What is the most effective way to measure forecast accuracy and at what level do you share forecast accuracy reporting with an executive audience? Total company gives an overly positive view of results, but lower levels seem too detailed for the audience. Any insights into how to share the forecast accuracy results and what is considered good (e.g., greater than 75%) would be appreciated.
Sort By:
Oldest
VP of Supply Chain in Transportation7 days ago
Great question. Forecast accuracy has different dimensions: revenue, commodities demand, supply-demand, material replenishment, profitability, and overall forecasting accuracy. To answer your question, one thing that stands out is whether the company has a mature S&OP or IBP program.If the answer is yes, based on industry best practices, a member of the ELT or multiple members, depending on the organization's structure, should participate in those monthly meetings. However, if this is not the case, it's a clear sign that an element of education is required. This is a key area where we can work towards improving forecasting accuracy. It's also important to understand the audience's expectations of what they want to see.
Does your company have data intelligence to inform the forecasting process? Are there real-time dashboards or reporting mechanisms that guide strategic decision-making and planning? Are there financial reports that the ELT reviews that raise questions regarding forecasting accuracy? Do you track COGS or COS?
KPIs or OKRs usually include forecasting accuracy. This is an important tool to not only communicate upwards but also identify areas for continuous improvement or corrective action within the operations.
The main point of interest here is that the ELT is not usually interested in the granular details, but they do want a narrative that can quickly explain any variance to forecast. From experience, ELT members are interested in the final results, and if not the goal, they want to know what is being done to solve or correct.
Global COE Lead S&OP in Manufacturing6 days ago
I agree with what Talbot has said in his post. You could use volume-weighted accuracy as it involves weighting the accuracy by the volume of each product, giving a more realistic picture. Along with FA, you can measure forecast BIAS that shows a tendency to over and under-forecast. The formula for BIAS is the same as the accuracy (Given by Talbot) but without taking the absolute value of errors.For reporting purpose, you could use segmented approach - such as product categories, regions, or periods.
Formula: Forecast accuracy = 1 - Sum(Abs(Forecast - Actual))/Sum(Actual). Applied across a large group of products this will give you a volume weighted accuracy based on units. If you have a very diverse product set you might consider measuring each product group separately or applying a $ weighting. You should also measure forecast bias. Same calculation as above without the absolute value [ABS()] function. This will show a tendency to forecast too high (positive) or too low (negative). Many companies will also measure new product introductions separately.
As far as benchmarks go, 85% at the SKU level 3 months out (3 month 'lag') is the standard in fast moving consumer goods and life sciences. High tech is typically lower in my experience (but should aspire to that number). Ultimately it depends as much on how uncertain your demand is as it does on how good your process is.