Subscribe to the Supply Chain Planning Blog

Keep up with the latest trends, research, and insights about supply chain planning, demand forecasting and inventory optimization.


Probabilistic Forecasting Can Extend the Life of SAP APO

By Jeff Bodenstab12 Dec 2017

Since the beginning of time – OK, since the beginning of demand forecasting — the standard approach has been a single number forecast that works relatively well with stable high volume demand. Traditional forecasting tools such as SAP APO, designed 25 years ago or more, generally hold their own in this environment.

But conditions have developed and grown over the years into areas where traditional signal number forecasting struggles to offer satisfactory results, and a new probabilistic forecasting paradigm has become the best approach. They include:

  • Aftermarket demand, spare or service parts, or MRO
  • Long tail, intermittent or lumpy demand, caused by revenue slicing from large product portfolios, shorter replenishment cycles or shorter product lifecycles
  • More channels to market, with aggregated demand coming from multiple “demand streams”
  • Online retailing, which typically deals with large product portfolios and less unpredictable demand

When any of these or similar conditions occur, aggregated time-series forecasting performance deteriorates, or can even break down altogether, escalating total cost of ownership (TCO). (See example below of Systagenix case study)

Firms are feeling the effects. “In today’s market, demand management success is like the flip of a coin,” says Laura Cecere, founder and CEO of Supply Chain Insights. “We see that most people think that demand management is very important, but only 44% feel that it’s effective.”

Cecere says that companies using demand planning software from ERP vendors have an even lower satisfaction rate. “One of the issues is that the engines in these tools are not as well-developed for the demand patterns we see in today’s market. Companies make the mistake trying to get precise on imprecise numbers. Instead, they need to manage demand flows—patterns caused by order frequency, order quantity or batch size, and types of demand, like trade promotions, new product launches, and seasonal consumption.”

And that’s what probabilistic modeling does. It analyzes a wide variety of demand inputs, flows and parameters, not as a single number aggregated forecast, but instead identifying the probability of a range of possible outcomes at a very granular level.


Under the Hood

An underlying problem with the traditional, top-down approach is that it smooths out variability from individual demand streams by aggregating demand. That makes it easier to generate a high-level forecast—but item-location level forecast quality is poor because specific demand signals are missed. For instance, two products with the same aggregated historical sales get the same forecast, even though a look at the details reveals widely different order patterns and therefore the need for different amounts of inventory to handle those different demands.When using the older top-down approach, a high-level forecast is made and then rather arbitrarily allocated down to an item-location level for inventory and replenishment—the level at which business decisions are taken. So crucial, granular information about volatility and error at that level is lost.

Probabalistic Forecasting - Why Demand Details Matter

Probabilistic forecasting understands there is inherent uncertainty in future demand, whether the SKU is a fast or slow mover. Its outcome—a range of values, each with a probability of happening—mirrors demand in the real world. Variability is part of the calculation, and the granularity of the baseline demand is as detailed as possible—by individual sales order line, daily by item and ship-to location.

Because probabilistic forecasting focuses on underlying demand patterns and causes, new influences or early signals that shape the forecast can also be added to the demand model – such as social media effects, trade promotions, and product life cycle profiles. Machine learning can refine the forecast by crunching external data.

The Probabalistic Forecast can be Foundation Layer for Full Demand Model

Making the Switch

One company who adapted probabilistic forecasting to their SAP APO is Systagenix. The company—part of global wound care company Acelity—combines both fast-moving and long-tail products with continuous new product introductions and global growth, constantly changing the dynamics and demand flows of its supply chain network.

Systagenix was using SAP APO for demand management with limited success. It didn’t have a structured way to capture market intelligence from its global sales organizations and distributors, and the planning team was always trying to second-guess other stakeholders and come up with a single forecast number. The company was struggling to meet the very high (approximately 98%) service levels its customers’ demanded.

A new, probabilistic approach pulls historical demand data weekly from SAP into a demand forecast that evaluates statistical variability for all SKUs. Demand planning gets additional inputs from sales and finance to generate a consensus forecast. This plan is used to create an optimized inventory profile, setting dynamic safety stock and target stock levels for each SKU at each DC according to the service level required. The data is then pushed back to SAP for DC replenishment and to inform production.

Systagenix squeezed service levels even further up to 99% yet significantly reduced inventory. It also dramatically improved the productivity of its planning team—from 2-3 people toiling for a total of at least 100 hours a week to produce the initial forecast, to one planner completing the task in a single day.

Subscribe to the Supply Chain Planning Blog

Keep up with the latest trends, research, and insights about supply chain planning, demand forecasting and inventory optimization.

Supply Chain Brief