Today’s retailers and manufacturers are awash in data that more precise and available than ever before.
Much of newly available data requires that we use a massively scalable, network-based, real time platform in order to provide both a single version of the truth from a data integrity perspective as well as a seamless functional environment between planning and execution from a modularity perspective.
As an example, consider my company, One Network Enterprises, which provides this type of service across some of the largest corporations in the world today.
With this type of platform we can continue to use the longer horizon and more aggregate level mathematical techniques to create the traditional forecasting baseline, and then layer on the newer techniques working in tighter time buckets to drive forecast accuracy from the traditional 60% to over 80% at the SKU/shelf/location level.
Greater than 80% accuracy levels have been the norm at One Network customer sites for many years now.
The key is that with the newer network- based platforms, the business processes from forecasting to order management through replenishment and ship can react in ways that maximize customer service levels while also reducing inventory requirements and transportation costs.
Traditionally when choosing a platform, companies have used an Enterprise Resource Planning (ERP) environment with Bolt-Ons, or a Best of Breed solution integrated through an Enterprise Service Bus sharing a common Master Data Management system. Yet while the ERP might be able to provide a single version of the truth if rigorously controlled, it is not modular and therefore too difficult to extend and support. Furthermore, Best of Breed, while solving for modularity, is not able to provide a single version of the truth - causing significant duplication issues and associated data integrity problems.
Without an advanced platform, a ceiling of about 60% accuracy is the reality for time series-based forecasting. From a platform perspective this ceiling isn’t governed so much by computing power and storage (which has always been available in the old ERP architectures), but rather by the fundamental limitations imposed by information theory and the fact that the data being used to drive the forecasting algorithms in our traditional systems does not reflect current events or market conditions.
What is Demand Sensing?
“Demand sensing” has become increasingly important over the past few years. Why? Because companies believe that improving demand forecast accuracy can drive higher levels of customer service through better shelf inventory availability while simultaneously decreasing overall inventory costs and increasing profits.
Furthermore, evolving consumer behavior and rising market volatility have underscored the opportunity to sense and react in near real-time to changes in the demand and supply network, Yet these shifts have also fully exposed the limitations of our traditional forecasting techniques.
So what is demand sensing? Simply put, it is a next generation forecasting methodology that greatly improves current levels of forecasting by employing an updated set of mathematical techniques which are designed to analyze daily demand information. The result is a much more accurate forecast of near-term demand based on the current realities of consumer sell through.
This jump in forecast accuracy helps companies manage the effects of market volatility and gain the related benefits of a demand-driven value network, including more efficient operations, increased service levels, and a range of financial benefits including higher revenues, improved profit margins, decreased inventory levels, better order performance and a shorter cash-to-cash cycle time.
Evolving consumer behavior and rising market volatility have underscored the opportunity to sense and react in near real-time to changes in the demand and supply network.
The principles of demand sensing discussed in this post and those that follow apply across industries and more directly to any company participating in a supply network, including retailers, manufacturers, suppliers, or carriers.
From an infrastructure perspective, demand sensing platforms must scale to process the high volumes of data associated with hundreds of thousands of item and location combinations every day. The sheer volume, frequency and compressed processing windows require increased process automation along with the application of advanced mathematics to ensure that the resulting demand signal used to drive the execution environment is accurate and consistent.
Of course significantly increasing demand accuracy is only half of the equation. To gain the potential network benefits, the platform must support both a seamless environment between planning and execution as well as the ability to replenish the high frequency demand signals with optimized execution.
A New Frontier For Forecast Accuracy
At some point most forecasting methods will hit the law of diminishing returns where the forecast accuracy will tend to flatten out, regardless of the formulae or analytics that are applied.
Mathematicians have traditionally approached forecasting based on various time series and curve fitting techniques which create a forecast based on data such as prior sales history, drawing on several years of data to provide insights into predictable seasonal patterns.
Supply chain professionals have also come to the realization that past sales can be a poor predictor of future sales, especially when considering the variations associated with capabilities like distributed order management.
Yet with today’s rapid pace of new product introduction and shortened product life cycles, having more than an 18 month sales history might be considered a luxury. Supply chain professionals have also come to the realization that past sales can be a poor predictor of future sales, especially when considering the variations associated with capabilities like distributed order management.
Couple those challenges with all the promotional activities in today’s markets, as well as the availability of increasingly precise and frequent data, and you can begin to understand why yesterday’s math is having difficulty solving today’s problems.
Recently, forward thinking companies have begun employing improved mathematical approaches that can leverage the movement toward real time and network-based business processes along with their increased data volumes to create a completely new ceiling for forecast accuracy.