Does Your Supply Chain Need a Dalmation?

Dalmatians are the unofficial mascot for firefighters in the United States. The reason? The Dalmatian breed loves to run. They are committed to playing a part in putting out fires.

As I work with supply chain leaders, I find they love reacting. While I agree that perfection should not stand in the way of progress, I am bothered by what I see today in the industry. I see many creating and putting out fires without searching for the root cause. Much of the focus is on supply. Here, I want to focus on demand.

The Focus Needs to Be on Demand

GenAI models like ChatGPT are fast and easy to build. Many leaders I work with are enthralled by the empowerment of building/running their engines. It is so much easier than the drudgery of refining the traditional models in supply chain planning systems. However, few ask the question, "Is this model capable of driving systemic change?"

I had discussions with three traditional planning consultants starting their businesses this week. Each had retired and was starting a journey to build ChatGPT models. However, as we spoke, I found that no one that I talked to understood Forecast Value Added (FVA) and the need to use the Naive forecast to calculate the baseline forecast to understand demand shaping versus shifting. They were also not clear on the definition of demand sensing versus forecasting. With the evolution of the knowledge graph and the ability to map demand narratives, the focus needs to shift from a discussion on reducing error to improving FVA and decreasing demand latency.

I find teams want faster engine output without understanding the latency of the order/shipment data. For example, let's examine the impact of the product platform's complexity on the tail's elongation (low-volume products that are difficult to forecast.) As shown in the Figure below, for these products, the Coefficient of Variation (COV) is usually higher than .7 (making them difficult to forecast with conventional processes), and the translation of channel purchase to the visibility of an order (demand latency) is weeks and months. Yet, teams want to deploy GenAI to drive a faster math engine to move stale data in real-time. (Can you understand the Dalmatian reference now?) The engine is quicker, but the data is stale. This behavior creates fires.

Next week, I will start my fifth class to train supply chain leaders on using the power of the graph and demand narratives to visualize demand as flow (versus time-phased data). In this transition, students gain an understanding of:

• The impact of demand latency.
• Definition of baseline demand. Measuring shifting versus shaping. (Demand shaping increases baseline lift while shifting only moves product from period-to-period increasing cost without driving growth.)
• How to focus on FVA versus error.
• How to use disparate data to predict future demand.
• The need to use channel data and how to redefine demand models to use market data.
• The difference between responsive and reactive behavior.

Why does this matter? In my research, I find that only 20-25% of demand flows can be managed as an efficient supply chain (lowest cost optimization with functional metrics as an objective function using order or shipment data). (This was the design of the traditional supply chain model.) This percentage decreased over the last decade, but few question their models. Instead, like the Dalmatian, I find that many supply chain leaders want to run faster, chasing quicker engines without asking the question of whether the input data is relevant due to changing market conditions. My plea is that let's not apply generative AI to today's demand models. Faster engines on old paradigms will not drive better results.

Next steps? Here are five steps to take:

1) Learn the language of demand. Teach your team the language of demand. Get clear on terms like demand sensing, Forecast Value Added (FVA), techniques to build the naive forecast for comparison, demand latency, demand shaping, and shifting. Measure the success of current processes on improving FVA and shaping demand.

2) Do an inventory within your organization of channel data. A mistaken belief is that companies do not have channel data. I find this not to be the case. Most of the channel data is used by the sales account teams. In my work with clients, I find that neither the supply chain nor the IT teams know the availability of channel data or how to change the current modeling paradigms to use channel data. This is an opportunity for all.

3) Stop the funding on traditional demand models. Invest in experimentation with channel data using Graph AI.

4) Map your demand flows. Measure the COV and volume of your flows and study the impact of current processes on the FVA of the flow. Understand which items are forecastable by conventional means. Work through backcasting and testing to try to improve modeling. (Attribute-based forecasting, Attach-rate forecasting, changing position in the demand hierarchy for modeling, etc.)

5) Side-step the hype. Focus on real business results. Narrow AI in the right use case gives us great value. GenAI is useful for data discovery, insights, and data cleansing, but only if we build with the goal in mind. Move away from a nonsensical dialogue on trying to get perfect on imperfect data. Focus less on error and more on FVA.

In summary, we can redefine supply-centric processes to be more demand-driven with new technologies, but only if we learn to speak the language of demand and unlearn current processes. The first step is recognizing that the order is not a good proxy for demand, and any processes based on order data need to account for demand latency.

One thing is clear to me, the answer is not a faster engine on stale data. Demand data needs to be synchronized based on latency to make it useful.

I hope that your organization does not need a Dalmatian. They are beautiful animals that I would like to see in a warm bed by a fire in a beautiful forever home.

For additional information, check out these resources:

Building Outside-in Processes

Outside-in Video

Supply Chains to Admire Video

Supply Chains to Admire Report