In our previous article, How to Improve Your Performance with Forecast KPIs, we explored how metrics like accuracy and bias tell you how right your forecast is. But there is another crucial piece of the puzzle that often gets overlooked: variability.
Variability measures how much your forecast changes from one planning cycle to the next. Its main purpose is to help you prioritize your review time — the SKUs with the highest variability are the ones that shifted the most since last cycle, and therefore the ones that deserve your attention first.
Here is a breakdown of forecast variability and how to use it in SKU Science.
What is Forecast Variability?
Forecast variability measures how much a forecast changes from one planning cycle to the next.
Imagine you make a 12-month forecast every month:
In January, you forecast for Feb–Jan.
In February, you update it.
Both versions share 11 overlapping months.
To measure variability, we look only at those overlapping months and see how much the numbers shifted between the old version (F old) and the new version (F new).
How It's Calculated
The formula combines two different ways a forecast can change, then scales it so you can compare a small product line to a massive one.
Variability = ( Σ|F_new − F_old| + |ΣF_new − ΣF_old| ) / ( (ΣF_new + ΣF_old) / 2 )
The numerator adds two things: the sum of month-by-month absolute changes, plus the absolute change in total volume. The denominator is the average of the old and new total forecasts, this normalizes the result so you can compare products of very different sizes.
Breaking Down the Math:
1. The Month-by-Month Shifting:
Captures how much individual months jittered up or down.
2. The Big Picture Volume Change:
Captures whether the overall total demand expectation increased or decreased.
3. The Fair Scale (Denominator):
Divides the changes by the average size of the forecast. This ensures a shift of 100 units on a product that sells millions doesn't look like a crisis.
Why You Should Care
A forecast that jumps around constantly causes a "whiplash" effect across your business.
High Variability Causes: Constant production tweaks, inventory panic, and planners losing trust in the data (leading them to ignore the system and guess instead).
Low Variability Delivers: Stable shipping and manufacturing schedules, predictable inventory levels, and a planning team that actually trusts the numbers.
The Sweet Spot: Stability vs. Responsiveness
A perfect variability score of 0% isn't the goal—that just means your forecast is frozen and ignoring real-world changes. The trick is balancing three elements:
The Forecasting Trifecta:
Accuracy: How close you are to reality.
Bias: Ensuring you aren't consistently guessing too high or too low.
Variability: Ensuring your path to getting there is smooth, not a rollercoaster.
Variability in SKU Science
SKU Science surfaces variability in three places:
In the Navigator, the variability percentage is calculated between the current forecast consensus and the latest archived consensus. Use it to get a quick read on how stable your overall forecast is.
In Forecast Edition, you can measure variability at an aggregated level. Sort by variability descending to immediately spot the SKUs or families that changed the most — those are the ones that need your attention first.
In Reports, you can export your SKUs along with their variability percentages and sort them as needed for deeper analysis or to share with your team.
The percentage displayed corresponds to the variability calculated between the current forecast consensus and the latest archived consensus, following the formula described above.
In Forecast Edition, you can measure variability at an aggregated level and sort the results to identify which aggregations are the most or the least stable.
Finally, in reports, you can export your SKUs along with their variability percentages, and sort them as needed.
Variability is your shortcut to a smarter forecast review. By sorting SKUs from most to least variable, you spend your time where it matters — on the forecasts that actually changed.