In demand planning, getting people to use the forecast is more important than sweating which metric to report. Don't lose your audience or jeopardize adoption because your accuracy reports sound more like Oppenheimer than Barbie.
No matter how precise or sophisticated a metric may be, it only matters if people understand it and use it. The best forecast accuracy metric is the one that clicks with your team and supports action rather than delays it.
If your model gives you better forecasts by tuning it with the latest Jumble-Of-Letters-Error metric, great, use it internally in your model. But like your Wordle streak or fantasy football points, you can enjoy your success without telling people about it.
Choosing Metrics that Work for the Model and the Business
Let’s say RMSE (Root Mean Square Error) is your analyst team's go-to because it minimizes large errors for your forecast model and also directly drives your inventory calculations. Dandy! But if other people find MAPE (Mean Absolute Percentage Error) more intuitive, publish your results in MAPE. It’s more important for everyone to understand and act on the forecast than to have a technically perfect metric.
The approach could be to have one metric for refining your model and one for helping the business make decisions.
Use simple metrics like MAPE or MAD for a clear, straightforward view that everyone can follow.
Also use advanced metrics like MASE, RMSE, NRMSE, or RTAEÂ under the hood if they give your models a boost.
If you have one metric for internal use and another for external users, your analysts are free to modify the internal metric without the change management overhead of switching the published metric.
Monitor the internal and external metrics over time to ensure they generally move in the same direction.
Don't be afraid of simplifications such as calling weighted MAPE just MAPE. Most users expect that larger volume products affect the forecast more than smaller volume products. Similarly, using a symmetric version of MAPE is handy for intermittent demand. But just call it MAPE.
Keep the detailed metrics behind the scenes and publish simpler ones for general visibility.
Making Forecast Accuracy Clear and Actionable
Beyond picking a metric, how you present it matters.
Traffic Lighting (red, yellow, green) can communicate the state of forecast accuracy at a glance.
Line charts are great for showing trends, and trends can be more important than point numbers. Instead of focusing on the accuracy level itself, ask: is it improving?
Control charts can help manage minor fluctuations in forecast accuracy that are simply part of the game. For instance, if tiny changes are handled by buffers like safety stock in your supply chain, there’s no need to panic over them.
Segmenting your thresholds by demand type will keep the focus on meaningful shifts.
Lastly, don't forget about tables. The ability to filter, sort, and have the system percolate up items and locations that need attention should be a requirement in any forecasting system.
Is Reporting Forecast Performance Worth The Trouble?
Yes, FCA is beneficial for many reasons, and I'll cover two here. First, accuracy metrics show the effectiveness of your forecasting process, especially over time. Forecast Value Add (FVA) is a tool that can use your chosen error metric (MAPE or another!) and tells you if your efforts are paying off. Even when certain products or customers are tough to predict, FVA can highlight whether planners and algos are making a positive impact.
Secondly, forecast accuracy can tell users how much faith to put into a forecast. They can do this by evaluating the error metrics independently, where higher accuracy means more predictive power. Better yet you can use recent forecast error to calculate a prediction interval, cone of uncertainty, or ranged forecasts that show how the forecast is really a spread of possible outcomes. This range of outcomes can be displayed in a table, or graphically, as above. The range typically widens as you move further into the future.
Conclusion
Focusing only on technical metrics can leave teams feeling lost or disinterested. Instead, pick metrics that work for the business, highlight trends, and focus on value add. Demand planning is about supporting decisions, not just perfecting numbers. By choosing user-friendly metrics, using control mechanisms like traffic lights and FVA, and prioritizing engagement, you’ll build a demand management process that’s impactful.
Comentários