Harnessing the Power of Downstream Data Through Better Mathematics
In a recent survey on Global Supply Chain Trends by PRTM Management Consultants, 75% of respondents consider demand and supply volatility and poor forecast accuracy to be the biggest roadblocks they currently face. Second on the list is lack of visibility into current market demand. High demand volatility and lack of visibility into consumer behavior caused manufacturers to lose money during the recession and will hamper their ability to meet consumer demand in the economic upturn. S&OP (sales and operations planning) was developed 25 years ago as a way for manufacturers to match supply with demand and access to Point of Sale (POS) data has been available for 40 years, but manufacturers are still generating forecasts that are wrong 50% of the time. Why does forecasting remain so difficult?
Manufacturers face three consistent obstacles to accurate forecasting: they don't receive the data in a timely manner, they don't have a structured approach to analyze the data and create an actionable response, and the information they receive is often missing data or incorrect. The only way for manufacturers to correctly match supply with demand is to improve forecast accuracy by gaining visibility into current demand. Only by accessing and intelligently processing all available data can manufacturers improve forecast accuracy.
The drive to become demand driven, to let current demand drive production and shipment of products, requires that manufacturers access information as soon as possible and analyze that information in the time horizon necessary to create an actionable response. They need to acquire daily data daily and shift the time horizon for planning from weekly or monthly to daily. POS data, for example, provides an accounting of every sale as the product is scanned and is an immediate snapshot of demand. The benefit of POS data is that the data gives a real-time snapshot of demand right here, right now. The value of that data is significantly decreased if the information is collected today and used three months from today. If demand for boxed meals has increased because of the recession, the time to act is now, while consumers are looking for ways to save money. Shipping more products a year from now, when the economy is recovering and people are back to going out to dinner, will not improve sales.
Improving forecast accuracy requires a structured approach to use the masses of data available to make better business decisions. Many large consumer products manufacturers collect and store POS data in demand signal repositories (DSR), but few use it for more than account-level inquiries, such as analyzing promotion effectiveness or looking for phantom inventory. Some companies have considered a flowcasting approach to forecast demand. Flowcasting collects data at the end of the supply chain, the retail store, and creates a sales forecast for every product in every store. The manufacturer then calculates time-phased requirements in the future all the way from the store shelf to the factory. While the concept seems simple and intuitive, it depends on correctly forecasting at the store level plus accurately modeling all intermediate supply chain processes and targets, which is essentially impossible.
Relying on only one type of information leaves manufacturers open to disaster, whether the information is retailer data or shipment history. For example, most consumer products companies use time-series forecasting methods to generate forecasts. This method of forecasting uses historical sales to predict future sales. These forecasting systems failed during the recent recession because consumer spending deviated from historical norms in unprecedented ways.Information is only useful if that data is still an appropriate indicator of present sales and future demand. When consumers moved stores, switched to private label and increased reliance on coupons and promotions during the recent recession, historically-based forecasting could not predict current demand. None of these changes were predicted by history. POS is not a silver bullet either. Some products, like ice cream, are seasonal. If you take a look at ice cream data leading up to summer you will see an increase in orders that precedes the increase in product sold, so POS is not a leading indicator. By implementing processes that look at all available data, not just one stream of information, manufacturers can forecast demand even if one piece of information is no longer valid, gets corrupted or is missing.
Processes have to be resilient so that missing, inaccurate or incomplete data does not create a catastrophic failure. Retailer inventory data is also well-known for having quality issues. In theory, next week's inventory is a simple math equation: inventory minus sales plus receipts. In practice average error for inventory is as high as 50%. The inherent error means that relying on a retailer's inventory number to generate production and shipments will result in a poor demand response if you follow the basic arithmetic. What is needed is predictive analytics that can find meaningful patterns in downstream data despite the high inherent error and processes that can deal with missing data.
Many consumer products companies have begun implementing this type of predictive analysis. Leaders are employing software with mathematical models designed specifically for supply chain analysis to automatically process downstream data on a daily basis. These new models analyze each data input to evaluate if that input is predictive of demand for that specific product and apply pattern recognition mathematics to develop forecasts for each product. For example, a large consumer products company may manufacture both a soap product and a snack food. These two products are both sold at the same large retailer, but they are consumed by different groups and consumption is based on different factors. The snack food may be seasonal, so demand is predicted by last year's sales during this season. The soap product sales may vary based on weather, time of year, current promotions or yesterday's sales. Using these new supply chain models, the manufacturer collects POS data, orders, shipments, inventory levels in the supply chain and at the retailer and historical forecasts. All of this information is analyzed and the mathematical models forecast demand using the information that most accurately predicts shifts in consumption for each individual product at each discrete location.
Harnessing the power of downstream data is the next frontier for manufacturers and the future of demand-driven supply chain management. Consumer products companies that master the data challenge and convert the masses of data into actionable information will make better business decisions and gain a competitive advantage. These companies will produce goods based on what is happening in the marketplace now by matching supply to actual demand -- in a real-time way that automatically adjusts to market volatility and helps protect companies in times of economic change.
Robert F. Byrne is President and CEO of Terra Technology, which provides supply chain solutions for consumer products companies.
Interested in information related to this topic? Subscribe to our weekly Value-chain eNewsletter.
