Depending on who you speak with you will get varying definitions and opinions regarding demand sensing and shaping from sensing short-range replenishment based on sales orders to manual blending of point-of-sales (POS) data and shipments.
Most companies think that they are sensing demand when in fact they are sensing replenishment, and shaping inventory safety stock. Or at best, they are forecasting POS or syndicated scanner data, and manually blending it with sales orders or shipments. Also, if you're using basic time series methods like exponential smoothing you can only forecast demand, not shape demand.
To truly sense and shape demand, you need more advanced predictive analytics methods to model those factors that influence demand other than trend and seasonality. For example, price, sales promotions, in-store merchandising (feature, display, feature & display, temporary price reductions), store distribution, advertising, and more.
According to Oliver Wight’s book, Demand Management: Best Practices (2003), true demand is POS/syndicated scanner data, not sales orders or shipments. Also, if you are upstream in operations planning, then most likely you are too far removed from all the downstream marketing activities to understand those key sales/marketing programming tactics that influence demand. If you don’t have the capability to model those demand drivers that influence demand, then it’s impossible to shape future demand using what-if analysis. Demand sensing isn’t just forecasting the demand signal it’s about sensing those demand drivers that influence the demand signal, and then, using what-if analysis to shape future demand.
If you are manually blending POS/syndicated scanner data with sales orders, or shipments then the accuracy of your replenishment forecast (supply plan) will most likely not be very accurate.
To gain the most value from demand sensing and shaping, it's important to connect demand to replenishment or shipments using the Multi-Tiered Causal Analysis (MTCA) process. MTCA uses analytics methods, like multiple regression and/or ARIMA(X) as well as other advanced algorithms to sense and model the demand drivers. It's a two-step process.
- Model the key demand drivers that influence POS/syndicated scanner data, and then, run what-if analysis to shape future demand.
- Take the demand history and the shaped future demand forecast, and use it as a leading indicator in a sales orders, or shipment model.
This process provides a framework to model the push/pull effects of a company's business, and has been proven to significantly improve the accuracy of the supply plan, which is normally referred to as the shipment forecast.
In other words, you link the two-data series together through the data using advanced analytics, not manually blending them together. We all know any manual blending and/or overrides add personal bias whether intentional or unintentional increasing error.
The key is being able to do consumption based modeling using the MTCA process for thousands of products automatically up and down the company’s business hierarchy across markets, channels, brands, product groups, SKUs, demand points, and/or customers. It requires large-scale automatic hierarchical forecasting technology, that's driven by artificial intelligence (AI). AI uses a rule-based production system that automatically builds and updates the models as more data and causal factors are added to the system.
So, beware of software vendors who say that they can do consumption-based forecasting – demand sensing and shaping – as well as AI/Machine Learning. Only a vendor with robust experience in forecasting and advanced analytics, like SAS, can pull this off.
If you can only sense and forecast historical demand and sales orders or shipments six weeks into the future you need to ask yourself if that is really demand sensing and shaping? It sounds to me like replenishment forecasting and inventory safety stock optimization. I’ll let you decide.
So, are you truly sensing those demand drivers that influence the demand signal, and shaping future demand, or are you just forecasting demand and manually blending demand with sales orders to sense replenishment for short-range (1-6 weeks) supply planning?
Ꭲhere's certainly a great deal to know about this suƄject.
Ӏ ⅼіke all the points you've made.
Great article Charlie. I appreciate you highlighting the difference that MTCA can make in sensing / shaping organic demand vs. the practice of creating a replenishment forecast. This is the same conversation we had a few weeks back and I often find clients that conflate the two because Demand Planning is usually a function of Supply Chain vs. Sales. For companies that have drank the Kool Aid on DDMRP I do believe there is a large opportunity for improvement by using your two step approach vs. the standard net flow equation.
Thank you very much for the great feedback.
My concern with the standard net flow equation is that it pretty much ignores the forecast completely, and relies only on sales orders flow, which is very short-term. So, for those companies with long lead times it requires increasing buffer (safety) stock. Also, how do you react to sales promotions that can eclipse 50%-100% increase in demand? Particularly, for products with long lead times. New flow can easily go from 100 units a week to 10,000 units. or more. That would require a lot of buffer stock, or whip and raw materials. In all cases, increasing buffer stock, and/or incurring a lot of backorders.
DDMRP is just another name for MRP. It focuses on inventory buffer stock to address the symptom, not the root cause. The root cause is poor forecasting. Also, most companies are forecasting the wrong signal. Transactions (shipments) is the supply signal, sales orders is the replenishment signal. True demand is POS/Syndicated Scanner data.
Who is responsible for demand generation? Sale and marketing. So, why are demand planners reporting into the supply chain upstream from the consumer? Because they are not forecasting at all, and most definitely not planning demand. The are planning supply. A constrained demand plan (not forecast) is technically a supply plan. Those companies can no longer leave money on the table by making to a constrained demand plan (supply) in the new digital economy. Why? because smaller lateral companies will fill that unfilled demand taking away market share from larger companies.
Oh well, what do I know.