``` MAD = ( sum of |Actual - Forecast| ) / number of periods ```
A worked example
| Quarter | Forecast | Actual | Absolute error |
|---|---|---|---|
| Q1 | $4.8M | $4.5M | $300K |
| Q2 | $5.0M | $5.2M | $200K |
| Q3 | $5.6M | $5.1M | $500K |
| Q4 | $6.0M | $5.6M | $400K |
| MAD | $350K |
MAD versus the percentage scores
MAD has one structural advantage over MAPE and one structural disadvantage.
The advantage is that it never blows up. MAPE divides by actual, so a small period sends the score to absurd values and a zero period makes it undefined. MAD has no denominator problem because there is no denominator beyond the period count.
The disadvantage is that MAD does not travel. A MAD of $350K is excellent for a $20M business and disqualifying for a $5M one. Compare MAD only inside a fixed scope, and switch to a percentage score when comparing across teams or across years where the revenue base moved.
Using MAD as a control limit
MAD earns its keep as a threshold rather than a report line. Compute a rolling MAD over the last four to eight periods and set a band at three times that value. Any period landing outside the band is an event rather than noise.
That distinction changes the response. A quarter inside the band means the model is behaving as expected and any lean should be corrected with calibration. A quarter outside the band means an assumption broke. ORM's read on why forecasts fail points at exactly that: the business or the market changed and the forecast was still running on old assumptions, which is a diagnosis you reach by investigating the outlier rather than by nudging a multiplier.
MAD also feeds tracking signal, which divides the running sum of signed errors by MAD to detect whether the misses have started leaning one direction. Read MAD alongside forecast accuracy and the direction of your error, and use both to grade the sales forecasting process that produced them.
Frequently Asked Questions
How do you calculate mean absolute deviation for a forecast?
Take the absolute difference between actual and forecast for each period, add those differences, and divide by the number of periods. Errors of $300K, $200K, $500K and $400K across four quarters give a MAD of $350K.
What is the difference between MAD and MAPE?
MAD reports the average miss in dollars. MAPE reports it as a percentage of actual. MAD is the better number for a plan that commits dollars, and MAPE is the better number for comparing forecasters who run territories of different sizes.
Is mean absolute deviation the same as mean absolute error?
In forecasting the two names describe the same calculation. MAD is the older term carried over from inventory and demand planning, and MAE is the term used more often in statistics and machine learning.
How do you use MAD to set a control limit?
Set a band at a multiple of MAD, for example three times MAD, and investigate any period whose error falls outside it. A quarter that misses by more than three MAD is not the usual noise and deserves a root cause review rather than a calibration adjustment.
Put these metrics to work
ORM builds custom revenue forecast models that turn concepts like mean absolute deviation (mad) into prescriptive action for your team.
Schedule a Demo