Personalization, Beyond Recommenders by Edward Chenard
1 like294 views
This document provides an overview of Edward Chenard's career in personalization and recommenders over 20 years. It discusses his work leading data science teams at Fortune 500 companies and building over 50 data-driven products. The document then outlines Chenard's quest to understand the missing pieces in personalization, as algorithms and more data did not always solve issues. This led him on a global journey and to consider factors like human behavior, context, design, and behavioral economics. The document emphasizes that personalization must solve the contradiction between standardization and individuality.
Personalization, Beyond Recommenders by Edward Chenard
- 2. About Me Data, Digital and Personalization Transformation A Few Highlights • Worked with personalization and recommenders for 20 years • Led data science and engineering at numerous Fortune 500’s • Built and led the creation of over 50 data driven and digital products • Grew up in Bermuda, now lives in Minnesota
- 3. Where the Story Begins It’s 2011
- 4. Vision of Data – Continual Customer Relevant Interactions Using data to learn about our customers. CONTEXT AGGREGATE GENERATE INSIGHTS SERVE MEASURE CAPTURE • Services that capture and combine real- time signals from the physical world with location data, online activities, social media, and many other types of contextual input. • Technology that enables rapid aggregation of data from multiple sources and delivers insights that can give users an immersive and valuable experience. • The power of context will enrich not only online experiences but real-world ones as well. • The emergence of context-based services will have a compounding effect— those new services themselves provide additional sources of context.
- 5. Metadata How Data Flows Context Capturing Pipeline Context Mapping Customer Actions Web-enabled Access Point Resource Online Context Capturing Pipeline Real-time Repository MapReduce Processing Real-time Processing Customer Actions Internal Context Capturing Pipeline Internal Access Point Data Algorithms Data Aggregation Insights Repository Insights Generation Pipeline Servicing API Servicing Views Experience Experience Feedback Business Rules Scheduling/Queue Mgmt Customer Reactions Archive Repository
- 6. Recommended Product Algorithm Calculation - Matrix Decomposition 1 0 0 1 0 0 0 … 0 0 1 0 0 0 1 … 1 0 0 0 0 0 1 … 0 0 1 0 1 0 0 … 0 0 0 0 0 1 0 … 0 0 0 1 0 1 0 … Input Matrix User X Item (~ 6 mill X ~20k) Tuples: (user, item, count) Filtering: • viral items • users with too few items • Items with too few users Row Index User 1 a1745690-96fb-11e2-9e96- 0800200c9a66 2 c86312f1-96fb-00e2-9e96- 0800200c9a66 3 e41a1b11-96fb-11e2-9e96- 06c03c85fb6e Col Index Item 1 1945531 2 2953816 3 8264042 Clickstream (web, e-mail) Purchases Social/Search/etc DataSources
- 7. Event Driven Calculations Logical Application Flow – ClickStream Data StoresIngest Batch Aggregations Also Viewed Most Popular Raw Data (S3) APIs 1 2 4 3 Event and Item Key Spaces C* Summarizes
- 8. Innovation Loves Adversity Building an unproven future. We had to Start Somewhere • Big Data (Hadoop) • Data Science • And lots of data What did Personalization do for Best Buy • Processed more data than the rest of the company combined • Helped Best Buy be a top 5 performing stock of 2013 • Changed how Best Buy engages and understands customers Headlines from 2014 - Present
- 9. SomethingWas Missing Unknown factors were in play We Started to Notice Cracks in the System • Algorithms would go flat • More data didn’t solve the issue • More complex algorithms didn’t solve the issue Personalization Big Data Data Science The Problem was Bigger than Best Buy • Other teams at other companies had the same issues • Our happy model of success was temporary • Other factors were influencing success, but we didn’t know what
- 10. “The STAV Cycle” Strategy / review Technology implementation Analytics Data visualization “gaining insight and telling stories with data” Developed in 2014 with David Quimby to help talk about data and business problems
- 12. A Quest for Information From 2012 – 2014, I sought out experts in tech and personalization Information was Plenty, True Answers Were Not • Data Science was the answer to everything according to most • All tried the same methods but few could produce a real answer for improvement • Thousands spent on conference, travel to speak with experts, and every book on personalization lacked of real answers
- 13. Aspects of Personalization as We Understood It Targeted offers Intelligent Customer segments Business Rules Contextual Ads Ad Server Relevant Guided Selling Recommendations Personas Targeted Messaging Targeted Content User Generated Content Cross-Sell Intent-Based search Up-Sell and Cross-sell User Profile Counter Offer Need Based Shopping Social Marketing Behavioral Targeting Social and interests Profile Location/Proximity #kpnicholas 2011-2016
- 14. Transformation Strategy Focus on Productivity Focus on Customers Enhancement Focus on a Platform HowtoPlay HowtoWin Capabilities and Operating Model Innovation Business Model Talent and Culture Partner Ecosystem Model Data and Connected Infrastructure Change the Game Harness the Platform Go Together, Go Far Building Data, Insights, Action into our DNA All About Outcomes
- 15. Framing Personalization Personalization Strategy: Customer and Business Value Analytics: Who, How, Where? IT: What Tools and Why Social Sciences and Design: Understanding the Human Condition - Measurable Results - Speed to market - What is important for the customer - ROI - Spark, Hadoop - Cassandra, The Cloud - Pig, Hive, Python - HDFS - Understanding the Customer - Social trends and how to use them - UX - Design Theory - Data Scientist vs. Statistician - Languages - Retain, Train - Recommenders - Social vs Technical Developed around 2014
- 16. Personalization Expands Unknown factors were still in play An Improved Understanding of Data • Human behavior doesn’t run on objective behavior, it runs on context and meaning • To change perceptions, you have to change the story • Independent and creative thinking is what always drove progress and civilization, yet data science is essentially a copy and paste field which is why it is has and will never live up to its promise until that changes Personalization Big Data Data Science Design Data Governance An Improvement But Not a Solution • Design helped to give depth of experience to customers • Governance helped us manage our data better • However, we still lacked a true understanding of customer intent
- 17. Personalization is a Lot LikeWorld Building Great stories draw the audience into their fictional world. Great personalization, draws the user into a new world of their own creation. Expanding my Quest Globally • 2014-2017, a global journey to learn how other cultures and countries approach personalization • Data is a second order simulacrum of the true self • Physical or digital, identity and spaces are constructed not by ourselves alone but by the network. • Perhaps we had some things all wrong
- 18. Activity CenteredThinking Understanding Self Perception vs Reality: People act on perception, not fact. Oversimplify and you Breed in Weakness • Our identity is tied up in the things we own and we seek to create our perception of identity with those things we own • Profundity – great depth of knowledge or thought. If you really want to attach with customers, profundity is a critical aspect. All marketers are liars, because they must lie to themselves. You want your product to remain in thought after the experience so that they become a part of the customer’s Actions and Touchpoints Social (online) Offline Work Offline Personal Online Persona Devices Phone
- 19. AwarenessWithout Understanding Lots of Ideas, how do they fit? Many Moving Parts Without a Theme • Personalization plays on both our desires to improve ourselves but without losing what makes us, us. Individuality and identity are a huge part of the context and conversation • The journey is a story because we as a species love stories and the arches the characters take. Often our personas are a character we adopt for a situation and those arches we wish to play are part of the storyline journey we want to create • What does an experience/story need in order to be relatable to its audience? Big Data Marketing Personas Behavioral Economics Design Theory Behavioral Science Data Science Abstract Concepts
- 21. Design and Systematic Innovation Personalization Solves a Contradiction Design Showed the Problem • Every truly disruptive innovation solves a contradiction. Every solved contradiction was once an unsolved contradiction • Organizations/teams usually forge ahead without defining a problem. Without defining the contradiction, you can’t actually solve the problem • Personalization solves a contradiction which associates it with a well-defined problem and has a non-linear solution (a radical innovation) Standardization (mass marketing/mass production) Personalization/ Mass customization Null configuration Customization (craft/artisan) High Scalability Low Low HighIndividuality
- 22. The Genius of Sergio Leone The Good, the Bad and the Ugly Sergio Leone’s Method • Most movies had the script written, then shot, they scored • Leone’s method was the score the movie as it was written and then shot • Leone played the music on set while filming, creating a deep emotional experience shown in the actor’s expressions and better timing of sequences in a shot than was achieved without the music • How something is presented is more important than other factors like price
- 23. How PersonalizationWorks Sublation – combination without a loss. Evoke In Motion • Evoke in motion – Emotions and motion go together. Helping a customer evoke emotion and motion, helps to engage them in their story arch. Making your customer feel and act upon that feeling to create a perception • Multiplicity - Undermining a single master narrative with a network of intertext that make meaning of choice. Personalization disobeys because most shopping experiences are devoid of choice and have a beginning and end. Personalization doesn’t have an middle or end, it just evolves Personalization Philosophy (Concepts) Behavioral Sci/Econ Design Data Engineering Data Science Social Science and Marketing Data Governance
- 24. Heart Graphic by Accenture Philosophy of Personalization ATEMPORALITY IDENTITY NETWORK CULTURE HETEROTOPIAS An unmooring from historical methods. We are currently experiencing an atemporal period of consumer preferences. Spaces like identity are constructed. Our abilities are limited by the space in which it is produced. Our identity is tied up in the things we own and we seek to create our perception of identity with those things we own. Extremely fluid, often poorly organized but seemingly persuasive even when wrong. The network is us as a group, understand the network to understand the person. A dynamic space of layers and meaning. Margin spaces to explore non-standard methods. We don’t have a dominant culture, we have many, personalization creates the space for the identity. The True Drivers of the Personalized World
- 25. BehavioralAspects It’s always better to reduce anxiety and ambiguity than improve speed. Human behavior is essentially social. In all forms of society, there are three constant characteristics: • We always live in groups • Groups have hierarchy with a defined structure of power • People need to have a system of meaning that helps them make sense of the world These characteristics help people to get along, get ahead and create meaning. Finding meaning, is the key to understanding why people make decisions. People have a need for meaning and structure in their life. How people achieve these goals are determined by their personality. Offline Self Device Self Online Self Clash between Today and Future Aspirational You Present You 1-1
- 26. Behavioral Economics JP Morgan: Everyone has two reasons for doing something, the real reason and the good reason. Strategic - “Transform the Business” Focused on the ‘top-line’ value drivers associated with growth, innovation and scale. Tactical – “Run the Business” Focused on the execution-based savings that can be realized based on optimizing the data model and data maintenance processes. Operational - ”Grow the Business” Focused on savings driven based on improving data quality, deploying a sustainable governance model and optimizing our data (and related) processes. KPI’s Cost of Business Time to Complete Number of Systems / Interfaces Attributes & Hierarchy (and related) Data Quality Customer Journey By Numbers Single Version of Truth Supply Chain Optimization and Planning Behavior Change or Encouragment Estableshment of Meaning Customer Profitablitiy Analysis Creation of a new Customer Journey Value Drivers
- 27. Design Strategy Determines which stories need to be communicated to whom and how to use the technologies and structures before us to do so. http://en.wikipedia.org/wiki/File:PazuzuDemonAssyria1stMilleniumBCE.jpg #kpnicholas
- 28. Desktop Mobile Tablet In Store Publications TV/Radio Product Packaging Customer Support Ecommerce Customer Profile Product and Service Content Social Media Opportunities Advertising and media Sales’ Support and Tools Apps Personalization Customer Profile Email Execution of Design and Behavior #kpnicholas
- 29. Wisdom Collective application of knowledge into action Knowledge Experience, values, context applied to a message Information A message meant to change receiver’s perception Data Discrete, objective facts about an event Experience Grounded Truth Complexity Judgement Heuristics Values & Beliefs Quantitative Contextual Evaluative Qualitative Intuitive Informative Quantitative Connectivity Transactions Informative Usefulness Quantitative Cost, Speed Capacity Timeliness Relevance, Clarity Adding Value: Action-oriented Measurable efficiency Wiser decisions Adding Value: Contextualized Categorized Calculated Corrected Condensed Adding Value: Comparison Consequence Connections Conversations Transitioning to emerging technologies + Human/Machine = Transformation Data Lakes without Substance lack any meaning. Data Engineering Plays a Better Role
- 30. Data Engineering Collection of data streams. The more ways you can look at something, the richer you are. Cultural Behavioral Product Clickstream Local Events • Probability of engagement /LTV • Probability of Growth/Churn • Network Patterns (Human) • Life Changing Events • Server Response Time • Data Reliability • Network Patterns (Tech) Enriched data for downstream teams
- 31. Data Science as aTeam Player Signals have attributes depending on their representation in time or frequency domain can also be categorized into multiple classes All signal types have certain qualities that describe how quickly signals can be generated (frequency), how often the signals vary (rate of change), whether they are forward looking (quality), and how responsive they are to stimulus (sensitivity) Rate of Change (Slow or Fast) Quality (Predictive or Descriptive) Sensitivity (Sensitive or Insensitive) Frequency (High or Low) Sentiment Expressed as positive, neutral, or negative, the prevailing attitude towards and entity Behavior These signals identify persistent trends or patterns in behavior over time Event/Alert A discrete signal generated when certain threshold conditions are met Clusters Signals based on an entity’s cohort characteristics Correlation Measures the correlation of entities against their prescribed attributes over time Humans don’t act logically.
- 32. The Soft Side of Data Science A Maturity Model: Four Phases of Data-Driven Culture non- quantitative (“intuitive”) quantitative / static (“statistics is not machine learning”) quantitative / dynamic (a culture of machine learning / experimental design) quantitative / dynamic with human intelligence (a culture of machine learning / experimental design) Human behavior doesn’t run on objective behavior, it runs on context and meaning
- 33. Design of Experiments (DoE) Social Science and Marketing Ideas Build Product Measure Data Learn Minimize total time through the loop Experimental design of new data products requires a framework that allows us to test ideas quickly and allows us to get to answers for production products faster. DoE and Lean innovation can help us do that. Working with production product teams, business owners and UX to frame up experiments will help Big data and Data science teams get to results faster, using this model. Innovation at the edges, build for the middle and you have competitors, for the edge and you own the market.
- 34. Social Science and Marketing We Begin To build A Trusted Relationship Imagine the Potential Realize the Potential Customer Journey Perception Journey Emotional Journey Moment of Truth Signaling Actualizing Personal Growth Personalization Influencers Social Marketing Network Influencers Search and Browse Peer Reviews Email Data Refining Recommenders Positive Psychology Community Building Growth via new experiences Network Influencing Validation Marketing is not about optimizing against our competitors, it is about optimizing the experience in the context you and the customer define. Don’t compete, monopolize so you can removed the competitors from the equation.
- 35. Traditional Data Governance is Not Enough Data Governance Risk Mitigation Data Governance Data Quality Master Data Management Legal and Ethics Team Building Data Strategy and Governance is about building a single point of truth which allows: • We should look to improve the experience of the customer • Innovation takes place around the edges, be comfortable with ambiguity • Over administrative often kills the ability to be successful. “The typical data quality issues, as well as the typical impacts of poor data quality. The increased costs are dramatic: between 8 and 12% in revenue and between 40-60% in expenses.” Thomas C Redman author of “The Impact of Poor Data Quality”
- 36. The Soft Side of Data culture strategy Culture Precedes Strategy strategy technology Strategy Precedes Technology Using mathematical models, often won’t work because how we perceive things is not objective. The way you get people to behave in a certain way, may be oblique
- 37. How PersonalizationWorks The Lessons Learned. Personalization is Unique, because We Are • We are not designed to see the world objectively. Evolution doesn’t care about accuracy, it cares about fitness. We notice contrast more than absolute values. This is why using math and logic, doesn’t always work, because we forget the basics of what nature does • Don’t just test or build the obvious, test and build the unconventional too. Everyone competes to be logical so they end up in the same place which is the most saturated part of the market. Real winners don’t win by being logical like everyone else, they win by being unconventional Personalization Philosophy (Concepts) Behavioral Sci/Econ Design Data Engineering Data Science Social Science and Marketing Data Governance
- 38. Brainswarming When a team runs into a problem they have difficulty solving, all teams come together in a brainswarm model to help come up with a solution. Use the network of the team to solve problems, more points of view makes a team rich and more effective at engaging customers in new experiences.
- 39. Does thisWork? Results matter, so here are the results of this method There is Still a Lot to Do. • Improved conversion rates by 400% in 3 months • Reduced the cost of data management from $8 million to $3 million a year • Increased customer lifetime value by improving repeat customers • Better employee satisfaction, turnover dropped from 15% to 2% Break Even Break Through Break Away
- 40. The Future Is our online self, representative of the whole? Can your online persona create a persona of its own and give you a different reaction online than in real life? There is Still a Lot to Do. • AI needs to learn to be trivial. AI doesn’t practice defensive decision making which can allow it to be creative where sometimes humans are not. But we have to make sure it doesn’t have the biases of the trainer • The meaning of a signal is changing, to what is the question still unanswered • Perception will continue to drive reality