Churchill White Paper
Five Requirements For A Successful AI-based Retail Forecasting System
By Harve Light
Demand Forecasting: Absolutely Critical For Successful Retail Planning
From the beginning of time, forecasting demand (sales) has been a critical retail planning task as well as a major planning headache. Today, with larger retailers operating hundreds or thousands of stores in an omnichannel environment, the idea of forecasting demand on the back of an envelope is impossible. Whether they like it or not, today it is absolutely necessary for larger retailers to utilize technology for the generation of detailed, accurate demand forecasts. To most Retailers’ credit, demand forecasting for some planning functions, such as basic item inventory planning, is quite advanced. Realistically though, demand forecasting related to other retail planning functions (assortment, allocation, price, promotion, etc.} remains a critical retail challenge. Simply put, gone are the days of loose, high level demand forecasting. Competitors, particularly omnichannel retailers, utilizing advanced demand forecasting as part of their planning processes, have set a new and very precise retail planning standard. The purpose of this White Paper is to share Churchill’s list of retail demand forecasting solution requirements.
Five Requirements For A Larger Retailer’s Demand Forecasting System
To support any retailer planning function, the related demand forecasting solution must satisfy five system requirements:
- Support Of Very Large Retail Data Volumes.
- Support Of Conflicting Data Quality.
- Support Of Multiple Demand Forecasting Methodologies.
- Integration Into The Related Planning System.
- Acceptance By The Retailer’s Organizational Culture.
Requirement #1: Support Of Very Large Retail Data Volumes.
Core to retail demand forecasting is acceptance of the retailer’s extremely large data volume environment. For example, a Retailer carrying 30,000 items in 800 locations requires many billions of very precise demand forecasts per year. The addition of cloud technologies has greatly helped the management of large data but any retail demand forecasting application must take advantage of these resources that are necessary to handle the large data volumes.
Requirement #2: Support Of Conflicting Data Quality.
While it’s true that advances have been made in the accuracy improvement of some types of data, for example POS data. Unfortunately, today’s advanced retail forecasting methodologies (and planning environments) require significantly more types of data than just POS (think of promotional event attribute data, etc.). And, in today’s real world of IT, every retailer has some applications that are newer than others (think legacy systems). Additionally, demand forecasting modelbuilding (think AI machine learning) requires summarized historical data and we commonly find that a retailer’s data summarization standards were designed and implemented prior to today’s demand forecasting modeling requirements. A complete list of data challenges goes on and on. The point is that today’s advanced retail forecasting solutions must include functionality that supports conflicting and/or missing historical data.
Requirement #3: Support Of Multiple Demand Forecasting Methodologies.
Not all retail demand types fit the description of “Everyday Regular Price Demand”. Actually, there are many differing types of consumer demand. For example, a retail planning process might require a seasonal demand (with specific start/end dates) forecasting methodology. Promotional event lift forecasting requires promotional event attributes (or surrogates). Clearance pricing requires price elasticity modeling. And let’s not forget product line or item cannibalization demand and its additional data requirements. Today, for a retail demand forecasting solution to be effective, the forecasting solution must accommodate many different types of demand forecasting methodologies. One forecasting methodology will not fill the bill.
Requirement #4: Integration Into The Related Planning System.
Ask any retail planner if they dream about demand forecasting at night. Of course they don’t, but detailed and accurate demand forecasts are a critical component of their retail planning. Therefore, today’s omnichannel retailing world requires that the relevant demand forecasting functionality be able to effectively and efficiently integrate into the related retail planning application.
Requirement #5: Acceptance Within The Retailer’s Organizational Culture.
Unfortunately, for many reasons including personality and politics, retail organizations have traditionally resisted the introduction and utilization of technology-based demand forecasting solutions. An added complication is that demand forecasting results are never exact, they are always approximations. But since today’s retail demand forecasting cannot be approximated on the back of an envelope and therefore must include advanced technology, the successful retail demand forecasting solution must include both a solution and support path that will be accepted by retail organization.
Churchill & Demand Forecasting
For over two decades, Churchill has been providing advanced demand forecasting solutions to larger retailers. Churchill was one of the earliest proponents for utilizing AI Machine Learning (e.g. neural networks) to attack the challenges of retail demand forecasting. Through our planned product development cycles as well as customer research prototyping, Churchill has evolved a very advanced set of demand solution requirements and the related retail demand forecasting solutions required for today’s increasingly complex retail planning environment.
The fact is that today’s ever-changing world of omnichannel retail planning requires that retailers adopt advanced, technology-driven demand forecasting solutions. This list of five requirements provides a strategic checklist in your search for retail demand forecasting excellence. As you conduct your search for demand forecasting excellence, consider placing Churchill on your list of retail demand forecasting alternatives.
© 2019 Churchill Systems Inc.