Forecast accuracy is often blamed by non sales functions in an organization for bringing increase to inventory and to reduce service levels by increasing stock outs frequency. We forget while blaming that it is estimation and the actual facts and this is the reason it is called as forecast. Forecasts will always be inaccurate though the level of accuracy might increase in some cases and will deteriorate in others. Now, as we know, the whole supply chain efficiency and performance is highly dependent on the forecast and since the forecast has their own dynamic inaccurate levels, it is important for us to de-risk our supply chain from the impact of inaccuracies in the forecasts. We generally have two major types of forecasts viz. demand forecast and the supply forecast. The demand forecast is to help manage your operations, and supply forecast to send to your suppliers to prepare them to respond to your requirements. Again, the supply forecast is a direct result of the demand forecast. Now, either we can improve the forecast accuracy by improving collaboration between customers and suppliers or/and we can de-risk our operations from the impact of forecast inaccuracy by incorporating quicker response to demand changes.
How can we improve the accuracy of forecast? The forecaster’s dream is to develop a process that predicts the exact demand that will be received in every future period. With so many factors influencing actual demand, an exact forecast will remain just that – a dream. So it can be helpful to think of your actual demand as a combination of both a predictable portion and a random portion. Your hope is that by considering all the “right” factors in your forecasting process, the portion of your demand that is random disappears, thus greatly minimizing the risk of forecast error. The ultimate measure of an accurate forecast is the number of times you had hit stock outs while improving or maintaining your inventory turns per annum. If your demand is very predictable, you could carry very little extra inventory and you would never have a shortage. Inventory can be used to buffer against the unpredictable demand and there are numerous costs which can be avoided by carrying just the right quantity of inventory for every part you sell.A lot of mathematical research has been done to generate forecasts from historical demand. The assumption is that past results are a reliable predictor of the future. As world economic events continue to unfold, it is quite clear that historical demand is, at best, just one possible indicator of future demand. Therefore, in today’s environment, “statistical forecasts” should be seen as just one input to the demand planning process.In addition to using mathematics to generate a better forecast from historical data, another approach to improve forecast accuracy is collaboration. Now, what is collaboration? According to Wikipedia, Collaboration is a recursive process where two or more people or organizations work together toward an intersection of common goals done by sharing knowledge, learning and building consensus. Collaboration seldom requires leadership and can sometimes bring better results through decentralization and egalitarianism. In particular, teams that work collaboratively can obtain greater resources, recognition and reward when facing competition for finite resources. Additionally, the Collaborative forecasting is the process for collecting and reconciling information from within and outside the organization to come up with a single projection of demand. Normally, a collaborative process is more suited to create a demand forecast that considers multiple, and sometimes competing factors like:
The collaborative forecasting process normally needs to get inputs as below:
- Historical trend of demand, including trends, similar products and seasonality
- Macro and micro economic trends
- Advertising & promotions
- New product introductions, competitor activities and pricing shifts
- Insights of demand and supply chain planning participants
- Sales force sends their data on forecasts. Most of the times, this will be at finished goods level and this finished good might be a complete assembly or a service part
- Customers, distributors, dealers and other sales channels also send their inputs as per the local demand patterns and event dependencies
- Marketing and Statistical teams check the demand history for respective period
- Market research team will capture the other market factors like any positive or negative events coming up, money flow and the factors it is dependent on, product trend in the target market segment etc.
- Run the process infrequently (once a month at best, often once per quarter)
- Bucket data by large time buckets (monthly forecast and monthly actual sales)
- Summarize data by product family or other superset of parts
- Determine forecasts for individual SKUs through disaggregation models from the level where the forecast is actually created