Future perfect

Food companies may not be in the habit of crossing gypsy palms with silver, but manufacturers depend heavily on accurate demand forecasting for the success of their businesses. Getting it right enables companies to pull off the tricky balancing act of minimising inventory and waste without impacting on their service levels.

Strawberries have traditionally been the ultimate example, with enormous fluctuations in supply and demand. "You know that in June demand will always outstrip supply if there's good weather, otherwise you could be stuck with a lot of fruit," says supply chain specialist Tim Knowles. "Soft drinks are another good example, where demand will double or triple if we have good weather during the school holidays, but bad weather leaves manufacturers with an embarrassment of riches."

At the other end of the scale are products such as pet food, where the demand is constant throughout the year. But most food and beverage products lie somewhere between the two extremes, showing demand fluctuations that are rather more dependable than the weather. More predictable demand drivers might include Christmas or Easter, for instance, or supermarket promotions such as BOGOFs (buy one, get one free). Even so, planning production schedules to fit demand can be extremely tough, especially if retailers decide to throw a spanner in the works.

"Promotion decisions are normally made three or four months in advance, but a retailer might change its mind," says Knowles. "Safeway used to be notorious for bringing things forward and creating demand for stock that wasn't available. There's only so much suppliers can do about maverick retailers."

But it's not all the fault of the retailers. If the supplier does not have the right systems in place, its sales staff could be promising stock that simply isn't available. "Perhaps the sales staff misquotes or leaves it too late in the day to sort out availability. It's a particular problem for pan-European operations with production sites in multiple locations. Many big firms can have lead times of weeks on manufacturing," says Knowles. "It can cause a lot of embarrassment if promotions in the UK and Germany both hit in the same week, for instance."

The process of overlaying forecast demand and current stock levels to come up with a production plan is known as sales and operations planning (S&OP). Again, this becomes extremely complex in companies with multiple manufacturing operations, because so much data has to be amalgamated and significant errors can creep in. For example, there can be problems with factory schedule adherence, where a plant may say on paper that it makes 1,000 units a week, when in fact it makes only 900. This sort of problem means that even some of the biggest companies don't always get it right. "97% fulfilment is generally regarded as bad service, but I know of at least one major manufacturer who's still down in the high 80s. It's an S&OP disaster," says Knowles.

IT or organisational systems?

So is IT investment the solution, or is accurate forecasting more a matter of having the right organisational processes in place? There's no simple answer. Almost every firm, apart from the smallest, will use a combination of both.

Bigger firms rely on IT to process data. But even here it's the organisational systems that hold the key to accurate forecasting, according to Matt Parker, development planning manager for Kraft Foods.

Kraft has developed a six-step process for advanced demand forecasting, which it uses in all its operations across the world. Although the size and complexity of Kraft's business means that IT is an important tool, Parker stresses that smaller firms could adopt a similar approach using the office PC. "It's possible to apply our six steps in a very low-tech way. You can work this through using a simple spreadsheet. It's more about the organisational set-up," he says.

Parker describes step one as "the dull bit". It involves preparing the data and making sure that each product and customer is captured in the system. "It's quite complicated because it will be ever-changing in most firms," he says.

Step two is the creation of a baseline forecast, which separates the baseline volume the company expects to sell from the extra sales generated by promotions.

Step three incorporates information from customers about planned promotions. Variables include the depth of the promotion (BOGOF or 20% off, for example), the position in store, marketing support, competitor activity and seasonality. "The impact is always difficult to predict because so many different variables come into play," says Parker. "This part of forecasting is all about talking to people. For predicting the baseline demand you can use software, but for promotions it's all about the relationship with customers."

Step four is the "sense check" in the system, in the shape of consensus meetings. "The process starts at a very detailed level, looking at each stock keeping unit (SKU) and specific customers. We then look at the brand group level for all customers and at this stage we meet to review the forecasts," he says.

The approved forecasts are then turned into operational forecasts for the company's manufacturing operations in step five, while step six rounds off the process by evaluating the forecasting performance so Kraft can make any necessary adjustments to the process.

Importantly, advanced demand forecasting is not about deciding whether or not the company can meet demand. That comes afterwards, according to Parker. "From a forecasting point of view we're looking to predict what the customer demands. That process needs to be 'pure', otherwise we're hiding demand. We worry about whether or not we can supply it later on."

So for some firms, IT simply provides a tool for organising and viewing large amounts of data, with human judgement at the centre of the forecasting process. Yet for others, IT-based modelling plays a more important role in improving forecasting accuracy by reducing the impact of any bias on the part of the people making predictions. For example, Irish drinks manufacturer C&C recently opted for a sophisticated modelling solution from Logility as a way of improving the accuracy of the projections coming from its sales teams.

C&C's main products are Magners Irish Cider for the UK and Bulmers Irish Cider for the Republic of Ireland. It also produces Tullamore Dew Irish Whiskey and Carolans Cream Liqueur.

In Logility's Voyager solution, C&C has essentially bought in a set of sophisticated statistical algorithms to model changing demand. Data about historical sales are fed in and the system runs up to 15 different statistical models to find the best fit with the actual sales by sku. The best fit model is then applied to predict future demand.

"It effectively removes human bias from the process," says supply chain manager Shane Hughes. "People always have a certain level of bias in putting projections together, whether it's conscious or subconscious. Sales staff might put an optimistic spin on their projections, for example. These algorithms put the science back in to give an unbiased statistical model."

Human input remains important, however, because there will always be factors that are invisible to the system that will impact on the accuracy of the statistical approach. C&C therefore passes the Logility projections on to its sales team, who then use their hands-on market knowledge to adjust the forecasts. "The system gives a baseline forecast that can be increased or decreased based on market intelligence," says Hughes. "The amended forecasts are then submitted to planning in the plant. It's a closed-loop system."

A balancing act

Seasonality and long lead times make forward planning a big part of the cider business, according to Hughes: "Apples grow once a year and we only have one window in which to buy them, crush them and turn them into juice. Then there's a minimum maturation of nine months for the cider," he says. "We have to err on the safe side. We don't want to have a really good year and get halfway through, only to find we can't make any more cider because we didn't buy enough apples."

But it's a balancing act, because the business can only afford to hold so much inventory. "We obviously want to offer as high a service level as possible but it's a constant trade off for us as it is for all businesses," says Hughes.

Knowles says that forecasting is ultimately about making the best of a difficult situation: "The issues facing us in improving demand forecasting are the sheer complexity of our own behaviour as manufacturers and that of our customers, competitors and the weather. Best practice ensures we do not shoot ourselves in the foot by ignoring information that lies within our control. That still leaves us with many issues that lie outside our direct control and will cause demand forecast errors.

"Those manufacturers that have invested in common sense processes to get it as right as possible will carry less stock, throw away less of the stock they do have, satisfy more customers and employ supply chain staff who do not need to live on tranquillisers," he concludes.

l Food Manufacture is hosting its first supply chain optimisation round table debate chaired by Tim Knowles in Whittlebury, near Towcester, on July 8. This free half-day event is sponsored by Infor and Preactor. Call Helen Law on 01293 846587 or email uryra.ynj@jvyyvnz-errq.pb.hx​​