Sales forecasting is not about complex algorithms

Many companies typically try to solve these challenges with one of two solutions:

1. We neglect the sales forecast and “accept” that it is a difficult and complicated process. As a result, the forecast accuracy remains poor, resulting in a non-optimised supply chain.
2. We outsource our forecast to a third party. As a result, we pay for an expensive “black box solution” which is non-transparent and difficult to figure out.

Our position on this matter is quite simple: “No matter how tempting one of these solutions might be, we can and must do better.”

• What if we argued that sales forecasting should not be considered a problem, but as the backbone of a stable organisation that has a significant impact on the effectiveness in our supply chain and thus also on our bottom line?
• What if we argued that there is no significant value added nor increased forecast accuracy in the sales forecast associated with receiving a number generated by complex algorithms based on sales history on which third-party forecasters otherwise pride themselves?
• What if we argued that advanced statistical algorithms alone do not create value?
• Would that not also require that we share our experience on how to deal with these challenges by looking at things differently?

The purpose of this article is to demonstrate why sales forecasting should be given priority and where to focus our resources most effectively. It can actually be done in a simple and efficient way.

But first, there is one essential question that managements and decision-makers in all companies should ask themselves: “Do we consider sales forecasting as a necessary evil or as a value-creating activity?”

This is a fundamental question, and the answer will help you understand why this area should be given priority, why you should think twice about outsourcing the sales forecast, and how you can and should increase the quality with other methods.

Even though it intuitively makes good sense to most people that a forecast must be as accurate as possible and of the highest quality, we often experience that companies underestimate the effects of a poor forecast. In a way, this is understandable seeing that it can be very difficult to track the actual effect of a good or poor forecast on the bottom line, which automatically turns forecasting into a necessary evil that has to be there, but is not given priority. The fact of the matter is that there are major consequences associated with a poor forecast. Not just in relation to costs and earnings, but in many other areas of the organisation as well. But before it can be analysed further, it is necessary to understand the decisions behind the forecast.

Different decisions require different forecasts

A forecast is not just a forecast – it can be divided into different decisionmaking horizons. The short-term forecast, focusing on the coming days to the coming weeks depending on the industry, is used for estimating the need for stocking the right products at the right time and in the right place. The medium-term forecast is used e.g. for making decisions on the size of commodity contracts, the amount of manpower and the number of machines. The long-term forecast is used e.g. for financial budgeting and long-term strategic CAPEX¹ decisions.

In addition to these production-related decisions, the forecast should be used as a management tool in the sales effort. If some specific sales goals are set up, a structured forecasting process will reveal e.g. if a sales goal cannot be reached. It is then possible to take a proactive approach and stimulate the demand on the market by utilising the already booked capacity.

When the goals have been listed, the effects of a lack of focus and a nonoptimised forecast process become more clear – this can end up being quite significant. For instance, if the short-term forecast is not accurate, it affects the delivery performance and results in lost sales, increased inventory levels and the risk of obsolescence and depreciation. However, if the short-term forecast is accurate, it eliminates the need for large safety stocks for a specific delivery performance seeing that the uncertainty in the forecast should reflect the level of the safety stocks.

The correlation between forecast accuracy and the level of safety stock is illustrated below. The illustration shows an improvement percentage from 60 % to 70 % in the short-term forecast accuracy reflected by a reduction of the safety stock of ≈ 30 %. Naturally, this affects the tied-up capital in the stocks. 

Despite the major cost implications caused by a poor short-term forecast, the most significant costs are the result of poor medium-term and long-term sales forecasts. On this horizon, the focus should be on forecast bias rather than accuracy with bias being the tendency for the forecast to consistently overestimate or underestimate sales.

Consistent overforecasting results in unutilised capacity of manpower and machines, too large quantities of raw materials and the risk of waste and scrap owing to overestimation of procurement contracts, increased handling costs and net working capital. For example, a consistent overestimation of future sales (8 %) can result in too much manpower (8 %), too high machine capacity (8 %), too high stock levels (8 %), too large procurement contracts (8 %) etc. This has a significant impact on the company’s variable costs and thus has a direct effect on the company’s bottom line.

On the other hand, consistent underforecasting will result in belated scaling decisions regarding capacity, resulting in overstretched resources, overtime and expensive temporary work as well as stress and the need to put out fires in the organisation – not to mention expensive rush orders on raw materials, packaging etc.

Hopefully, it is now clear why it is important not to underestimate a good sales forecast. However, keep in mind that getting there is not so simple, and if you do not keep your eye on the ball, you risk drowning in complexity.

“Keep it as simple as possible, but not simpler”

It is a good idea to make a mental note of this famous quote by Albert Einstein, when designing the solution for the process and the organisation in connection with the preparation of a sales forecast. That said, it is easier said than done in a field of numbers and analyses, where complex solutions with several variables intuitively seem more correct and reliable. Broadly speaking, this mechanism is what the providers of sales forecast outsourcing solutions rely on in the form of complex algorithms – including, of course, more accurate forecasts as the pot of gold at the end of the rainbow. However, this does not always prove to be correct compared to using much simpler models, which is illustrated by the following two examples:

• If there is large variance in the sales demand and the future sales ARE NOT a reflection of the past, no advanced statistical models will be accurate because the history, in fact, CANNOT be written. Instead, a simple statistical forecast can be used for identifying an average level to which Sales and Marketing contribute with input concerning market knowledge of trends, campaigns and other situations that the history cannot reveal.
• If there is low variance in the sales demand, and the future sales ARE a reflection of the past, no advanced statistical models are necessary. Once again, it is most beneficial to use a simple statistical forecast, because the history, in fact, is reflected in the future. Input from Sales and Marketing should be incorporated in the event of changes in market conditions, e.g. in connection with new customers and competitors – all of which are elements that no advanced history-based statistical models are able to predict.

The observant reader will probably wonder whether there is in fact a need for outsourcing the forecast in order to gain access to the advanced statistical algorithms as argued for by the providers?

No, in popular terms, it is a nine-day wonder – it adds no value. In spite of the providers’ persistent efforts to convince us otherwise, there is no crystal ball to reveal what the future holds based on the history. What is even worse is that for companies that use advanced statistical algorithms, a “black box” will be forecasted with numbers that often do not make sense to the people who need them. Consequently, it reduces the trust in the generated forecast which is ignored and overridden by manual entries.

The value of a good and accurate sales forecast lies in the process, structured manual input and a differentiated approach

As stated above, the value of a good and accurate sales forecast is not found in advanced algorithms. The value is found in having implemented a structured process with a regular takt, where Sales and Marketing are involved at the right time and in the right areas. The sales forecast process can be broken down into the following phases:

1. Make sure that the historical sales demand is of good and representative quality. This means that campaigns and other outliers, which are not representative to the future, are removed. Set up a couple of threshold values to identify these exceptions in the history. Work out a simple statistical forecast and include seasonal patterns, if any. Segment products/product groups and identify where to focus your attention. Find out which products are stable and predictable – these products will typically be “hands off”, whereas unpredictable products require far more manual input and market knowledge.
2. The part of the company with the largest market and product knowledge and the latest knowledge are typically not the same people who prepare the forecast. Therefore, involvement of the responsible individuals from Sales, Marketing and Category management is necessary when collecting relevant market input. For instance, will any new products, customers or campaigns be introduced? All input, which is crucial to an accurate forecast and which no algorithms can predict. Make sure to have a fixed weekly or monthly takt time for how and when these stakeholders should be involved.
3. Adjust and correct according to input and KPIs, e.g. forecast bias. Agree on a final forecast to be communicated to the rest of the organisation and on which all decisions should be based. Assess and discuss the “root cause” of major changes or errors in the forecast and make sure to continuously develop and improve the process.
4. Communicate the forecast to relevant stakeholders in the organisation in the right format, e.g. units for Production, kilos for Purchasing and currencies for Finance to ensure a good and solid consensus forecast. This is the process that creates value, if it is designed correctly and improved on a continuous basis. And, of course, it cannot be outsourced to a third party, because the important and crucial knowledge of the market exists internally in the organisation. Once again, this turns the postulate on advanced algorithms – based on history can do all of this – into a utopian dream and wishful thinking.

But then why do an increasing number of companies decide to outsource their forecast? First of all, it is probably because the importance of the sales forecast is underestimated, and outsourcing is a convenient way to “get rid of the problem.” Secondly, it is probably because we often overlook the fact that the real value creation is found in the good and structured process – NOT in advanced algorithms.

In any case, we argue in favour of the following: “Yes, sales forecasting is an important activity which must not be underestimated under any circumstances.” In the same way, we argue that: “No, the value added cannot be outsourced, as value is created internally via a structured process” and, last but not least, “No, it does not have to be scary and complex – it can be done simply and effectively.”