Better Decisions are in Demand: changing market conditions and AI innovations are creating new opportunities in demand-based decision making
CONTENTS
- Introduction
- Demand-Based Decision-Making Overview
- An Action Plan For Senior Executives
- Integrating Demand Predictions into Management Tools
- Conclusion
ABOUT THE AUTHORS
Tom Seddon — CEO, Predion.ai (by Foundry.ai)
Tom leads Predion.ai. Prior to joining Predion.ai, Tom was CEO, CMO for Extended Stay America, CMO for InterContinental Hotels Group and CEO of the Subway Franchisee Advertising Fund Trust. Tom holds a Masters in Data Science from the University of California, Berkeley and a Masters in Electrical and Electronic Engineering from the University of Bath, UK.
Scott Setrakian — Vice-Chairman, Foundry.ai
Scott leads Foundry.ai’s San Francisco office. Prior to joining Foundry.ai, Scott was co-founder and Managing Director of Applied Predictive Technologies. Previously, Scott sat on the Board of Directors of Mercer Management Consulting and ran the firm’s global Oil, Gas, Chemicals, Pharmaceuticals and Process Industries Consulting Group. Scott received an MBA and an A.B. in Human Biology from Stanford University. He sits on the Board of Directors of the Buena Vista Funds, the William Saroyan Foundation and the San Francisco Zoo.
INTRODUCTION
Businesses of all types — retail networks, restaurant chains, branch banks, grocers, airlines, doctors’ offices and hospitality chains to name a few — depend on accurate demand predictions to support a wide variety of key operating decisions on a daily basis. Decisions about how much product to order or build by SKU, hourly staffing levels by function and location, inventory quantities and even price points are driven in whole or part by an hourly, daily or weekly demand projection. The accuracy of this forecast is, therefore, a key ingredient in maximizing profitability.
Yet, demand prediction itself has for many years been considered a relatively ‘solved’ issue, addressed by models that were built around the turn of the 21st century.
It is only recently, as a result of several factors including new commerce channels, shifting consumer behaviors and improved technology capabilities, that this area has resurfaced as a source of both challenge and opportunity for the Global 2000.
In particular, we see four phenomena occurring simultaneously that are causing many companies to replace their legacy demand prediction and associated decision making systems with modern tools and processes:
Ultimately, all of these trends are underwritten by the rapid, ongoing reduction in unit costs for storing, transmitting and processing digital information. This is what has enabled the growth of e-commerce channels and the resulting changes in consumer behavior. It is what has created the digitization and accessibility of new forms of data. And it is what has allowed the practical implementation of new AI-based analytical methods.
DEMAND-BASED DECISION MAKING OVERVIEW
For many leading corporations, demand prediction was the AI of the 1990s. Advancements in data processing, storage and transmission efficiencies had reached the point that models could estimate future sales for a location with a reasonable degree of accuracy.
But limitations on the techniques and data available at that time imposed a ceiling on what ‘reasonable’ performance could be. The fact is that many business demand patterns do not neatly conform to textbook mathematical assumptions with, for instance, strong spikes from holidays, frequent trend changes from promotions or competitive actions, and seasonal demand variation, to say nothing of the significant ‘noise’ driven by otherwise random day-to-day differences. Combined with the operational need to generate demand predictions across hundreds or thousands of outlets or SKUs, these factors often encouraged the use of relatively simple statistical techniques, as any more sophistication was beyond the capability of the technology of the time to generate material improvement.
Nonetheless, these 1990s-vintage prediction models were generally perceived to be effective for a majority of the last 20+ years. With stable performance, companies moved on from trying to improve them. However, over the last several years, many companies — especially customer-facing ones — have observed that the accuracy of their models has been significantly eroding. We have heard this observation consistently from senior executives, and we believe that changing customer behavior, and in particular online commerce, is the predominant cause.
It is hardly news that online commerce has completely transformed sales patterns for every customer-facing business. As a channel, it now represents over 10% of all US retail sales and, in many product categories, it commands a significantly higher share. Moreover, e-commerce is growing at roughly 15% per year, while retail sales in the traditional channels are basically flat. There are also ongoing variations to customer behavior such as ‘order online and pick up at store’ that further complicate demand patterns.
The challenge posed by this dramatic change in channel strategy and consumer behavior is that legacy demand prediction models were developed and calibrated around the primary sales drivers relevant at the time they were built. But the importance of these drivers has shifted over time, and these models also often fail to consider newly available datasets that are highly pertinent to predicting more recent consumer behavior patterns. The more that underlying customer behavior has changed, the less precise these legacy prediction models have become.
In response to the decline of their prediction models’ accuracy, many companies have investigated the possibility of refreshing their legacy analytical methods with new datasets and models. However, with the passage of time, they have discovered some combination of:
- The models were written in code that is no longer in use or supported by the organization or original vendor; and/or
- Model documentation has gone missing or is incomplete or incomprehensible; and/or
- The software engineers that wrote the original models are long gone; and/or
- The current IT team has other pressing priorities and little appetite for complex revisions of a methodology that is unfamiliar and outmoded.
This challenge is hardly academic. Small differences in forecast accuracy have been proven to lead to significant differences in profits. Gartner benchmarking found that for every 1% forecast accuracy improvement — narrowing the gap between median and best-in-class performance — companies on average realized benefits including:
So — if forecasting is closely tied to profitability and many companies are grappling with a changing environment and declining prediction models, where should they be focusing in order to right the ship?
AN ACTION PLAN FOR SENIOR EXECUTIVES
Advances in data modeling and processing are providing companies with a unique opportunity to optimize their demand prediction and decision making capabilities, which will enable them to improve the many critical business processes that rely on accurate demand forecasts.
Our experience is that executive management teams implicitly understand when their demand predictions are not creating sufficient value, but are commonly daunted and exasperated by the challenge of upgrading. To most reliably achieve success, we recommend the following practical guidelines for executive management of demand prediction initiatives:
CONCLUSION
A dramatic shift in selling channels and consumer behavior is driving a recognition among senior executives that their demand predictions, and associated business decision processes, need an overhaul. Technological advances in AI hold the potential to make this not just a corrective exercise, but an opportunity to create material performance improvements. We have seen executives who follow the guidelines in this paper successfully capture this opportunity, creating significant value within surprisingly short timelines. We hope these concepts help you in this important domain and more broadly as you consider how to drive focused, practical and profitable AI initiatives within your organization.
