Integrated Demand Forecasting For Retail Demand
Demand Forecasting: Absolutely Critical To Successful Retail Planning
From the beginnings of time, forecasting sales has been a fundamental retail task as well as a major headache. Today, with large Retailers operating hundreds or thousands of stores, 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 large Retailers to utilize technology to get successful forecasts. To most Retailer’s credit, forecasting for some planning functions such as basic item inventory planning is quite advanced, but realistically demand forecasting most other retail planning functions such as assortment, price, promotion, assortment, etc. planning continues to be a retail challenge.
Churchill & Demand Forecasting
For two decades, Churchill has been providing demand forecasting solutions to large Retailers and organizations that support large Retailers. Through planned product development as well as customer trial and error, Churchill has developed a very advanced, integrated retail-planning demand forecasting system called “Churchill’s DFM Demand Forecaster™”. The purpose of this White Paper is to overview Churchill’s application footprint for DFM Demand Forecaster. Key components of DFM are also introduced.
Five Requirements For A Large Retailer’s Demand Forecasting System
Requirement #1: Support Of Very Large Retail Data Volumes. Core to retail demand forecasting is acceptance of extremely large volumes of data. A Retailer carrying 30,000 items in 800 stores requires over one billion demand forecasts per year. The good news is that computing and storage power have advanced to the point where large data sets are not a limiting factor when demand forecasting applications are built to take advantage of available capacity.
Requirement #2: Support Of Conflicting Data Quality. Great advances have been made in accurate collection of POS data. Unfortunately, much work still remains. Summarization of historical POS data remains an issue at many Retailers. And, collection of “causal data” such as promotional event data is very limited. We recently worked with a large Retailer who is currently supporting eleven competing Master Item Files (Resulting from acquisitions).
Requirement #3: Support Of Multiple Demand Types. Unfortunately, all retail demand is not basic, always-in-stock demand and therefore cannot be forecasted as if it were. Virtually every Retailer has seasonal, one-time and promotional event generated demand. Also unfortunately, each unique type of demand requires its own unique forecasting methodology(s). Traditional demand forecasting methodologies have focused on basic item demand forecasting and then utilized compromised or manual methodologies to support forecasting of non-basic item demand with the resulting inadequacies.
Requirement #4: Integration Into Related Planning Systems. Ask any Retailer if they live to generate demand forecasts. Of course, they don’t. Demand forecasting is a subfunction of a larger retail planning process such as inventory, price or promotion planning. This means that both storing resulting forecasts in data base formats that can be efficiently accessed from the Retailer’s databases as well as the capabilities to present formatted demand forecasts to the retail planning systems that they support.
Requirement #5: Acceptance Within The Retailer’s Corporate Culture.
Think about the emotions that can arise over demand forecasts e.g. promotional event forecasting. Often merchants compete with other merchants for adv space on the covers of weekly flyers. As part of the selection process they may have to justify that their promotional offering is going to sell better than the offers of other competing merchants. A sales forecast can be a part of their argument for inclusion and a computer-generated demand forecast may not present the sales predictions that would support their position. Today’s demand forecasting applications must present credible sales forecasts and retail organizations must learn to accept and utilize these forecasts to their benefit.
Churchill’s Forecasting Solutions Meet These Important Requirements
Churchill’s demand forecasting software solution, Demand Forecast Manager™ v3.8, includes all of the functional components that are necessary to meet the above listed application requirements. Demand Forecast Manager v3.8 includes modules for the following application functionality:
- Demand Forecasting Methodologies
- Demand Forecasting Model Initiation (Setup)
- Toolkits That Provide Supplemental Functionality
- Information Systems User Interface
- Business User Forecasting Process Interface
Demand Forecasting Methodologies
While there are many different perspectives that can be utilized to segment demand forecasting methodologies, Churchill utilizes the following three methodologies to forecast most types of retail demand.
- DFM Replenishment Demand Forecaster™ v3.8--Basic Item forecasting (Always In Stock)
- DFM Short Life Cycle Demand Forecaster™ v3.8--Life Cycle forecasting (Seasonal, Fashion, New)
- DFM Price & Promotional Demand Forecaster™ v3.8--Price & Promotional Event forecasting (causal or lift forecasting)
All three modules have multiple variations that will cover the Retailer’s demand forecasting requirements. Churchill’s strategy is to consider each module as a set of independent forecasting techniques that are designed to work together and provide a single, total demand forecasting capability.
Demand Forecasting Model Initiation (Setup)
All high-scale demand forecasting methodologies require formal initiation or “setup” functionality. And the importance of the setup functionality will vary by demand type. To properly accommodate these differences, Churchill’s DFM forecasting system includes unique setup modules for each and every major type of demand forecasting. For example:
- DFM’s Replenishment Demand Forecaster(RDF) includes the model selection functionality that is required to initially setup the forecasting function for “Always-In-Stock” items. RDF’s Forecast Model Optimizer is also utilized to refresh the initial forecast models on a regular reoccurring basis.
- DFM Short Life Cycle Demand Forecaster(SLC) includes a PreSeason or initial demand forecasting module. SLC PreSeason is very important for the demand management of seasonal categories. SLC PreSeason helps answer the questions: (1) What do we need to buy?, (2) When do we need the product? and (3) Where do we need to place the product?
- DFM Price & Promotional Demand Forecaster(PDF) is a set of causal-based forecasting methodologies and includes a setup module named PDF AutoTuner. PDF AutoTuner’s function is to integrate PDF’s PreTrained forecasting software with six months of a Retailer’s demand history for the purpose of configuring the Retailer’s price and/or promotional lift forecasting models sometimes building forecasting models as specific as item/store.
All of DFM’s initialization software components are high-scale, automated applications that are managed and adjusted based on forecasting performance metrics.
Toolkits That Provide Supplemental Functionality
All forecasting methodologies require some use of supporting toolkits. One example of a supporting toolkit is the DFM toolkit, DFM Profile Cluster Builder™ v3.8 (PCB), that builds and maintains the clusters and profiles necessary for constructing and grouping seasonal and size demand profiles. PCB call also be utilized to build price zone and assortment clusters.
Another example of a Churchill DFM toolkit is the DFM module that audits and transforms input and output data. This Churchill DFM Toolkit includes a set of data scoring algorithms that identify outliers and other anomalies needing additional review.
Without these application support toolkits, the job of demand forecasting in a large Retailer can become cumbersome and very manual. Additionally, demand forecast accuracy will definitely suffer.
Information Systems User Interfaces
If one thinks about the five system requirements listed above, one will recognize that an effective, efficient Information Systems User (and technical application management) Interface is imperative. Churchill’s DFM Operation Center v3.8 IT User Interface module elevates the technical management of the various demand forecasting methodologies to a more sophisticated level. As one example, it is a simple task to build a 1 or 2 row demand forecasting model, but it is much more difficult to build a 100,000,000 row demand forecast application that must produce appropriate forecasts within a prescribed, limited time window.
Functionally, Churchill’s DFM Operations Center™ technically supervises large demand forecasting applications including the acquisition of large amounts of required data history. DFM Operations Center also provides the facilities to embed unique or proprietary forecasting algorithms. DFM Operations Center is the “IT” core of Churchill’s demand forecasting solution.
Business User Forecasting Process Interface
Retail demand forecasting applications occupy a unique business role in that their purpose is generally to support one or more retail planning applications such as inventory planning, price planning, promotional event planning, merchandise planning, store labor planning etc. As such, many times the business user interface functionality is embedded directly into the specific retail planning application(s) that the forecasting application is supporting. A side benefit of this embedding process is the minimization of forecasting-related business process changes that need to be learned by the users. Emerging as an alternative business user strategy is the availability of demand forecasts through a demand forecast viewer that users and/or demand forecasting specialists can utilize to view and approve demand forecasts.
Churchill’s DFM forecasting system provides for both user perspectives. DFM can embed demand forecasts directly into retail planning application databases and alternatively, DFM includes a standalone demand forecast view/approval module called DFM Viewer v3.8 that is customer configurable and includes such functionality as filters, drill-downs, etc.
An Extra Benefit: Supplemental Planning Tools That Extend Forecasting Functionality
Frequently, Churchill is asked to provide extensions to its forecasting application that reach into the supported retail planning application. One popular example is markdown price optimization. Another example is inventory model stock calculation. Through DFM’s Operations Center functionality, DFM has the capability to provide and integrate these and other types of retail planning components.
With over twenty years of demand forecasting software development as well as implementing many large retailer demand forecasting systems, Churchill has surfaced lots of the application challenges and, as a result, has generated lots of functional ideas. Churchill professionals have distilled these demand forecasting requirements and practical ideas resulting in a comprehensive demand forecasting system specifically designed for larger Retailers. Today, Churchill’s DFM includes all of the components of a world-class demand forecasting system---forecasting methodologies, User and IT interfaces, supporting toolkits---all coordinated by a comprehensive operational environment. Churchill’s Demand Forecast Manager v3.8 is an economical, efficient and effective demand forecasting solution for larger Retailers.
© 2015 Churchill Systems Inc.