Quant5 planning ness-050613_final
9 likes6,644 views
Doug Levin of Quant5 presented on how to apply predictive analytics to marketing challenges. He outlined challenges facing marketing departments like making decisions without insights and a lack of data scientists. Levin discussed gathering different types of customer and transactional data to determine key metrics like product relationships, customer scores, and targeted promotional offers. He emphasized using machine learning algorithms and predictive modeling rather than spreadsheets to glean non-obvious insights from data. Levin demonstrated these concepts using a sample ecommerce company called Lucy Couture.
Quant5 planning ness-050613_final
- 1. © 2013 Quant5, Inc. How to apply Predictive Analytics to Marketing Challenges For the Planning-ness Conference May 10th, 2013 Doug Levin | doug@quant5.com
- 2. © 2013 Quant5, Inc. Agenda Planning-ness Approach ~Time Allocation (Minutes) Teaching 45 Putting teaching into practice 45 – 60 Evaluation and discussion 20-30 2
- 3. © 2013 Quant5, Inc. Teaching 3
- 4. © 2013 Quant5, Inc. Challenges Facing Marketing • Frequently make critical decisions without: – The information they need – The insights in their business & environment they need – Access to data in other parts of their organization – A supporting cast & crew (aka data scientists) • Hours are spent each week searching for data – “Connecting silos”
- 5. © 2013 Quant5, Inc. Challenges Facing Marketing Departments • Roadblocks to Success: – Being asked to come up with brilliant new insights • Shortage of data scientists to do statistics, math, etc. • No tools – An avalanche of data from mobile, social media and other sources… and growing by the minute! – Data located in legacy systems run by IT • Have to connecting silos through organizational means not direct reporting authority 5
- 6. © 2013 Quant5, Inc. How ambitious are you? • Do you want to have a “data centric” business? Business decisions no longer based on gut instinct • Do you want to have a “data centric” marketing dept.? Fact-driven relying on measurement & feedback Real-time data At the point of impact Everyone’s Involved & Connected Ubiquitous optimization Automated Relying on predictive analytics & validation Characteristics 6
- 7. © 2013 Quant5, Inc. Data in your organization that can help you… discover trends and opportunities • The top 5 sources of data tagged for predictive analytics: • 54% Sales • 67% Marketing • 69% Customer • 55% Product • 51% Financial In addition, 40% of companies surveyed indicated that “Social (Facebook, Twitter & LinkedIn) had potential value in predictive analytics Source: SAP Analytics 02/08/13 All related to revenue 7
- 8. © 2013 Quant5, Inc. Persistent, Deep Questions Who are our • core customers? • frequent buyers? • best customers? • poorest payers? • Best sales guys? The Who? How do we: • Attract the best customers to buy more? • Reduce the cost of customer acquisition? • increase first purchase size? • increase subsequent purchase size? • increase cross-product purchases? • Reduce fraud The How? 8
- 9. © 2013 Quant5, Inc. Predictive Analytics CAN help? A LOT! • Predict market trends • Predict customer needs • Predict price volatility • Create customized offers for each segment and channel • Predict changes in demand and supply across the entire supply chain 9
- 10. © 2013 Quant5, Inc. Use Predictive Analytics… When your spreadsheet runs out of gas Data Variables Logic Speed 10
- 11. © 2013 Quant5, Inc. Predictive Analytics Solutions Horizontal Vertical Embedded Database Analytics •Hadoop •Unstructured and Structured Databases Consulting Services 11
- 12. © 2013 Quant5, Inc.© 2013 Quant5, Inc. Functional / Business Unit Outcomes Goal: More new & incremental sales Sales Goal: ROI + efficiencies + incremental rev’s Marketing Goal: Better products, prices & competitiveness Product 12 • Customer Analytics • Prospect analytics • Sales cycle analytics • Price analytics • Competitive Prospects & Intelligence • Industry trends
- 13. © 2013 Quant5, Inc.© 2013 Quant5, Inc. Functional / Business Unit Outcomes Goal: More new & incremental sales Sales Goal: ROI + efficiencies + incremental rev’s Marketing Goal: Better products, prices & competitiveness Product 13 • Market trends & Drivers • Competitors, threats & vulnerabilities • Opportunities & Budget Optimization • Improving positioning & messaging • New products, markets and partners • Marketing activity optimizations • Business risks • Threat detection
- 14. © 2013 Quant5, Inc.© 2013 Quant5, Inc. Functional / Business Unit Outcomes Goal: More new & incremental sales Sales Goal: ROI + efficiencies + incremental rev’s Marketing Goal: Better products, prices & competitiveness Product 14 • Product Management Analytics • Actionable Product Intelligence • Competitive Analysis • Partner Analysis • Supply Chain Analysis • Launch Plans & Positioning • Price Analytics
- 15. © 2013 Quant5, Inc. Steps to Successful Predictive Analytics Design Implement Measure • Goal setting • Resource Assessment • Questions to be answered • Tests • Deployment(s) • Feedback • Assessment of KPIs • Improvements • Validation 15
- 16. © 2013 Quant5, Inc. Non-Obvious Knowledge and Probabilities Predictive Analytics for Business Analyze current and historical data in order to better understand customers, products and partners, and identify potential risks and opportunities 16
- 17. © 2013 Quant5, Inc. Putting the teaching into Practice 17
- 18. © 2013 Quant5, Inc. Situation Analysis • Lucy Couture: – A 3-year old eRetailer • High-end “Juicy” couture – Bags, business attire, dresses, intimate apparel, parts, shirts, shoes, skirts – Demographic: • Women (25+) • In College (19-25) • Other (gift purchasers) – Generates a couple of millions in gross revenues p/year – Has 15 employees – Limited data centricity Marketer What are the prices sensitivities? Product Manager What are the product relationships? Marketer What are the demographics (age cohorts) of purchases? Marketer What sort of financial data can be used? 18
- 19. © 2013 Quant5, Inc. Situation Analysis • You are the Director of Marketing – With a marketing manager (“marketer”) with an MBA reporting to you • He/she is not a data scientist • The Marketing Department – Maintains the website – Uses ConstantContact as a email campaign management system – Has access to: • POS data & the customer database – Has a Limited budget – Has limited data centricity but a desire to transform the company into a data centric culture 19
- 20. © 2013 Quant5, Inc. Your goals: 1. Increase revenue 2. Increase efficiencies of marketing activities 3. Improve customer communications 4. Evolve into a data centric organization • Here are the steps involved: 1. Gather data from current systems 2. Determine the product relationships 3. Determine the customer set that is most & least receptive 4. Determine the next product and price to be promoted via email 5. Integrate back into current systems 6. Measure & improve results 20
- 21. © 2013 Quant5, Inc. Your Marketing Mix 4P’s • Price • Product • Promotion • Place Which element(s) of the marketing mix is most effective increasing revenue? 21
- 22. © 2013 Quant5, Inc. Your Marketing Mix 4P’s • Price • Product • Promotion • Place Which type of promotion is going to be most effective? Direct or indirect sales Advertising Marketing Promotions Events Direct marketing PR Email ! Lucy has email addresses from all kinds of potential customers and a relatively small number of actual customers 22
- 23. © 2013 Quant5, Inc. Which data to use? • Do not use data from: – Social Media • Facebook, Twitter, Blogs, surveys, etc. • Customer Sentiment – Mobile Data – Machine Data • RFID, sensors, etc. – Images • Video, audio, emails – “Real Time” • Use data from: – Customer transactions – Legacy systems – Web site: Google Analytics 23
- 24. © 2013 Quant5, Inc. Can A Spreadsheet Do The Analytics? 24
- 25. © 2013 Quant5, Inc. Prospect Customer Scores – Step Two • Data needed – ( Any demographic data is ok here, we can take advantage of a lot of disparate types of information) – Customer ID – Age – Household income Equation: Machine learning algorithm which mines customer demographic and descriptive data to determine which characteristics are indicators of success. 25
- 26. © 2013 Quant5, Inc. Product Relationships – Step Three • Data needed – Product transactions data – Transactions in common Equation: Determine which groups of products are highly correlated and purchased together. 26
- 27. © 2013 Quant5, Inc. Targeted Offers – Step Four Available Data: • 10,880 email addresses • 1,360 customers – ∑ 4,080 transactions (3 per customer) – x transaction = $138.00 (2013) • Days since last purchase • Purchase prices by product, category and SKU number 27
- 28. © 2013 Quant5, Inc. Targeted Offers– Step Four • Equation: Determines the similarity of market baskets by analyzing past customer behavior, and determining which products are most likely to be purchased next by each customer. 28
- 29. © 2013 Quant5, Inc. Targeted Offers – Step Four • Promotional offer: – Who should receive these targeted offers? A select group of established customers – How should the customer info be presented? Customer names and customer IDs – What other information would be helpful to know? Lifetime value Risk of Churn • Validation – Past KPIs (# of emails per period, opens, sales) – Closed loop? 29
- 30. © 2013 Quant5, Inc. Demo 30
- 31. © 2013 Quant5, Inc. Discussion & Assessment 31
- 32. © 2013 Quant5, Inc. How YOU can apply Predictive Analytics to Marketing Challenges For the Planning-ness Conference May 10th, 2013 Doug Levin doug@quant5.com