Building Economical Simulators
Download as PPTX, PDF1 like260 views
The document discusses building economical simulators by using historical data from news articles, social media, and web searches to simulate causal relationships between events. It provides examples of simulating how an oil spill off Singapore in 2010 could cause gas prices to increase using data about the tanker collision and spill that occurred on May 25, 2010. The goal is to connect macro and micro economic factors to enable simulating the market.
Building Economical Simulators
- 1. Building Economical Simulators Dr. Kira Radinsky, CTO & Co-founder
- 2. 2 Customer Lifecycle Intelligence Predictive Marketing Predictive Customer Success Predictive Sales
- 3. 3 Uriel’s Journey IT Senior Director IT ? IT Are demographics alone enough for predictions?
- 4. 4 Customer Lifecycle Intelligence Articles read Social Posts
- 7. access NAC Network Access Control Access Control Solutions Physical Security Unmanaged Servers
- 8. What is Dell Reading About?
- 9. Month Identity Management Is It the Right Time to Contact Dell?
- 10. Trend Analysis: Is it the Right Time?
- 13. What is Uriel Writing About?
- 14. 14 You Get a Summary Screen IT Senior Director IT Lead Twitter Topic Fit IT Low Influencer Company Interest Trend High Integrated in Salesforce and Marketo
- 15. 15 Customer Lifecycle Intelligence Articles read Social Posts Website Interactions Webinar Attendance
- 16. Uriel’s Website Visitation Patterns Webinar Email Response Contact Me Call 2nd Interaction 87% 10% 69% 34%
- 17. 17 Dynamic Predictive Lead Scoring IT Senior Director IT IT
- 18. 18 Customer Lifecycle Intelligence Articles read Social Posts Website Interactions Webinar Attendance Campaign ResponseEmail Response Time
- 19. 19 Historical Response Analysis Response Time Fast Response: 70% Slow Response: 30%
- 20. 20 How Fast Does Uriel Respond? Fast Response: 70% Slow Response: 30% Response Time
- 21. 21 Lead Score Changes Automatically IT Senior Director IT IT Slow ResponseResponse Rate Responds to Emails?
- 22. 22 How Fast Does Uriel Respond? Response Time Mid Response: 60%
- 23. 23 Several Factors Make Him an “A” Prospect IT Senior Director IT IT Medium ResponseResponse Rate Responds to Emails?
- 24. 24 Customer Lifecycle Intelligence Articles read Social Posts Website Interactions Webinar Attendance Campaign ResponseEmail Response Time Support Cases Usage Patterns
- 25. Cases with High Severity Usage of basic authentication Usage of advanced NAC feature Uriel’s Engagement and Usage Level?
- 26. 26 Predictive Customer Success: Renewals FAQ Campaign How To Ramp-Up Campaign
- 27. Cases with High Severity Usage of basic authentication Usage of advanced NAC feature Uriel’s Engagement and Usage Level?
- 28. 28 Predictive Customer Success: Renewals Good candidate for Upsell campaign Upsell Campaign – did you see our new features?
- 29. Success Model Opportunity Score + Insights Wins Losses Internal Data Sources External Data Sources Demographics Firmographics Behavioral 30 On-the-spot Predictive Scoring + Insights Update 1x/week 1x/min
- 30. 31 Insights from Historical Data Analysis
- 31. 32 Installed in <1hour Qualified Pipeline Renewal Rate 1.75x Conversion Rate 2x +10% Easy Install, Fast Time to Value
- 32. 150 years of news articles Billions of tweets Millions of web searches
- 33. Event1 Merchant shipCollide Tanker 05/25/2010 11:00 SingaporeLocation Action Time- frame Event2 spill 05/25/2010 15:00 Action Time- frame Time “Tanker and merchant ship collide” “Oil spill off Singapore” Oil Location Singapore
- 34. Event1 Go up Gas prices 05/25/2010 11:00 Action Time- frame Event2 spill 05/25/2010 15:00 Action Time- frame Time “Oil spill off Singapore” Oil Location Singapore “Gas prices go up”
- 37. Simulating the Market Connecting Macro & Micro Economics
- 38. Thank You!
Editor's Notes
- #2: Mainstream economic theory models the world in a single almost static model. In this talk, I will present how to mine the ever changing web, flows of information in company’s internal data to build dynamic micro and macro economic models. I will share algorithms that mine millions of data points and news to build simulators from text and other unstructured data. I will share some of our customer results of using those models to boost their revenue by 10-30%.
- #3: Many marketers use very limited data to target prospects and segment their lists but the buying cycle is changing and there is a wealth of data and powerful tools out there that you Mr./Ms. Marketers can use to be more effective (uncover hidden "gold" in your Marketo & CRM, deliver more SQLs, increase conversions, make Sales team happy). Then go into Uriel example to illustrate the point.
- #4: We are a security company trying to sell to Dell
- #15: We did all the work for you behind the scene! You just get this summary page!
- #30: Additional Data Points: Intent Product installs: online and offline
- #31: Additional Data Points: Intent Product installs: online and offline
- #32: Explain that this is the conversion and how many deals! Find hidden sweet spots
- #34: So I started collecting more and more data – we got all the NYT data from 1850 till today, social and real-time media, and human web behavior.
- #35: I was trying to teach the computer to find and understand human text and learn causality patterns from it. It was looking for patterns like: “Colorado Flooding Imperils Oil and Gas Sites, Causes Spill” “Texas oil pipeline fire causes evacuation of town near Dallas” “oil causes war” I believed if we could model the cause and effect of every possible event – we can start predicting better. So for each event reported we started automatically finding who performed the action, who the action was performed on and using what instrument and we started building a model of those causalities.
- #36: I was trying to teach the computer to find and understand human text and learn causality patterns from it. It was looking for patterns like: “Colorado Flooding Imperils Oil and Gas Sites, Causes Spill” “Texas oil pipeline fire causes evacuation of town near Dallas” “oil causes war” I believed if we could model the cause and effect of every possible event – we can start predicting better. So for each event reported we started automatically finding who performed the action, who the action was performed on and using what instrument and we started building a model of those causalities.
- #37: Causality Graph Building Built on 20 machines 300 million nodes 1 billion edges 13 million news articles in total
- #39: Let’s take this one step further! Predict the behavior of a complex market. Learning across industries What to do when you get to a new industry? Collaboration between companies/ channels/ distributors Two distributors are turning to the same customers – prevent this (How do we suggest the same lead to several companies)