Predictive Analytics for Retail: Understanding Customer Behaviour

Transcription

1 Predictive Analytics for Retail: Understanding Customer Behaviour Friday 10 th October 2014 London, Southbank

2 Agenda 9:30 10:00: Arrival, Registration & Coffee 10:00 10:20: Introduction: What is predictive analytics and is it realistic for you? 10:20 11:00: Applying Advanced Analytics in Retail Predictive Applications for Retail Customer Examples 11:00 11:30 Refreshment Break 11:30 12:15 Technology overview: Demonstration of tools and techniques :30 Getting started: what to consider and how to begin Lunch

3 Premium, accredited partner to IBM specialising in the SPSS Advanced Analytics suite. Team each has 15 to 20 years of experience working in the predictive analytic space - specifically as senior members of the heritage SPSS team

4 Current Themes In UK Retail Retail is important to the UK economy, with turnover in 2012 of 310 billion The value of internet retail sales in 2012 was 29 billion, around 9% of total retail sales Seven retailers (between them worth around 76 billion) are in the FTSE 100, and many pension and investment funds have significant retail holdings It is the largest private sector employer around 3 million people

5 Current Themes In UK Retail Examples of UK retail news in the last 14 days

6 Current Themes In UK Retail Hot off the press Evidence of changing consumer behaviour in retail

7 Current Themes In UK Retail Source Deloitte

8 So how can Predictive Analytics help? Demand Side attract grow fraud retain risk

9 So how can Predictive Analytics help? Supply Side Product Lifecycle Planning Forecasting & Pricing Internal fraud Store / channel segmentation & profiling risk

10 Predictive Analytics for Retail

11 What do we mean by Predictive Analytics? Predictive analytics encompasses a variety of techniques from statistics and data mining that analyze current and historical data to make predictions about future events Analysis of structured and unstructured information with mining, predictive modeling, and 'what-if' scenario analysis.

12 What do we mean by Predictive Analytics? It s different from Business Intelligence or MI reporting Actually, it s not always about prediction However, Predictive Analytics does creates important new data These data take the form of estimates, probabilities, forecasts, recommendations, propensity scores, classifications or likelihood values Which in turn can be incorporated into key operational and/or insight systems

13 Types of Predictive Modelling Predict a particular type of outcome Propensity/ Classification Clustering Identify groups within a population displaying homogeneity (based on a wide array of data) Forecast a future value over a defined time period Time Series Association / Sequence Identify repeatable patterns of behaviour or sequence

14 Types of Predictive Modelling Classification / Propensity Who is most likely to respond / convert based on historical response data and the array of behavioural data we have about them? Clustering How can I divide my prospects/ client base into meaningful and usable groups as a framework for marketing planning / customer insight? Association & Sequence What is the optimal sequence & frequency of events and interventions that lead to a response / purchase / dormancy? Time Series What will site traffic / revenue be next day / week / month?

15 Analytics in Retail Segmentation Life Time Value Loyalty Purchase Behaviour RFM Store clustering Propensity Modelling Response Cross-Sell Churn Reactivation Voucher Redemption Other Applications Affinity/Basket Analysis LTV prediction Forecasting Text Mining Satisfaction Modelling

16 Why is this important to organizations? Acquiring customers is expensive Not unusual to cost 6 times as much as retaining them Understanding who is most likely to convert is very cost effective 80% of a company s profits come from 20% of its customers Need to understand these customers needs How they behave and what keeps them happy Increasing customer retention rates by 5% increases profits by 25% to 95%. Study by Bain & Company, working with Earl Sasser of Harvard Business School Customer Lifecycle Economics is amplified for e-businesses (B2B & B2C) by Frederick F. Reichheld and Phil Schefter

17 Even when the application isn t a predictive model We can still use historical data to better understand our customers Find strong correlators or drivers of behaviour Carry out text mining and analyse changing sentiment Develop a segmentation strategy that is data-driven Go beyond recency, frequency monetary to incorporate Who Demographics What Product/Service Categories Interactions Website, Social Media, Call Centre, Payment methods History Tenure, Customer Journey,

18 Competitive advantage How do our clients maximise success? By exploiting a wide data landscape Social Media Data Interaction Data Descriptive Data Transaction Data Degree of intelligence

19 How do our clients maximise success? By utilising a powerful, proven methodology CRISP-DM: Cross-Industry Standard Process for Data Mining Each application can be developed and progressed through a series of key phases

20 Is it practical for your business? Do you need to be statistically trained? No Appropriate skills can be transferred and learnt Business understanding is just as important as technical skill Analytical skills can be taught Data readiness? You need some data, certainly Does not need to be organised into a tidy single supporter view Data is often [always] fragmented and disparate (occupational hazard)

21 Is it practical for your business? How do we get started? You can start small and develop in phases Demonstrate initial ROI before releasing further investment Plan your first projects using a recognised analytical methodology Minimum set up components Key analytical tools implemented Data access, management & manipulation exploratory analysis, profiling & modelling capabilities Offline deployment Skills transfer and knowledge to continue independently First project (s) completed Actionable results, deployed in the business

22 Applying Predictive Analytics in Retail

23 Advanced Analytics in Retail John McConnell

24 Objectives To talk about some of the things that have been done in retail with AA and to introduce more concepts to help you work out what (else) you might do

25 SPSS Retail Users Include

26 The 3 pillars top level AA types

27 Predictive Customer Analytics: Driving Actionable Insight Acquire customers: Understand who your best customers are Connect with them in the right ways Take the best action maximize what you sell to them Grow customers: Understand the best mix of things needed by your customers and channels Maximize the revenue received from your customers and channels Take the best action every time to interact Retain customers: Understand what makes your customers leave and what makes them stay Keep your best customers happy Take action to prevent them from leaving

28 Predictive Operational Analytics: Driving operational Efficiency Manage operations: Maximize the usage of your assets Identify the impact of investment Ensure inventory & resources are in the right place at the right time Maintain infrastructure: Understand what causes failure in your assets Maximize uptime of assets Reduce costs of upkeep Maximize capital efficiency: Improve the efficiency and effectiveness of your assets Reduce operational costs Drive operational excellence in all phases: procurement, availability and distribution

29 Predictive Threat & Fraud Analytics: Driving mitigation & prevention Monitor environments: Identify leaks Increase compliance Leverage insights in critical business functions Detect suspicious activity: Identify fraudulent patterns Reduce false positives Identity collusive and fraudulent merchants and employees Identify unanticipated transaction patterns Control outcomes: Take action in real-time to prevent abuse Reduce Claims Handling Time Alert clients to transaction fraud

30 Some more example retail apps Customer Campaign response modelling All the web site stuff All the call centre stuff All the CRM stuff Customer segmentation Operational Risk and Threat Developing sales baselines Shrinkage Retail location Click fraud Retail outlet segmentation Payment fraud Demand/stock forecasting Credit scoring Merchandising Store performance modelling Resource planning

31 The CRISP-DM process 1.Business Understanding 6.Deployment 2.Data Understanding 5.Evaluation 3.Data Preparation 4.Modelling Data Preparation Time

32 Data at the heart of Predictive Analytics High-value, dynamic - source of competitive differentiation Interaction data - / chat transcripts How? - Call center notes - Web Click-streams - In person dialogues Attitudinal data - Opinions Why? - Preferences - Needs & Desires Descriptive data - Attributes - Characteristics Who? - Self-declared info - (Geo)demographics Traditional Behavioral data - Orders -What? Transactions - Payment history - Usage history

33 Definition I IBM/SPSS Statistics IBM/SPSS Modeler Inferential Statistics Inferring parameter values in a target population based on sample statistics. Often using parametric assumptions. Data Mining Applying historical patterns to predict future outcomes. Tested empirically

34 Definition II Data mining Discovering previously undetected patterns and relationships in data Predictive analytics Applying historical patterns to predict future outcomes

35 Types of Predictive Modelling Predict a particular class of outcome e.g. churn Classificati on Value predictio n Predict a value e.g. revenue, visits to a web site Forecast a future value over a defined time period Forecasti ng values Associati on / Sequence Identify repeatable patterns

36 A classification of predictive types Type Overview Classification Predicting Values Forecasting Values Probably the most widely applied. We are predicting a discrete set of targets ( categorical ). Typical examples of classes we predict are: Good credit risk/bad credit risk (loss)/bad risk (profit) Will buy/will not buy Legitimate transaction/fraudulent transaction When predicting classifications we create a propensity score for each class Here the target is a real measurement ( continuous ). It can often take a wide range of values. For example sales my vary from 100 to depending on the day Typical examples are: Revenue Call centre calls Demand in all its forms Predicting values where we explicitly account for previous values over time and look to predict at future points in time e.g. for the next 7 days. The example targets are largely the same as those for Predicting Values Association/Sequence The most specific and therefore less common of the 4 Looking for associations between entities. Most often used for Market Basket Analysis/product affinities. Sequences can look for event sequences that lead to successful outcomes or failures e.g. across repair processes or patient treatment paths

37 Where is Segmentation? Data mining Predictive analytics Discovering previously undetected patterns and relationships in data Here we find segments without an outcome in mind Segments used more strategically Applying historical patterns to predict future outcomes Here we find segments with an outcome in mind (e.g. repurchase propensity or high value propensity) Segments used more tactically

38

39 Creating meaningful segments Descriptive Gender, age etc Lifestyle Behavioural Browsing Purchasing Responding Converting Interaction Attitudinal Brand empathy Satisfaction

40 Segmentation strategies Deterministic Discovery based Rules Hierarchies Filters Clusters Associations Patterns Correlations

41 Behaviourally customer segments Behavioural segmentation based on a core set of attributes Segments profiled using other behavioural data and also additional demographic, survey and/or customer data

42 Homemakers purchase behaviour A mid sized segment with a profile bias towards younger females from low income families Product purchasing: Domestic items Furnishings Nursery Temperature control Purchasing profile suggests this segment are homemakers, probably buying appliances, furnishing and nursery items on a budget low average item value Within all repeat shoppers they are fairly average in terms of volume and frequency of ordering Entry point is through ranges such as floor care and nursery More likely than average to be acquired in Q1 and Q3 Order patterns Value Index Number of orders Number of items Lifetime spend Average order value Average items/order Segment size All shoppers Repeat shoppers % shoppers 2% 9% % orders 5% 9% % items 5% 8% % sales 5% 7% Mixed 20% CC only 58% HD only 22%

43 Homemakers - demographics Index vs Our Brand online shoppers Symbols of Success Happy Families Suburban Comfort Ties of Community Urban Intelligence Welfare Borderline Municipal Dependency Blue Collar Enterprise Twilight Subsistence Grey Perspectives Rural Isolation

44 Homemakers browsing behaviour Category browsing Browsing behaviour Value Index Avg Number of visits Visit lifecycle (days) Categories browsed Index vs repeat shoppers % B2B 1.2% % opted-in 18% Relative to other repeat shoppers this segment focus on browsing their core categories (Baby, Homewares and Household). May also be browsing Furniture as part of their repertoire.

45 Back to prediction - Predictive Analytics for CRM Attract Grow Retain Revenue Profit Less Loss Loss Time Think of events in the journey (visitor, customer, subscriber, applicant, product, machine, etc.) Events are the target for predictions.

46 Types of Predictive Modelling Predict a particular class of outcome e.g. churn Classificati on Value predictio n Predict a value e.g. revenue, visits to a web site Forecast a future value over a defined time period Forecasti ng values Associati on / Sequence Identify repeatable patterns

47 The Business Analytics Continuum What is happening? Why? Business Intelligence What will happen next? What should I do about it? Predictive Analytics Sometimes the data mining approaches give us BI style analyses that BI doesn t

48 Thinking multi-channel means taking a broader view Attract Grow Retain Revenue Profit Less Loss Loss Time The journey is more complex e.g. Research may happen on-line acquisition conversion offline. If we weren t at Big Data already we probably are now.

49 Predicting Life Time Value (Classification) A retailer wants to identify - early in the relationship customers who will go on to become high value customers e.g. here!

50 Brammer Leading retailer/distributor reduces cost of carrying surplus stock and improves customer service Applications & Benefits IBM/SPSS predictive analytics helped Brammer to manage its inventory more efficiently, significantly reducing the need to carry surplus stock, resulting in a total inventory reduction of 31.1 million in one year Inventory turnover improved from 3.2 times at the end of 2008, to 3.7 times at the end of the first half of 2009 Greater understanding of patterns and trends in customer purchasing data helps Brammer forecast marginal stock products more accurately and improve customer satisfaction by making a wider product range available for immediate dispatch Detailed insight into inventory requirements has helped Brammer develop closer relationships with strategic suppliers leading to further cost benefits

51 Suggested further reading

52 Contact us: +44 (0) Thank you

53 Jarlath Quinn Highly Effective Technology For Predictive Analytics in Retail

54 CRISP-DM: Cross-Industry Standard Process for Data Mining

55 What do we offer Anna? 31 years old Estimated income > 28K On average spends 26 Usually pays with credit card Not eligible for discount offer In the last 6 weeks bought these items - What is the next most relevant product to offer her?

56 Why IBM SPSS Modeler? Most intuitive interface on the market Supports the entire process Data agnostic Automatic Model Generation Very powerful data manipulation/transformation functionality Direct scoring to multiple data formats Highly scalable Exploits existing database functionality SQL Pushback In-Database Mining

57 Let s look at an example

58 To finish Options to get started!

59 Common Misunderstandings Revolutionary results overnight! You ll need a Ph.D. In fact, data literate, business focussed people learn how to do this all the time. The more accurate the model the better You need a clean, single-customer-view warehouse

60 Advice to get started Build Internal Credibility: Think about where you would get biggest impact for the least effort. Consider adopting a proven methodology e.g. CRISP-DM ( Consider the full data landscape Consider the sorts of roles involved /impacted Consider integration with other business insight systems (e.g. MI/BI) How will you know its worked? Focus on measuring the benefit e.g. response rate lift, increased cross-sell, revenue/profit impact Don t get hung up on modelling techniques - focus on Business Understanding and Deployment

61 Working with Smart Vision Europe Ltd As a premier partner we sell the IBM SPSS suite of software to you directly We re agile, responsive and generally easier to deal with As experts in SPSS / Analytics / Predictive Analytics we will deliver classroom training courses offer side by side training support offer skills transfer consulting run booster and refresher sessions to get more from your SPSS licences Give no strings attached advice We are a support providing partner so if you already have SPSS you can source your technical support directly from us (identical costs to IBM) We offer telephone support with real people as well as web tickets / queries We offer how to support to help you get moving on your project quickly

62 Options to get started Predictive Analytics for Retail (Starter Pack) Single client software tools, set up, training & first project deployed 10K to 20K Predictive Analytics for Retail (Enterprise Automation Pack) 2 x client plus server software tools, automation plus set up, training & first projects 75K Predictive Analytics for Retail (Enterprise Collaboration & Deployment Pack) 4 x client plus server software tools, automation, collaboration & deployment, set up, training & first project 140K+

63 Contact us: +44 (0) Follow us on Linked In Sign up for our Newsletter Thank you