C-b4 White Paper Demand Forecasting Retail April 2011

Copyright © 2011 by Context-Based 4casting (“C-B4”) Ltd. All rights reserved.

March 2012

Predictive Analytics in Retail Systems

White Paper Context-Based 4casting (“C-B4”) Ltd.

Abstract: C-B4 provides automated data-analytics solutions for

retail systems. C-B4 applications include demand sensing, anomaly detection in sales data, forecasting control and validation, optimization of inventory levels including out-of-stock & slow moving items, smart classification of products and services, analysis of stores potential and root-cause analysis for demand changes. C-B4 tools identify hidden patterns in Key Performance Indicators (KPIs) in retail data.

Predictive Analytics in Retail Systems—White Paper

Copyright © 2011 by Context-Based 4casting (“C-B4”) Ltd. All rights reserved. Page 2

Table of Contents

1. Solution Overview 3

2. About C-B4 4

3. Background to the C-B4 Solution 4

3.1 Enhancement and control of forecasting systems 4

3.2 SAS – Proof of Concept 5

4. C-B4 Solution for Demand-Forecasting Systems 6

4.1 Improving the demand-forecasting accuracy 7

4.2 Combined products’ forecasting—Context-Based Clustering 8

4.3 Complementary products and Cannibalization—Context-Based Root-Cause Analysis 10

4.4 Retail networks—analyzing all links and subsidiary supply chain—Context-Based Monitoring 10

5. Summary 11

5.1 Comparison with Existing Technologies 11

6. Contact Details 12



1. Solution Overview

Context-based 4casting (C-B4) Ltd has developed a new technology—context-based predictive analytics—which detects anomalies and correlations in a unique way, much earlier than traditional performance-management applications. This useful early-warning signal saves analysts precious time when refining demand forecasting and fine tuning production-rate and supply-chain parameters, allowing the setting of inventory levels to cater fully and effectively to the actual demand. The new technology leads to a significant increase in the accuracy of the supply-demand metrics. It reduces the risks and costs associated with out-of stock, slow-moving demand items and high levels of safety stock.

More importantly, the new C-B4 technology allows modeling and analysis of data patterns in KPIs (Key Performance Indicators) of retail data, such as selling events, holidays, promotions, relations among SKUs, complementary or cannibalism effects, temporal dependencies, customer and product profiling, etc. C-B4 patented technology provides executives and analysts the unique opportunity to analyze and reveal hidden relationships between KPIs embedded within their existing operations and business data and between different products and categories.

Using the C-B4 engine to model operations and business data, it is now possible to achieve the high level of analysis required in today's competitive marketplace. Implementing a range of applications, C-B4 solutions transform raw data into valuable

business information by identifying hidden relationships and predicting anomalies early.

Traditional software monitoring tools often have the following shortcomings:

■ They find anomalies too late.

■ They do not detect many hidden anomalies that affect system performance.

■ They do not support a pattern based root-cause analysis to better understand what

corrective action is needed for a network.

■ They do not respond automatically to a changing environment and other dynamic

factors related to the demand level of products and services. Thus, models for

forecasting—even if they exist—quickly become obsolete. In such cases, it is

impractical to constantly involve statisticians and other experts in remodeling

dynamically changing demand-forecasting models.

■ They cannot handle the huge amounts of actual demand data. For example, they

cannot provide a unique model for each product in a supply chain, and thus cannot

distinguish between what is important and what is not.

■ They are based largely on traditional models that use unrealistic assumptions.

■ They rely on specific behavior patterns or on a limited set of models.

■ Most significantly, they do not take into account changes in the behavior of the

patterns, but instead focus on data thresholds that may trigger an alarm too late.

The C-B4 solution overcomes all these shortcomings. Section ‎4 describes the

suitability of the C-B4 solution for demand-forecasting systems.



2. About C-B4

The underlying, patented theory behind C-B4 solutions is based on rigorous and extensive R&D work performed over the past decade in the Faculty of Engineering at Tel Aviv University in Israel. Context-based 4casting (C-B4) Ltd was founded in September 2007 by a group of engineering professors. The C-B4 team includes experts in data mining, information systems, and algorithmic design (third-degree level). All team members have post-secondary degrees, with extensive experience in industrial

organizations.

Three months after the company's establishment, C-B4 sold its first predictive-analytics solution to Cellcom Israel (NYSE:CEL), a leading mobile phone company. Cellcom continues to implement C-B4 applications to control and monitor its network.

In parallel, C-B4 has also developed solutions for the retail industry that focus on:

■ Enhancement and control of forecasting systems

■ Analysis and optimization of inventory levels

■ Classification of products, clients and services

■ Evaluation of store potential

■ Root-cause analysis of hidden patterns among products, such as cannibalism and

complementary effects

C-B4 ran several successful analytic projects in large retail companies. For example, a Product Manager at a global retail software company noted:

“We did an internal test of the C-B4 engine’s results versus the (company)

clustering engine, and it proved that the C-B4 clustering and pattern analysis

technology can significantly improve our current forecasting results.”

C-B4 has several projects and joint ventures with international companies such as General Motors, Applied Materials, and Proctor and Gamble. C-B4 is cooperating with SAS and with Oracle Israel.

3. Background to the C-B4 Solution

At the core of the C-B4's patented technology stands a unique network-modeling engine that provides the basis for analyzing the dynamics of Key Performance Indicators (KPIs) within complex systems and processes. The engine’s construction algorithm automatically optimizes the size and the statistical efficiency of the network model.

The C-B4 network model provides a compact description of all the significant patterns in complex systems and processes. Once the model is constructed, it captures all the significant dynamics and dependencies in the data. This information can be used by various applications, such as monitoring, prediction control, and classification and clustering, as explained below. The C-B4 network model obtained from this analysis is flexible, making it possible to implement different modules in a variety of demand-forecasting domains.

3.1 Enhancement and control of forecasting systems

Understanding and predicting customer demand is vital to manufacturers and distributors to avoid stock-outs and to maintain adequate inventory levels. While forecasts are never perfect, they are necessary to prepare for actual demand. To maintain an optimized inventory and effective supply-chain metrics, accurate demand forecasts are imperative.



Traditional systems are based on historical information and use statistical models to generate a stock forecast. The level of confidence we have in this forecast determines whether we will act according to the predicted value or not. Often, the confidence level of a prediction is not as high as analysts may desire, and certain corrections—either human or automatic procedures—are needed.

In addition, retail or vendor management must deal with the delivery of hundreds or thousands of products, goods, and services, and it is impractical for them to monitor all the forecasts all the time.

The C-B4 solution provides a reliable indication of where to look and where to direct one’s efforts to improve the accuracy of demand forecasting. The unique C-B4 technology enables us to discover crucial information about historical data and the way a prediction was made based on that data. C-B4 technology monitors and models patterns of data used to generate prediction values. It can identify cases where although the data values are acceptable, the patterns of data prior to the "prediction" event indicate some abnormality. This information is used to generate warning signals, prompting us to examine the demand forecasts more closely, and to employ further refinement algorithms to them.

3.2 SAS – Proof of Concept

SAS is a leader in business analytics. In a joint case study, SAS and C-B4 analyzed the retail data of a world-wide retail firm with hundreds of products for home usage. In this case study, the objective was to use the C-B4 engine as an enhancement to SAS forecast server, which is often considered the best in the market.

R. Bello Lapidor, Sales Manager at MIA Computers (SAS distributors in Israel) remarked:

“We checked C-B4 technology with SAS worldwide and did not find currently

a software tool that provides the predictive-analytics capabilities generated

by C-B4 tools. This is particularly true regarding the C-B4 ability to detect

anomalies in large datasets without requiring preliminary assumptions, and to

find complex relationships among a large number of KPIs.

To summarize, we strongly believe that C-B4 technology can provide an

excellent complementary solution to those obtained from the SAS systems,

and accordingly we look forward to continue our cooperation with C-B4.”

SAS software has been used to conduct an ETL (Extract, Transform, Load) process on the data. Then the SAS forecast server constructed a demand model for each product and category. Finally, the C-B4 engine was applied to the forecast results, and it achieved the following results:

■ Provided early detection of anomalies in data patterns, which were the source for the

generated estimation.

■ Generated alerts where special intention needed to be given to a specific forecast,

showing the ability to identify one out of thousands of products that needed correction.

■ Directed the users to understand the root-cause of the low level of confidence given to

a prediction by a traditional approach.

C-B4 provided early predictions of problematic forecasts that could be dealt with in real time by a product analyst. Some of these problems could have been detected by human interference, but this would have been extremely time-consuming. The C-B4 solution accurately identified the time of the change point in the data, enabling quick root-cause analysis.



The following figure illustrates a real example for controlling a demand-forecasting system. The red dots (circled at the top right) provide an early-warning indication that the forecast model is no longer valid for the considered product. A substantial time before the actual sales (marked by the black dots across the graph) exceed the forecast threshold, the C-B4 early-warning indication has already anticipated a problem.

Figure 1: Real-life prediction of an anomaly (a few days before the actual event)

4. C-B4 Solution for Demand-Forecasting Systems

Demand forecasting is a crucial process for retail networks, and encompasses all the links of the chains and sub-chains: raw materials, inventory, and stocks (both at the vendors and at the producers), on-shelf inventory, and much more.

Since most retail firms use some demand-forecasting software, achieving a competitive advantage is the desired goal of every demand analyst. Experience shows that 30–50% of forecasting is inaccurate, and this influences the accuracy of the aggregate, thereby affecting forecasting performance and inventory costs. C-B4 improves this by identifying bad forecasting and exceptions in demand, relative to the normal behavior. By performing root-cause analysis, the user can identify the causes of change in behavior and incorrect forecasting. C-B4 then helps to indicate where these forecasts need to be further refined.

Today, the existence of a forecasting system, either for retail network or its subsidiary

supply chain, is a basic operation tool. In order to plan the production rate, inventory

levels, transportation needs, and other factors, it is crucial to have a demand-forecasting system. However, the standard, traditional system is statistically based and has inherent flaws:

■ There is an assumption regarding some known distribution model, and historical data

is used to extrapolate future estimations.

■ The way by which "irregular values" are treated is not always clear. Many times they

are excluded so as to avoid deviations in the generated values. Other times they are

simply “smoothed” out.

■ The generated forecasts are usually based on an aggregation of values.

All these and other reasons result in a low level of confidence in the forecasts. The competitive advantage of a demand analyst is to achieve more accurate forecasts.



C-B4 can dramatically improve the level of confidence, since its unique technology is based upon modeling the actual pattern of the historical data used to generate the forecasts.

C-B4 context-based monitoring is designed to directly indicate exceptions, thus reducing the time taken to detect and respond to changes and to produce accurate short, medium, and long-term forecasts.

Identifying abnormalities in the actual way any prediction model works—by relating to the very specific demand pattern of each product—gives range of possible actions to be implemented by the demand analysts. The first distinguishing fact about the C-B4 methodology is that even if all values are within "acceptable" limits, and thus taken into consideration by a standard prediction module, C-B4 can detect whether the pattern of the data is abnormal, and indicate whether special care is needed for the specific case.

Another achievement of C-B4 is that since the user looks into the very heart of each product’s demand pattern, additional information regarding the demand may be manifested. This information can be of business value and enables the user to extract more competitive advantages over other forecasting systems. An overview of C-B4 technology “Smart Forecasting” is presented in the following figure.

Figure 2: Smart forecasting solution integrated with C-B4 technology

4.1 Improving the demand-forecasting accuracy

This solution module facilitates accurate forecasts across a variety of demand types (fast, slow, seasonal, erratic, etc.) by indicating exceptions.

C-B4 monitoring detects anomalies in data sequences, while maintaining relevant and readily interpretable results. The unique pattern-recognition monitoring module enables the users to detect anomalies even when all analyzed values are within acceptable limits.



C-B4 technology learns and models the demand data of each product. The actual demand level of an individual product or service is monitored. If provided, this can be done with KPIs like purchase location, quantity, time, price, stock at hand, and more. For each product, a specific model is generated, taking into account the pattern of all these parameters jointly.

The C-B4 monitoring module issues alerts regarding exceptions and residual patterns. These elective indications and their corrective actions (either human or automatic) lead to more accurate forecasts. These forecasts form the basis for analyzing actual demand versus the refined forecast for individual products.

C-B4 automates the process of analyzing complex demand patterns from multiple sources. By directing the user to exceptions, C-B4 greatly reduces the time taken to detect and respond to changes and to produce accurate short, medium and long-term forecasts.

Context-based monitoring can be used to:

■ Monitor complex demand retail chains, involving seasonality, time of day, place of

purchase, other SKUs sales, events such as holiday and promotion, and more factors.

■ Detect changes in demand pattern, which are exceptions that require special care, for

example, the beginning of an out-of-stock situation, pick sales, safety stock too high,

and slow-moving inventory items.

■ Find anomalies in KPI correlations, even when the KPIs themselves behave normally,

between products, categories, point of sales, client types, and more.

■ Pinpoint the generated forecasts that require refinement out of hundreds or thousands

of monitored products.

The monitoring of single product or service:

■ Supports for both slow and fast-moving SKUs.

■ Manages multiple KPI metrics of demand for a single product, category or service.

■ Gives higher accuracy and supports dynamic manufacturing/inventory policy in line

with real demand, especially with many exceptions.

C-B4 may be applied to:

■ A single product or multiple product demand forecasts

■ A single product demand forecast for a specific sale point or in multiple sale points

■ A design of service providers (for example, a call centers or cash registers)

■ Many more situations

The cost effectiveness of C-B4’s higher accuracy and optimized forecasting stems from less dead inventory, fewer out-of-stock situations, and in general, a higher confidence level for the estimations generated by the forecasting system.

4.2 Combined products’ forecasting—Context-Based Clustering

C-B4 technology may be integrated in a demand-forecasting system to model the actual behavior of demand or inventory levels of a group of products or services.

Monitoring the actual demand and inventory levels across various products and services provides knowledge regarding cause-and-effect phenomena. For example, finding hidden dependencies in the demand levels of different products can reveal cannibalism effects or complementary products. A correct business strategy can then be implemented. Figure 3 shows the demand level of a cluster of 87 products that are associated with each other



temporally – during the holiday period only. Such clustering is based on pattern analysis and cannot be found by conventional clustering methods.

Figure 3: A cluster profile of 87 products in a US retail chain during a holiday

C-B4 technology can provide a full understanding of the impact of a particular promotion (in addition to known aggregated data such as average demand). This allows us to group products and services in a very accurate way.

Since C-B4 deals with patterns of data and can correlate data such as time, date, place, coupon-related purchase, and many more KPIs, and since the C-B4 models detect both time-domain and space–domain correlations, clustering the appropriate products and services can be done to provide a competitive advantage.

External influences to demand patterns, such as promotions and new product introductions, are also managed effectively within the system with the minimum of human input.

Context-based clustering is applicable in the following areas:

■ Global planning helps manage stocks in complex multi-echelon supply chains with

multiple stock locations and suppliers in the most effective and optimal way. Global

planning assists in the decision-making process of where stocks should be held in the

supply chain. Stock allocation and stock re-distribution proposals become more

accurate.

■ Strategic modeling enables 'what-if' analysis to help with modeling various logistics

scenarios, including the impact of changes to planning strategies, product

segmentation, alternative suppliers, etc. Any new strategies that may have an impact

on stock investments or client service levels can be simulated and the results can be

analyzed prior to implementation. The "real-life" pattern recognition of a group of

services and products gives strategic issues a more reliable confidence level.



4.3 Complementary products and Cannibalization—Context-Based Root-Cause Analysis

C-B4 technology can analyze the correlation between various Key Performance Indicators that characterize a given demand-forecasting system. Since the specific pattern of demand for each product is monitored, information can be generated regarding the correlated effects between products.

More importantly, since the detection and correlation of anomalies is based on patterns and not on aggregated data, root-cause analysis can be implemented "in advance" or "on the fly". We don't have to wait for the end-of-day results or end-of-sale information and make a post-facto analysis. Rather, the analysis can be done periodically by the C-B4 engines as the data is received, and any indicated trends will be the basis for the root-cause analysis. This analysis can provide an understanding of which KPIs represent an anomaly at a certain time or within a given time lag. This information can be used to:

■ Understand the correlation between different products, such as competing

(cannibalization effect) or complementary products.

■ Applying a central root-cause analysis tool for policy determination—the ability to

compare various sectors' patterns, for example, may provide information for root-

cause analysis.

■ In-advance detection for corrective and preventive action. Since C-B4 detects

abnormalities even when all values are within control limits, one may react even

before a phenomenon has reached its peak, and thus reduce the impact of non-

effective cannibalization, or enhance effects such as complementary products.

4.4 Retail networks—analyzing all links and subsidiary supply chain—Context-Based Monitoring

C-B4 technology can analyze a complete retail network, meaning actual demand and all the relevant supporting processes, such as marketing, production, supply-chain logistics, and point-of-sale data.

Since C-B4 engines are scalable, and operate in a cloud-computing architecture, it is very easy to join together as many KPIs as required, and analyze a whole retail network operation. KPIs from the point of production, inventory levels along the supply chain, transportation alternatives, and even shelf setting at the final point of sale, can be modeled together to find the correlation between all of them.

Since the model is dynamic and self-adjusting, it can produce a different correlation for the same products on a different supply chain or in a different point-of-sale situation.

Understanding these correlations (most of them are hidden and cannot be distinguished by regular tools) is of notable business value. Some applications are:

■ Applying a central "war room" for a retail network—dealing with on-time corrections for

the production rate, transportation needs, levels of inventory, suggested promotions,

and marketing steps.

■ Providing an ever-improving and dynamically calibrated mode of operation for the

retail network. Based on some regular rules, the abnormalities and correlations

stemming from the overall analysis is a source for on-the-fly orders that maximize the

effectiveness and efficiency of the network.



5. Summary

The following C-B4 features and benefits can boost business intelligence efforts:

■ Unique pattern-based solutions—C-B4 models and monitors both individual and

groups of products, demand values, and KPI patterns. Only the C-B4 solution uses

context-based modeling to detect significant exceptions in the data patterns. Unlike

other tools that indicate some change in a process if a KPI value exceeds a particular

statistical threshold (for example, a very low confidence level), C-B4 can automatically

identify a change in a model even when a KPI maintains its regular values, yet begins

to follow different patterns.

■ No need for preliminary knowledge—C-B4’s solution does not require any a-priori

knowledge or assumptions about the system or the process distribution or dynamics.

The C-B4 modeling data structure is generic and can automatically represent a wide

range of both simple and complex processes.

■ Adaptive Learning—The C-B4 machine algorithms adaptively learn the modeled

process at any time and for each device. This is a key feature in an ever-changing

environment. Unlike other tools, there is no need to rescale or remodel the tool, even

if the monitored system is rapidly changing. Moreover, by using C-B4 solutions,

predictive maintenance is tailored to each monitored component or KPI individually.

■ Simple to use and simple to implement—The construction of the data model as

well as the parameterization of various applications are transparent to the end-user.

There is no need for statistical consulting or model validation prior to the

implementation of C-B4. This feature leads to considerable financial savings

compared to other tools on the market.

■ Integration with traditional IT tools—The C-B4 can be integrated with various

databases and IT tools, such as Oracle databases, performance-management

software, and MS Excel, enabling easy implementation and modification of the various

C-B4 outputs.

5.1 Comparison with Existing Technologies

The following table provides a summary of the differences between C-B4 technology and existing technologies.

Characteristics Existing Technologies C-B4 Technology

Data presentation: Specific statistical models Generic statistical models

Monitoring of: The data itself Data patterns and data itself

Presentation of results:

Graphs, control diagrams, and more

Graphs, control diagrams, user-defined filters, and more

Updating and learning:

Requires periodic investments

Automatic "learning" model

Adaptation to sub-system components:

Adaptation to all sub-systems separately according to fixed model principles

Individual adaptation to each KPI, i.e., predictive maintenance is tailored to each component

Number of simultaneous KPIs:

Usually one dimensional Multiple KPIs



6. Contact Details

For further information, please contact:

Email: [email protected] or [email protected]

Phone: +(972)54-324-7220

Cellular: +(972)54-495-4465

Fax: +(972)9-955-2784

Website: www.c-b4.com

mailto:[email protected]

mailto:[email protected]

http://www.c-b4.com/

C-b4 White Paper Demand Forecasting Retail April 2011

Documents

retail systemswhite

key performance

subsidiary

produce accurate

retail systems

b4 technology

low level

retail networksanalyzing