.png)
"How accurate is your forecast?" is the most common—and best—question to ask any forecasting provider—and it deserves a more honest answer than just a single percentage figure. After all, forecast quality isn’t a simple number; it’s a matter of using the right metrics. This article explains how to measure forecast quality effectively—and how to spot marketing figures.
Why "95% accurate" means nothing
A general accuracy figure without context is worthless, for three reasons:
The level of aggregation makes all the difference. Forecasting the weekly sales of an entire business is easy—errors in individual items cancel each other out. Predicting the daily sales volume of a single item in a single store is many times more difficult. Yet it is precisely this granular level that production requires. Anyone advertising accuracy should specify at what level it was measured.
Small quantities can skew percentage figures. For an item that sells three units per day, selling just one unit less than planned already represents a 33% error—even though the forecast was practically perfect. Percentage errors always look bad for slow-moving items, while absolute errors look bad for fast-moving items.
Chance sets a lower limit. Demand has a genuine element of chance—the spontaneous bulk order, the tour bus. No forecast in the world can predict this element. Perfection is not an attainable goal; the relevant goal is to fully capitalize on the predictable portion.
The Standard That Really Matters
The honest question isn't "How accurate is the forecast?" but rather, "How much better is it than my current method?" The benchmark is the previous planning method—an estimate based on experience, the previous week's figure, or the Excel average. Progress can be accurately measured against this baseline: same time period, same items, forecast versus actual planning versus actual sales.
And even more importantly: In the end, it’s not the statistical metric that counts, but the operational results—return rates and availability. A forecast may look impressive statistically but be of no practical use if it isn’t translated into concrete quantities. That’s why GoNina measures success by the numbers that make a difference on the ground: up to 52% less excess inventory and up to 6% more revenue compared to the previous forecast.
Where Forecasts Are Reliable—and Where They Aren't
Stark: recurring patterns. Days of the week, weather, vacations, seasons, holiday schedules—these account for the majority of demand fluctuations. It is precisely on atypical days, when human estimates are most off the mark, that the data provides the greatest advantage.
Weak: truly one-time events with no data trail. That’s where humans come in—every GoNina forecast can be adjusted manually if the company knows something that isn’t in the data.
Here's How to Evaluate a Provider's Reputation
1. Based on your data, not on sample slides. A test using your own sales data shows how well the forecast works for your business— GoNina offers a 4-week trial period with a money-back guarantee.
2. Compare at the item-by-day level. The level at which production takes place is the level at which measurements must be made.
3. Base it on key performance indicators. Returns and clearance sales before and after the launch—that’s the proof that counts.
Frequently Asked Questions
Does the forecast improve over time?
Yes—every day of sales provides new training data, and changes in purchasing behavior are continuously factored in. The biggest improvements occur in the first few weeks; after that, the forecast improves steadily.
What happens when data quality is poor?
The forecasting model can only learn from the data that has been entered—clean product master data and recorded returns directly improve data quality. Gaps tend to appear during the onboarding process and can usually be resolved quickly.
Conclusion
To honestly assess the quality of a forecast means: measuring it at the right level, comparing it to your own previous method, and evaluating it based on business results. A single marketing percentage figure doesn’t answer any of these questions—a four-week test using your own data answers all three.
To learn how AI sales forecasts work in detail, check out our comprehensive guide for bakeries.
