Forecast | Actual | Error |
|---|---|---|---|
| 12 | $2,400,000 | $2,100,000 | +14% |
| 8 | $2,300,000 | $2,100,000 | +10% |
| 4 | $2,200,000 | $2,100,000 | +5% |
| 1 | $2,150,000 | $2,100,000 | +2% |
The curve is the team's own benchmark. Next quarter's curve against it is the read.
Bias = mean of signed error over the last four quarters, per rep, at a fixed horizon
| Rep | Q1 | Q2 | Q3 | Q4 | Bias | Mean absolute error |
|---|---|---|---|---|---|---|
| A | −12% | −9% | −15% | −11% | −12% | 12% |
| B | +18% | +22% | +15% | +20% | +19% | 19% |
| C | +6% | −8% | +4% | −5% | −1% | 6% |
| Team | +3% | +1% | +1% | +2% | +2% | 2% |
The team is within 2 percent every quarter. A sandbags by a tenth; B is optimistic by a fifth; they cancel. C is the only accurate forecaster. The forecast bias worked example computes this table by hand, and the direction piece covers what to do with A and B.
The team error is the sum of signed errors; the spread is the mean of absolute errors per rep. Two percent and twelve percent are both the team's forecast accuracy, and the second is the honest one.
| Measure | Formula | From |
|---|---|---|
| Weekly forecast snapshot | Per rep, per week, saved | CRM, exported weekly |
| Error by horizon | (Forecast at h − actual) ÷ actual | Snapshots and closed deals |
| Bias per rep | Mean signed error, four quarters, fixed horizon | Same |
| Mean absolute error per rep | Mean of absolute errors | Same |
| Team spread | Mean of per-rep absolute errors | Same |
| Horizon curve vs own prior year | Same curve, prior four quarters | Same |
No snapshots. The forecast is always the current number, and accuracy is measured on the day the quarter closes.
Horizon unstated. Ninety-five percent accurate, the day before the end.
Team total only. Two percent, from two reps who cancel each other.
Error chased, bias ignored. Training on forecasting technique for a rep who is sandbagging deliberately.
A good forecast accuracy is one measured from a saved snapshot at a stated horizon, within the team's own prior curve, with per-rep bias near zero across four quarters and a per-rep spread that is narrowing. Within 15 at the start of the quarter and 5 near the end is a fair expectation for most B2B teams; the bias table is where the improvement comes from. Covirage snapshots the forecast weekly from the CRM export and produces the horizon, bias and spread tables each quarter.
Roughly within 15 to 20 percent for a team with a normal sales cycle, tightening to 10 by mid-quarter and 5 in the last fortnight. A team that is within 5 at the start is either very predictable or forecasting the number it was given. The horizon curve for the team's own history is the benchmark.
Bias is the more useful finding. Random error is the sales cycle being uncertain. Bias is a rep or a team systematically low, to be safe, or systematically high, to be optimistic, and it is correctable per person once it is shown across four quarters. Most forecast accuracy programmes fail because they chase error and never separate out bias.
Both, and the per-rep view is where the action is. Team accuracy is what the board sees; the per-rep bias table is what the sales leader uses on Monday. Offsetting errors at the team level are the normal way a bad forecast looks good.