Reboot Forecasting and Demand Planning

This article was originally published in the Supply Chain Alpha newsletter on LinkedIn. Read the original here.

A strategic function that needs a bold, fresh take to unlock growth.

Over the last few years, I have steadily noticed a degree of stasis setting in the field of forecasting and demand planning. Planners slowly giving up on the promise of technology, the reversion to excel, more meetings and bureaucracies to make decisions, teams looking to senior executives to divine out decisions and similar observations. If demand planning is strategic, why are we so stuck? I was determined to find out why. I spoke with over 70 Demand Planners and what I found was instructive in re-shaping my mental model.

---

"The Forecast is always wrong"..."Don't waste time forecasting, focus on using pull strategies"... "You'll never have perfect data, so why bother investing in AI for forecasting".

I have noticed that companies and teams that make statements like the ones above are also often the ones that haven't critically examined forecast and demand planning. But let's give them credit for having an opinion and who knows, maybe it comes from having genuinely tried but failed at using demand planning effectively. My goal today is to restore forecasting and demand planning as a frontier capability in the strategic toolset of the enterprise. And one that is poised to take off with the emergence of AI and Machine Learning.

Notice I use the terms forecasting AND demand planning - that is intentional because there is a distinction.

To keep it simple, I define Forecasting as the analytically rigorous aspect of converting historical information and relevant forward-looking inputs into a useful range of estimates of the future.

On the other hand, Demand Planning is an essentially collaborative, human process that injects human judgement, intuition and the ability to shape the future with levers we uniquely possess as innovators, marketers, investors and competitive beings. Demand Planning is a corporate function involving sales, marketing, finance and supply chain meant to crystallize the final, aligned enterprise demand plan that everyone executes to.

Ok back to that survey (thank you LinkedIn !). Here's what I found -

---

• Over 65% of respondents defaulted to excel in forecasting and demand planning
• Cross-functional processes like new product planning, promotions and competitive analysis were key barriers to accuracy
• Ability to integrate diverse data sets internally or externally is hard
• Scenario modeling is difficult or nearly impossible to conduct - excel use was common
• Difficult to interpret the outputs and even harder to prioritize Planner bandwidth around high value actions
• Teams are stuck with expensive solutions touting AI/ML features that are one-size-fits-all, with "stale" training and are losing relevance as the business shifts

---

The verdict is clear. Planners are tired of the friction across data, reasoning and UI layers. Friction that constantly blocks their flow, creates barriers to executing cross-functional workflows and ultimately, hurts speed and productivity.

Teams want a single solution which recognizes that -

1. Demand Planning is a multi-player collaborative sport, and facilitates end-to-end cross-functional workflows, and
2. Forecasting is a dynamic activity. It must have the ability to bring in rich data and context on-demand, deploy a library of AI models, conduct rapid scenario plans and re-train models in response to demand shifts.

This is a tall order. But the stakes are even higher. Why should you care as a leader or senior executive? Today, your organization's demand planning process is brutally siloed. Nowhere is this more prominent than in cross-functional processes like New Product Planning and Promotions Planning. Marketing teams are striving to find new vehicles to grow the topline, Finance is busy exercising control to achieve the financial guidance on revenue and margin, Sales teams are building revenue pipeline forecasts to meet their commits and Supply Chain is building statistical baseline to feed inventory plans. And every single team is operating to their own version of THE PLAN, none of which tie to what customers might be planning! This is an incredible amount of inefficiency and friction that threatens your ability to grow and innovate meaningfully. Is it really a surprise then if teams are scrambling to explain to you why the operating plan is so far disconnected from what you thought or why cost-to-serve and waste are egregiously higher than expected?

We must reboot this process and the underlying technology by re-architecting it from first principles.

---

Demand Planning - the second coming:

To restore Forecasting and Demand Planning to its strategic status, I believe we need to empower Planners, Finance and Marketing Analysts and Managers with tools that are analytically rigorous but malleable at the process level. I would emphasize the below as a practical roadmap:

1. Data is a central discussion point - particularly its breadth of scope and depth of quality. I think this is a red herring to the extent that teams often under-utilize the data that they already possess or could get their hands on. For instance, most teams create univariate forecasts using just a single variable like historical orders/shipments. This is painfully inadequate. Strive to bring in internal data like pricing, product attributes, lifecycle flags, instock levels etc. Critically examine what external data is crucial. For instance, Consumer goods companies need to leverage point-of-sale data, retail pricing, competitive prices, channel inventory levels - all this is available and can be brought in relatively easily. Manufacturers and Industrials can get distributor or channel data. Most companies can buy Syndicated data that provides macro trends and seasonality. Start somewhere and add as you go. Do not be paralyzed by broad brush statements about data quality and availability.

2. Forecasts are inherently probabilistic - Embrace the idea that we need to move forecasts away from brittle point estimates to probability distributions. Operating in ranges allows more realistic conversations between teams and is a critical input into the subsequent decisions in allocating inventory and capacity.

3. Leverage a library of AI/ML Models that are hyper-relevant and learn alongside your business - Don't settle for one-size-fits-all solutions. Deploy a library of models that can support seasonal, intermittent, trend, promotional, lumpy and other demand patterns. Too much energy is spent agonizing over forecast accuracy on a single time series, but far too little on uncovering the complex relationship across the inputs that ultimately affect demand - this is where AI/ML can really shine. Do not settle for mediocre models, demand excellence in model quality and diversity. That said, AI/ML models do will need to be retrained every once in a while as demand patterns shift.

4. Workflow Orchestration - The end-to-end process from data ingestion to final plan must be conducted entirely within the toolset. If employees have to take the process offline, it adds friction and adoption suffers.

5. Change Management: Do not underestimate how hard adoption of process and technology is. The best technology can be rendered ineffective because teams do not understand its purpose or how to use it every day. Invest early and often in communicating the end state, quantify the impact to the company and its customers and hold your technology providers accountable to train your teams. I will be writing more on this aspect soon.

But these don't have to be 5 separate efforts - these can be one thing. That is the bar we should aspire to in rebooting demand planning. That is what we are attempting to build at Strum AI. We are empowering supply chain users with curated libraries of AI/ML models that are hyper-relevant to their business and thoughtfully injecting AI Assistants for end-to-end workflow orchestration. This is an exciting time to reimagine outcomes with AI in Supply Chain. Email us at info@strum-ai.com to learn more or to get a live demo. We would love to chat with you.

Good Luck!