How to improve forecast accuracy?
To begin within any business, one should first evaluate if forecasting is required at all. Many a time sales based replenishment is ignored as impractical without further thought. However, after implementing concepts such as lean and TOC, many companies have found pull based systems to be much more suitable for supporting the business. A point to note here is that forecasting is easier in some industries than others and so some techniques would work better in some industries than others. Forecasting new products is again a different matter altogether and is covered in a short section separately later. But given a set of fairly established products, it should be possible for demand planners to achieve forecast accuracy above 70%.
Sales, Marketing and Demand planners have a role of reviewing and validating the forecast made in the previous forecasting cycle. Those in a supervisory role need to quickly review the work of their subordinates. While experience does play a role, here are some basic checks. An obvious place to start with is comparing the forecast with past sales, say the last 6 months. Changes in trends are easily discernible when numbers are in a series. Reasons for trend changes are usually because of ongoing marketing efforts, but their impact needs to be reviewed every forecasting cycle. Depending on the effectiveness of the marketing efforts, the forecast numbers will change every month. You also need to look out for spikes or dips in the series. Spikes probably indicate a trade offer or marketing promotion. The period immediately after a trade offer is followed by a steep dip in sales. Just as sales peak during an offer period, it is equally important to reflect the sales dip in the periods immediately after the offer in the forecast. Another check is comparing the forecast with the budget especially at the beginning of the year. The two need to be aligned. A forecast that is different from targets has very likely not been viewed by the sales team. Another good comparison is the previous year’s sales in the corresponding period. This is useful in checking for seasonality and also for considering business activities impacting the forecast which are very likely to be repeated from the previous year - trade offers, discounts, TV commercials, etc. However one of the best checks is comparison with the forecasts given in the previous months. Changes made to the forecast can indicate a lot of things - changes in sales and marketing tactics, attempts to cover up for lost sales or shortages in the same product or related products, lack of long term planning, etc. It can also explain reasons for shortages. For example, if the numbers in a forecast have increased dramatically by 3 to 4 times, it is most likely going to lead to a shortage situation before production catches up with the demand and replenishes the depletion in inventory.
At an aggregate level tonnage, forecast value and gross margin are some of the parameters used for checking the quality of the forecast. These can be viewed using several hierarchies. Geographical involving areas, regions or Customer hierarchies involving stores within a retail chain, based on Products involving brands and businesses or time based with months consolidating into quarters, year and so on.
How is it different when forecasting for a new product? Forecast review here is with data of some other product which is the most comparable with the forecasted product. Then there could be comparisons with lots of external industry and channel level data. For example footfalls per month at store level is used in the apparel industry. Industry growth is a common parameter in many industries. Lead indicators are yet another set of data. For example, automobile sales would be a lead indicator for forecasting in several automotive components. Basics being the same, the number of parameters to be compared with is more in number and more external in nature for new products.
Will a tool help? As mentioned above there are several different types of data that a forecast is compared with during review - recent sales, sales in the previous year, budgets, forecasts belonging to previous few cycles, etc. It would indeed be cumbersome to combine these many different data types into one sheet even if the formats are standardized. The forecaster will end up working on the data crunching for most of the time rather than the forecast. Besides, the kinds of checks one needs to have on the forecasts are so many that it is easy to miss on an important one. One needs to remember that the checks points are independent. A forecast could fail on any one or on several checks at the same time. For example, a forecast may look alright by looking at the past trends, but may be off way when compared with the targets. It may still be further off when compared to last year sales in the same time period. There may be a conflict when more than one parameter is compared. For example, in the above case, the previous year’s sales may indicate that the forecast should be trimmed downwards considering seasonality, but increased considering the budget. In such a case it may not be possible to resolve the conflict completely and here it is important to record the reasoning behind making the forecast for future reference and learning. A tool helps in throwing up consistent exceptions based on the rules set-up and also in maintaining these records. Manual review without tools is tiring with many chances of errors. So in the absence of tools, the review is only partially done after looking at only one parameter - sales in the previous few time periods. Having a tool is always better and this would show in the results - better accuracy and hence lesser resulting total inventory.