Sign in

Blog · Forecast and pipeline

Forecast bias on five reps and four quarters: the whole arithmetic on one page

The complete forecast bias and error calculation on five reps over four quarters, small enough to check by hand: each rep's week-six forecast and closed result per quarter, the signed miss, error as the mean absolute miss, bias as the mean signed miss, the four kinds of rep, the adjustment for the predictably-high rep with its range, the reps with too few quarters to adjust, and the identity that the adjusted total equals the raw total plus the adjustments, so a reader can reproduce every figure and then run it on their own snapshots.

The short answerFive reps, four quarters, each with a forecast made at week six and the closed result. The signed miss is forecast less actual over actual; error is the mean of its absolute value; bias is the mean of the signed value. One rep is high every quarter by about 17 percent: low error relative to bias, predictably wrong, adjustable. One misses by 15 percent in alternating directions: high error, near-zero bias, not adjustable. One has three quarters and is shown unadjusted. The adjusted forecast for this quarter is the raw forecast plus each adjusted rep's bias correction, and the identity is that the two totals differ by exactly the sum of the adjustments. Every number can be reproduced by hand.

Forecast bias is a mean of signed misses per rep, and on five reps and four quarters it can be computed by hand. This page works error and bias, the four kinds of rep, the adjustment with its range, the reps too young to adjust, and the identity.

The forecasts at week six and the results

Rep Q1 fcst / actual Q2 Q3 Q4
A 1.20 / 1.02 1.10 / 0.95 1.30 / 1.10 1.15 / 1.00
B 0.90 / 0.92 1.00 / 0.85 0.85 / 1.00 0.95 / 0.82
C 0.80 / 0.78 0.85 / 0.84 0.75 / 0.72 0.90 / 0.86
D 1.00 / 0.80 1.10 / 0.90 0.95 / 0.80 1.05 / 0.85
E — 0.70 / 0.66 0.75 / 0.69 0.80 / 0.72

Values in millions of dollars. E joined in Q2.

The signed miss per quarter

Miss = (forecast − actual) ÷ actual

Rep Q1 Q2 Q3 Q4
A +17.6% +15.8% +18.2% +15.0%
B −2.2% +17.6% −15.0% +15.9%
C +2.6% +1.2% +4.2% +4.7%
D +25.0% +22.2% +18.8% +23.5%
E — +6.1% +8.7% +11.1%

Error and bias

Error = mean of the absolute misses; bias = mean of the signed misses

Rep Quarters Error Bias Kind
A 4 16.7% +16.7% Low error relative to bias: predictably high; adjust
B 4 12.7% +4.1% High error, low bias: unpredictable; do not adjust
C 4 3.2% +3.2% Accurate; as given
D 4 22.4% +22.4% Predictably very high; adjust, and coach
E 3 8.6% +8.6% Too few quarters; shown, not adjusted

The adjustment, with its range

Adjustment = −bias, applied to this quarter's raw forecast, with the range from the four misses.

Rep This quarter's raw Bias Adjustment Adjusted Range of misses
A $1.25m +16.7% −$0.18m $1.07m +15.0 to +18.2
D $1.00m +22.4% −$0.18m $0.82m +18.8 to +25.0
B, C, E $2.55m none $2.55m
Total $4.80m −$0.36m $4.44m

Adjustment for A: 1.25 ÷ 1.167 = 1.07, a haircut of $0.18m. For D: 1.00 ÷ 1.224 = 0.82, $0.18m. Total adjustments $0.36m.

The identity

adjusted total = raw total + Σ adjustments = 4.80 − 0.36 = $4.44m

Every dollar of the difference is attributed to a named rep.

Where it goes wrong, even at five

Misses measured at different weeks. A's Q3 forecast taken at week twelve reads +2 percent and the bias collapses.

Error read without bias. B and A both miss by about 15 percent; one is adjustable and one is not.

E adjusted on three quarters. A nine percent haircut on a bias that may not hold.

Adjustment applied silently. The leader sees $4.44m and does not know A and D were cut.

From five to fifty

The same table per rep from the weekly forecast exports and the closed results. Covirage runs it every quarter at the stated week. The forecast bias guide covers the measure, and the forecasting methods guide covers where the adjusted roll-up sits among the methods.

Questions people ask

Why week six?

Because a forecast at week twelve is nearly always right and says nothing about the rep. A fixed week every quarter, at the same distance from the end, is what makes the four quarters comparable. Six of thirteen is a common choice and it is stated.

How many quarters before adjusting?

Four is the floor for a stable bias. Rep E has three and is shown with the bias and the range but not adjusted. The count is on the row, and the rule is stated.

Is the adjustment applied automatically?

It is computed and shown beside the raw forecast. The sales leader commits one of them. On this page the adjusted total is $260,000 below the raw, and the two reps whose adjustments make up that difference are named.