What are the key critical issues in demand planning and forecasting today? How has pandemic disrupted planning? What assumptions and practices need to change to deal with uncertainty of the kind that pandemic brings?
There are two sides to this question. First there are new critical patterns in the industry over the past decades that require changes in how we plan and forecast. Second is the special situation of the pandemic. I will answer the second part now and we can discuss the first part after.
This pandemic is a black swan. It occurs once in a 100 years. From an operational perspective our primary tool is in response. But the forecast is important in how we predict the path after the pulse back to normal. It needs to filter out the pulse and rapidly identify new patterns. From a strategic perspective we do a lot to prepare for the next black swan and make our supply chain more resilient.
What are the different approaches to planning? What are your views on S&OP and IBP as an approach to help organizations become more supply chain resilient?
There are countless different approaches to planning. Most are hype or fad, some are really good. But we are still on a journey to find the perfect approach. Many parts of which we have discovered over the years and some that are getting traction today, they just need to be combined and complemented. Some broad patterns are that we need to move from periodic to responsive concurrent planning, remove functional silos (horizontal), connect strategy to execution (vertical), align metrics and incentives, and in general we need to remove the huge gap between our plans and reality. Some of the innovations along these lines are demand driven models and probabilistic math.
Violation of any one of these lessons guaranteed leads to the most expensive solution that does not solve business problems and causes all real work to occur on spreadsheets instead of the system
S&OP and IBP fit into this vision. S&OP is a key tactical piece that aims to create alignment to drive operational plans. IBP is the most misunderstood part. It is not S&OP on steroids. It is a strategic piece that drives outcomes. The bigger the company, the more important IBP becomes. In view of the pandemic, S&OP has a responsive role of how to ensure the path back to normal plays out, whilst IBP has a prescriptive role to ensure resilience for future events.
Currently in India and other countries a number of practicing professionals are advocating DDMRP as model for resilient supply chain planning? What are your views on DDMRP?
DDMRP is not any more or any less resilient than other approaches. It is however more responsive than most, allowing a quicker recovery. DDMRP is part of a larger DDOM model and complemented by tactical DDS&OP and strategic DDAE model. DDMRP cannot be seen separate from these other models. It is the easiest, most fundamental and mature part of the model so it gets a lot of attention. I see all of these as a great starting point, especially for companies still using MRP and those not having control over their processes yet. Supply chain is very complex. The DD model makes it easy to get started with the confidence that it is not a dead end street. That said, I do not think it is a magic elixir that will work best for all companies or for all parts of a company for which it is a good fit. In summary, I think DD will further mature and get complemented to fill those gaps.
What according to you is good and not so good in DDMRP?
DDMRP (or rather the larger model) provides 3 things
- A methodology. This is its greatest strength, allowing any company to adopt it quickly and easily.
- A cohesive planning and execution framework. This is another strength. This will allow companies to implement parts knowing that it will not become another silo.
- Formulas and approaches to do calculations. This is the main weakness. I would consider these starting points or educational to help gain conceptual understanding. Any company implementing a tool should do thorough diligence to determine if the math is sophisticated enough for their needs. The model is easily extended or tweaked, off-the-shelf commercial software that implements it is typically a different story.
What do you think of demand sensing? Are orders good enough to take care of the famous bullwhip problem? How do you think demand sensing be done in order to align the end-to-end supply chain? How do we measure variability? The variability that helps you plan your inventory buffers or safety stock appropriately and bring agility and build resilience?
DDMRP is not any more or any less resilient than other approaches. It is however more responsive than most, allowing a quicker recovery
Demand sensing to me is closest to true demand-driven. DDMRP is a misnomer, which should really have been called order-driven. Sensed demand, sales orders, and forecast are all demand. True demand is what the consumer really wants. When we sense demand, we try to get closer to the consumer. There are many variations, but in the most extreme we can sense cash register scans in the store, which is called POS DS. What this gives us is a demand signal that is more granular and known sooner. This is the only way to remove bullwhip effect before it reaches our company. DDMRP is then a way to remove bullwhip effect that we would add ourselves once we receive sales orders.
Two important notes one demand sensing:
- DS software typically does two things: sensing and forecasting. Sensing is getting the data further downstream. Forecasting is all the stuff with pattern recognition and statistical algorithms. That second part introduces inaccuracy and must be weighed carefully against other options whether it is beneficial.
- For many companies we cannot get data downstream. And true demand sensing is not a realistic option. The best we can do is the forecasting part daily applied to ship-to locations for the granularity benefits in forecast consumption. When we can get downstream data, it may not be available for every customer. In that case we need to carefully design how to blend that demand signal with other signals like sales orders.
You have been a proponent of probabilistic forecasting and have been advocating probability distribution in place of statistical forecasting and statistical error. What according to you is bad and good math here?
Whenever uncertainty exists, using exact numbers is naïve math. In some cases, you can get away with it, mostly at very high levels of aggregation where variability is relatively low and normally distributed allowing use of standard deviations and such as an adequate approximation. But in modern supply chains this will cause tremendous inefficiency and instability. “The forecast is always wrong” and “plans are infeasible by the time they are published” are direct result. Probabilistic math provides a model that reflects reality, and can produce forecasts that are 100% accurate and plans that will sustain the test of time much longer. In 10 years all forecasting and planning systems will be probabilistic or on their way to bankruptcy. Nobody will want a system that is always wrong when they can get others that are always right.
Can you give a simple example to explain what probabilistic math is?
Imagine weather forecasts 30 years ago. It either said tomorrow it will rain. Or it said tomorrow it will not rain. Today the forecast says there is a 40% chance of rain. If you were planning an outdoor kid’s party, you now can make an informed decision whether to make indoor backup plans. In the old days, the weatherman made that decision for you. And in the 40% chance he was wrong you would blame him for ruining your party.
Connected planning and control towers are here to stay, but they too will become more sophisticated in time.
This modern forecast comes closer to a probabilistic forecast. If a forecaster says sales will be 1000 units next month and it ends up being 200 he will be blamed for the error. But if the forecaster indicates there is a distribution of possible demand and 200 has a reasonable probability of occurring he cannot be blamed. Rather the decision makers would have picked an amount to stock knowing there was a chance it would be much less, or much greater. It would be a business decision, not a math error.
Can probabilistic forecasts help us build scenarios in the near and the short run and also build the long run?
Yes. A probabilistic forecast will provide time series of distributions. Every uncertain number is replaced by a full distribution. A probabilistic plan does the same: every uncertain number replaced by a full distribution. The traditional plans can tell you the average, but not the extremes. In the traditional approach every other scenario will need to be created explicitly. The probabilistic approach provides every possible scenario implicitly. You want to know the 5% lower confidence you can see it. You want the 1% confidence, same. The traditional plan is simply the 50% confidence level of the probabilistic plan.
Connected planning, control towers, and SAAS based solution – are they going to replace the need for all planning approaches? Or we can leverage them to build and broaden the horizon of planning?
I think historically grown approaches will slowly morph over time and may be complemented by new solutions or piecemeal parts may be ripped and replaced. Especially the parts that are core and fundamentally limited will be those that are replaced. Probabilistic math is one of the core foundations of any planning solution of the future. Systems will need to replace their math in newer releases and that would allow seamless upgrades for existing customers. This is going to be crucial for operational planning processes, but will be competitive for tactical and strategic processes too.
Connected planning and control
towers are here to stay, but they too will become more sophisticated in time.
SaaS today is primarily a means to deploy. It will help faster adoption by
smaller companies who do not have the hardware, know-how and workforce to
manage a complex SCM system landscape in house. In future SaaS has a lot
broader potential. Imagine demand forecasts that include big data, or services
firms handling planning in place of in-house planning departments, or huge
server pools in the cloud that can generate multiple scenarios on the fly
If you have to redefine this approach of demand forecasting and planning what would be your recommendation?
Naturally, I would strongly recommend to move to probabilistic systems sooner rather than later. But also before you decide on any one forecasting or planning solution first map out the larger needs of the company. If the best tool for one department for the current most urgent need does not fit within the larger strategy it must be discarded. The alternative is the introduction of yet another silo that will hurt business.
Design the ideal framework, from strategy down. Minimum requirement for any part will be to be an enabler for the larger picture. For this larger framework determine the key metrics. All lower level metrics need to serve the higher level ones, and all need to complement each other within functional processes and between them. Finally all incentives need to be aligned to the strategy and the metrics that serve it.
What’s your closing comment Stefan?
Learn from mistakes of the past. The most important ones:
- Do not replace one system by setting requirements of the next to match what the old one did. RFP’s are the enemy, at least how they are used today.
- IT is not a decision maker in planning system selection. They should have one non-preferential vote since they need to support it. The purpose of IT is the same as the purpose of the software: to enable business. The purpose of software is not to provide job security or career paths for IT.
- If you get advice from consultants or analysts, protect yourself. Do not take advice from parties that are incentivized to favor one approach or system. Very strictly separate parties that advise from parties that implement. For example if you use one company to help select a system, do not use that company to perform any work installing or setting up that system. Their incentive needs to be the best possible solution for you, not the most billable hours for them.
Violation of any one of these lessons guaranteed leads to the most expensive solution that does not solve business problems and causes all real work to occur on spreadsheets instead of the system.
Finally, supply chain and demand planning are exciting domains. You can set your business and yourself up for long-term happiness and gratification if you stay ahead of the curve. Enjoy it! ♦♦♦♦♦