Technologies and planning.
The hype surrounding new technologies has identified that many tasks will become automated and jobs are at risk. This will include supply chains, but how long until there is concrete evidence of job replacement actually occurring?
Recently, Sears, a US based retailer announced that all items sold through its stores will have an RFID tag attached for traceability. This announcement took me back to the late 1990s, when many media articles convincingly argued that RFID would transform B2B and retailing. The US based retailer Walmart announced that it would become the leading adopter of RFID and failed. Now, 17 years later, there is the potential for success.
The timing of 15 – 20 years between hype, pilot projects and more general acceptance of a technology is not unusual. Will this be true of Cognitive Computing and Artificial Intelligence (AI), used to simulate human thought processes for planning supply chains? A few software companies are at the pilot stage for applications that address elements of this activity, which are applicable in Fast Moving Consumer Goods (FMCG) and Consumer Packaged Goods (CPG) businesses. Examples of applications in development are:
- Access and evaluate the current ‘outside-in’ data from social media, weather forecasts, analyst reports etc.;
- Reduce the bias and forecast error associated with new product releases using a Forecast Added Value tool; this evaluates the potential revenue, trade promotion cost and demand planning inputs against multiple sales forecasts
- Identify the range of probable sales outcomes for items with high forecast error and intermittent (lumpy) demand, such as service parts and SLOB (slow and obsolete) inventory, caused by the long tail of SKUs targeted at small market sectors
While these and other applications are being developed and trialled, the planning of supply chains still contains many areas in which planners and other supply chain professionals can build their understanding of principles and techniques and improve the traditional spreadsheet approach.
Measure the forecast error
Planning relies on forecasts, but forecasts of future events will never be ‘correct’, as they are based on the most likely scenario of past actual sales and perceptions of the future. Therefore, to compensate for ‘errors’ in the forecast, safety stock must be calculated for each SKU. The term ‘forecast error’ is a statistical term to identify the difference between planned and actual situations.
To improve accuracy, forecasts are often done by product group, with the planning system breaking down the group forecast to SKU, based on the product mix of prior sales within the group. The forecast for particular SKUs is then adjusted to account for known promotions. In the inventory system, the SKU should be grouped by category/class (discussed in the blog Inventory Policy is built on the inventory structure).
The preferred measure of the ‘forecast error’ is the mean absolute percentage error (MAPE). Logisticians define MAPE as the difference between forecast and actual sales divided by actual sales. This measure provides an indication of the forecast accuracy for all SKUs in a category/class and by distribution centre (DC).
Given that forecasts consider the future, how accurate should they be? For SKUs in the high selling STEADY category, a MAPE of 10–15 per cent can be expected. Moving towards the tail of SKUs, a range of 20–30 per cent forecast error is not unusual and it can be up to 60 per cent; due to bias in the forecast and unknowns associated with the sales of a SKU.
As with any measurement process, there is a temptation to make the numbers look better than they actually are. This can impress senior management and give Logisticians a false sense of security. Two approaches are not advised:
- limit the forecasts to the product group level and
- weight the errors – called the Weighted Mean Absolute Percent Error (WMAPE)
The Tracking Signal assists inventory control by SKU. It is calculated by dividing the cumulative variation for the number of periods under review by the standard deviation; a control limit is established by category. For SKUs in the high selling STEADY category, the acceptable tracking signal will be up to about 4. For lower categories, forecast error for an SKU is likely to be more varied; therefore, use a tracking signal of more than 7.0 as the trigger for a review.
While MAPE and the Tracking Signal are currently in use, the increasing number of SKUs and marketing promotion events means the approach to forecasting at product group level and allowing the planning system to calculate the item level forecast, based on historical patterns, is proving to be less satisfactory. Forecast errors are increasing, which affects the accuracy of planning.
Therefore, as an improvement, consider changing the current approach of providing one sales number per month for each SKU. Future sales should be shown as a range, identifying both the optimistic and pessimistic forecasts, with probabilities that reflect the likelihood of each occurring. Multiplying the forecasts by their probabilities and adding the results will provide a more likely indication of future demand.
Forecasts based on ranges can help Logistics Operations. Supplies of long lead time items can be acquired at the optimistic forecast level, while materials that are easy to obtain can be acquired at the pessimistic forecast level, but with provisional orders in place to increase the amount to be acquired at short notice.
Your objective for supply chain planning should be to reduce the MAPE (become more accurate) through investment in training, process improvement and systems. While achieving this objective, understand the developments in planning applications, using technologies such as Cognitive Computing and AI and how they can potentially be used in your organisation.