Maintaining Optimal Forecast Accuracy

Lori Mitchell-Keller, Manugistics Senior Vice President of Global Marketing and Product Management, shares her insight on the value of implementing and maintaining an optimal forecast accuracy chain.

Last Christmas in the US, Restoration Hardware, a furniture manufacturer, created a small red cabinet with 25 drawers that counted down to Christmas day. The product became one of Americas hottest selling items, and one of Americas must have items.

Then, mid-way through the season, it sold out. The company could not build anymore because, by the time it had restocked its shelves, the Christmas period would have been over. Conversely, it had also produced a similar white cabinet, but with a less festive feel. Restoration Hardware had plenty left by the end of season.

Restoration Hardware is not alone in making mistakes with its forecasting. In the UK, MFI was widely reported in the media to have suffered greatly from not updating its forecasting tools, which contributed to poor sales of its Hygena kitchen cabinets. MFI has a very complex supply chain, as it builds, distributes and sells a wide range of household furniture across 192 UK stores. Failing to forecast accurately meant a huge waste of resources and high costs for MFI.

Other companies are attempting to avoid the hiccups suffered by MFI. It is estimated that companies will spend over 210m worldwide on forecasting applications this year. If Restoration Hardware had properly forecasted the sales of the red and white cabinets it could have maximised profit and continued to ride the waves of its success instead of being unceremoniously discarded in the middle of the season. Of course, hindsight is a wonderful thing and Restoration Hardware will almost certainly learn from its mistake this year. A global industry group, made up of companies including IBM and Proctor and Gamble, has issued a new collaborative planning, forecasting and replenishment standard with guidelines for sharing forecasting data across a supply chain to help out manufacturers and suppliers.

Firms such as Manugistics, have sophisticated demand forecasting technology to deliver improvements in forecast accuracy and offer faster response to demand fluctuations, says Mitchell-Keller, Senior Vice President, Manugistics. Research has proven that small percentage improvements in forecast accuracy can dramatically reduce inventory costs. The benefits of better forecasting include lower inventory holding costs, less inventory obsolescence, less working capital investment, better on-time order fulfilment and faster order-to-cash cycles, and fresher products for customers. It also means companies can make better estimates of how its most profitable items will sell, and ensure they have a good in-stock position.

Visibility throughout the supply planning process will also help to eliminate the risks that exist because of fluctuations in demand. Using the capabilities of an advanced replenishment planning tool such as forecast consumption logic, inventory optimization based on statistical models and visibility into planning exceptions in conjunction with the output from a demand forecasting tool will allow for a proactive approach to managing potential supply chain issues.

Forecasting is always a moving target, and, over time, product demand patterns change. Seasonal products can move into the mainstream, thereby losing their seasonality and exhibiting more stable demand characteristics. Meanwhile, products that have historically exhibited stable demand patterns may begin to show a trend, especially near the end of the products life cycle. Yet while the demand patterns change, the forecasting system continues to use the forecasting parameters that were configured into the system during implementation, when the demand patterns for products were first analysed. As a result, forecast accuracy inevitably begins to deteriorate as demand profiles change.

The difficulty of maintaining and improving forecast accuracy is two-fold. First, even detecting changes in a products demand profile is difficult. Demand profile changes are challenging to detect because it is difficult, if not impossible, for a demand planner to recognise an actual profile change versus a temporary demand fluctuation. Second, even if a demand profile change is detected, resetting forecasting parameters is a time-consuming and often tedious process. As a result, demand profile changes go either undetected or detected well after the actual profile shift has occurred. The time required to adjust forecasting parameters means demand profile changes are not reflected in the system or worse, updates are significantly delayed until long after the fact.

The issue is further complicated by the bewildering choice of forecasting techniques available to companies. Techniques range from the naive to the statistically complex, and if associated forecasting parameter settings are also taken into account, the number of forecasts that can be generated from the same set of historical data is almost unlimited. In the face of these challenges, many manufacturers have been tempted to leverage modern computing power by simply trying a wide range of forecasting methods and then choosing the one which happens to yield the best mathematical fit with past history. But such approaches can yield widely divergent results on different samples of historical demand data even though the data all have the same underlying demand structure. Companies fall into the trap of hoping they can rely on information from previous years. Forecast fit is not the same thing as future forecast accuracy, and forecasting the past however skilfully is not the answer to achieving more accurate forecasts.

Many forecasting techniques are unsuited to addressing intermittent demand and result in biased forecasts. Recent research has identified a break point between continuous and intermittent demand, and although such results do not have the immutability of a law in the physical sciences, they do offer a powerful guide for decision-making. Is the demand seasonal or non-seasonal? If there is an identifiable seasonal effect on demand, firms should incorporate it into the forecast model and thereby achieve greater forecast accuracy. However, if a seasonally adjusted forecast is generated for a product which is not subject to seasonal effects, then the forecast error will be needlessly increased, no matter how well the forecast appears to fit the history.

Trends can be analysed and considered to see what sort of trend might be present in the product. Is it linear, is it short-term or long-term, or is there no trend at all? The task of identifying these characteristics is, however, not nearly so clear-cut as it might appear. A decision on seasonality is in principle straightforward, provided it is known that no other factors are present. A carefully planned step-by-step approach, supported by the right statistical tests and incorporated into a decision table, enables these questions to be examined systematically so that by the end of the process the most suitable forecasting technique emerges. By investing heavily, a company will have more accurate forecasts overall and a reduced or eliminated need to tune forecast parameters. This reduces dependence on user skill, so that better forecasts can be generated more consistently and automatically. Investment in software solutions will also bring a significantly reduced incidence of forecasts being upset by changing events.

Another approach with great promise is Collaborative Planning, Forecasting, and Replenishment (CPFR). The idea is that trading partners throughout the supply chain work together to determine inventory and replenishment strategies and make efforts to align their sales forecasts accordingly. If successful, this collaborative initiative should lead to lower inventory throughout the supply chain, more inventory turns, fewer stock-outs, and higher profitability.

Yet, areas will always exist that no one can predict. For example, employees from different functional areas have agendas that can diminish the accuracy of forecasts. Salespeople who are not hitting their targets will favour lower numbers so they have a better chance of hitting their goals, while finance departments, attuned to the costs of excess inventory, are likewise conservative, while production planners over produce to avoid being blamed for lack of product. Safety stocks will always remain for that disaster waiting to happen.

Firms will never expect to hit the mark every time, but they are realising the fact that companies like Manugistics can offer them large improvements in their supply-chain and forecasting. One things for certain: finding a little red advent cabinet should not be difficult this Christmas.