Today hurricane Nate’s north winds are pushing against the bayous of Louisiana.
In offices across the United States, demand planners are scratching their heads. The impact of multiple hurricanes this season differ by commodity. What are they seeing? Shifting demand and rising prices for cotton and orange juice. Higher spikes for lumber and building supplies. Surprising demand in automotive for car replacement. A shortage of pharmaceuticals in Puerto Rico. One thing is clear. History is not a good predictor of current demand. They need accurate and timely channel data. They do not have it.
Sensing markets and translation of channel demand into an accurate demand signals for the company is the foundational principle of market-driven processes. Traditional processes are inside-out. They do not sense, or adapt to market shifts. While many technologists wave their hands advocating that the Internet of Things (IOT) is the answer, I say not so fast. When I hear this, I raise my hand to ask some basic questions. If we look back at history, 70% of companies implemented Vendor Managed Inventory (VMI), however two decades later only 1% of companies use VMI processes to drive a better demand signal. In most companies, sales teams use VMI processes to develop orders. They operate in isolation. The reason? Today’s demand processes are inside-out. In today’s architectures, there is no place to put outside-in data like VMI, Point-of-Sale (POS) or rating/review data. IOT will be the same. So I ask, “What will be different? Won’t we have the same issue for the Internet of Things data? Instead of jumping up and down on IOT shouldn’t we be discussing the redefinition of demand?”
Demand discussions raise emotions. I know of few areas in supply chains’ discussion that raise as much dialogue or ire. On one extreme, there is an argument which states forecasts are always wrong, “Why do them at all?” At the other end of the spectrum is the argument that “Forecast error is the most important metric to improve.” I am in the middle. I do believe in demand planning, but most companies overstate forecast improvements. Here I share my world view.
Everyone has a bias. Let me be transparent on mine. I worked for a software company for almost a decade and implemented demand management solutions in the 1990s for multiple companies. At that time, the demand processes were largely regional. I also worked in manufacturing during 1978-1992 trying to plan demand. I have been an industry analyst covering the market–Gartner, AMR Research, Altimeter Group and my own company Supply Chain Insights– since 2001. I have seen a lot of fads–CPFR, VMI, Flowcasting, Collaborative Sales Planning– they all come and go. Each was going to solve the problem of getting a better forecast. Not so. The gaps between what people want and what they have is great. This was the subject of my last blog post, Upending the Apple Cart. In that blog post I stated we must start by examining the apples.
Today’s demand management processes have many issues. They are not well-understood. Here are some top-of-mind issues:
- Bad Output. Many companies implemented demand management processes as a technology project. The output of demand planning engines were never validated, and the engines were never refined, honed, and tuned. As a result, the forecasting models are not a good fit to drive improvements. (The validation of the system’s output is an important step which is often overlooked.)
- Need for Tuning. The implemented demand management engines weren’t fine-tuned through regular testing. Like a car, demand engines needs continual tuning.
- Human Systems. At the time of the initial implementation, training happened. However, the understanding of the systems was lost through turnover and the lack of career paths.
- Executive understanding. There is a overstated belief that demand error can be greatly improved. In most companies, it is what it is. Bottom line: improving the forecast is possible within a range. However, companies should start by defining the reasonable range. Let’s take an example. In new product launch, while the error for new product launch improves through the implementation of demand planning processes based on attribute-based forecasting, the average demand error of new product launch is 75%. It is not feasible to reduce this error to 30% MAPE. Processes with high error need a design based on reasonable expectations, inventory strategies, and processes to absorb the error.
- Engines and Flows Need to Be Defined by Process. Most supply chain leaders can easily conceive supply flows, but not those of demand. Mapping the demand processes should align flows with technology capabilities. Techniques like attribute-based forecasting, probabilistic demand planning, attach-rate planning, and demand sensing are not well-understood. The design for most systems focused on high volume items that were easy to forecast. The answer is not 80/20. In demand planning forecasting the tail is critical to driving revenue. There was not equal focus on tuning the engines for the products on the long tail of the supply chain which are frequently high margin with lumpy demand patterns. As complexity increased (more items on the item master), forecastability (the ability to forecast) decreased. Most companies have not measured forecastability or aligned techniques based on demand flows. They also do not hold the planners accountable to improvement through Forecast Value Add. New business models, e-commerce, custom projects, and localized assortment make the demand pattern lumpier and more difficult to manage. Figure 1. represents the long tail.
Figure 1. The Long Tail of the Supply Chain
As a result, many companies have systems, but satisfaction is low (45% of the planners are satisfied with today’s technologies). Business leaders are questioning the head count in planning. Planners working long hours because the processes do not support the business requirements, question the processes, but they don’t know the answers. Because the systems and work processes are not aligned, most work in demand management happens outside of the demand-management systems using spreadsheets. Most demand planning happens in Excel Ghettos, not in the expensive technologies implemented in 90% of manufacturing companies. Because of these issues, loyalty with today’s systems is low.
I believe there is a need to forecast demand for tactical planning. The tactical planning horizon is from the slush period (order cycle period of confirming orders) for tactical planning (usually 10-18 months in the future). There is also the need for an operational planning period (usually 6-13 weeks in the future). Demand flows are the basis of the design. Each time horizon needs a redesign.
- Operational Planning for Demand Planning (Demand Sensing). Map the operational planning horizon to the tactical horizon to drive replenishment. Conventional systems use rules-based consumption. (This is the case for companies like Adexa, Logility, and JDA.) In the last year, demand sensing capabilities introduced by John Galt and OM Partners entered the market. (Test new solutions against the traditional demand sensing providers of E2open (Terra Technology), SAP and ToolsGroup. The capabilities of these technologies are not equal). In the operational planning horizon, demand sensing (the use of statistics and pattern recognition) replaces rules-based consumption to drive replenishment.
- Tactical Demand Planning. Use The tactical planning horizon to make asset decisions, determine the best network design, design form and function of inventory, and establish sourcing strategies. The tactical forecast forms the backbone of the S&OP process. Yes, it is true that the demand plan will always be wrong, understanding demand error and demand flows helps with building “what-if” capabilities.
- Strategic Planning. The most advanced companies map demand flows into network design processes/definition and align Applied Planning Systems (APS) to demand flows. This is ‘Planning by Design’. With companies having 5-7 ERP systems and 3-5 APS solutions, the definition of push/pull decoupling points, form and function of inventory targets, postponement strategies, and node definitions of factories, contract manufacturing and distribution centers within network design, cascades to the planning systems to synchronize the output.
While many argue that the definitions of the time horizons change with concurrent planning, I say not so fast. I think that within a global organization there is a need for a design group, an S&OP plan, and a replenishment process. Define the boundaries of the time horizons by work process definitions, not technology capabilities. Planning at companies varies by governance and cultural DNA. While concurrent planning is easier in a regional supply chain, collapsing the boundaries of the planning horizons, a global supply chain is more complex requiring more planning to achieve company goals.
Answering a Question
Recently I wrote a blog post on upending the apple cart. It outlined the gaps in demand management and the need to redefine forecasting. One reader rightly pointed out that while I drove the argument to change, I did not share an answer. With this as background, here I answer his question. He wrote, “While you make a compelling argument to change, what do I do?”
Step 1. Map Flows. To have the discussion, companies need to map demand outside-in, from the channel back. Demand data should not pool, and not be used in sales teams, VMI processes, and customer-data sharing. Channel data as input is essential. This includes structured and unstructured data like weather, social sentiment, rating/reviews, POS, and localized assortment decisions. Outside-in demand flows are market-driven to sense and adapt the demand to market changes with minimal demand latency. (Demand latency is the time from channel purchase to translation into an order. The longer the tail, the greater the demand latency, and the greater the distortion in the order pattern for modeling.)
Step 2. How Do We Plan by Design? Map demand flows into the network design activities and evaluate
Step 3. Redefine Tactical Demand Planning. Tactical demand management processes are a discussion of inputs, engines, data models and outputs. With the evolution of cognitive computing, machine learning, ontologies and data lakes. the world of demand planning will change dramatically. The building of a learning ontology enables the use of structured and unstructured data to drive machine learning/cognitive computing. Machine learning also helps with the cleansing of data to feed inputs from the data lake to eliminate duplicate and spurious data sets. Short term, a probabilistic engine might be used. (For a great overview of the use of this approach check-out the Spairliners case study from the Supply Chain Insights Global Summit. Probability forecasting is a great technique for lumpy and unpredictable demand. This case study is one of the best discussions that I have seen. )
Figure 2. Current and Future State: Tactical Demand Planning
Step 4. Redefine the Operational Horizon. Using statistics and streaming data, translate market data in the short-term horizon to redefine replenishment.
Step 5. Define Work Processes. After testing, define new work processes.
The shift reduces planning labor and improves the time to make a decision. It also reduces the required support team by 60-70% while improving job satisfaction. The evolution will take time. Today we only have experimentation.
There are many questions:
- Should companies place cognitive computing and machine learning on top of technologies like SAP APO? Should the ERP solutions be used as a system or record, or should cognitive computing replace existing technologies?
- Is it possible to combine revenue management/trade promotion management/ and demand planning into a new demand insights layer based on outside-in data?
- What is the right definition of concurrent planning capabilities within the redefinition of demand management?
- How should demand sensing technologies be mapped to manufacturing and distribution? What is the role of IOT? What do the processes of demand translation look like?
- How fast will this transition happen? What will it mean for early and late adopters?
- How will streaming data architectures combine with the statistics of demand sensing technologies to translate market demand to improve replenishment?
- What does this transformation mean for the planning teams? What are the required skill sets for the future?
What do you think of this vision? I would love to hear from you.
We don’t know the answers to these questions. The market is at an inflection point. Traditional vendors are improving conventional solutions through business analytics, while new best-of-breed entrants flood the market. To answer these questions and facilitate change, we are starting a new share group on demand planning starting in January 2018. The sessions will be held four times a year (face-to-face discussions) with monthly networking calls. In each session, I will organize case studies to drive ideation. The teams will use the experimentation of early adopters to work together to ideate on the future of demand management, and initiate testing with technologists to share with the larger group. The group will be no more than 35 business and technology leaders. (I will hand-pick the technology leaders based on their contribution to the industry.) There will be no sponsorship. Each attendee will pay a nominal fee of $750-1500 depending on the venue. (If we can use a host site at a manufacturer, the cost will be $750.) After we see the interest, we will pick a location and a date. (My thought is the midwest US to start. I am also open to doing this in Europe.) The format will be consistent with the work that we have been conducting on the network of networks. If interested in joining, please drop me an email at firstname.lastname@example.org.
For more thoughts on demand planning, check out these prior blog posts: