[Webinar] Demystifying Predictive Lead Scoring
13 likes4,118 views
This webinar led by Tony Yang and featuring guest presenter Kerry Cunningham focuses on demystifying predictive lead scoring, covering its evolution and key considerations for implementing a program. It highlights the shortcomings of current lead scoring methods and introduces the potential of big data and data science for deeper qualification of leads. Key takeaways include the importance of understanding past behavior for future predictions and the necessity of utilizing larger, more reliable data sets.
[Webinar] Demystifying Predictive Lead Scoring
- 1. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Demystifying Predictive Lead Scoring Host: Tony Yang Director of Demand Gen Mintigo @tones810 Guest Presenter: Kerry Cunningham Research Director SiriusDecisions
- 2. © 2014 Mintigo. All Rights Reserved. www.mintigo.com HouseKeeping Audio Check Audio is delivered via your computer speakers Please let us know in the chat window if there are audio issues Webinar Replay Available We will send you a recording of today’s session a4erwards
- 3. © 2014 Mintigo. All Rights Reserved. www.mintigo.com HouseKeeping Tweet With Us @mintigo @SiriusDecisions #PredictiveLeadScoring Ask Ques3ons In The Chat Window Ask ques8ons at any8me & we will answer them during Q&A
- 4. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Mintigo Enterprise Predictive Marketing
- 5. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Our Guest Presenter Today Kerry Cunningham Research Director at SiriusDecisions • More than 20 years in B2B • Expertise in inside sales, telemarketing & marketing-sales alignment • BA, Indiana University & MS, San Francisco State University
- 6. Propensity Modeling 6 Kerry Cunningham Demand Creation Services
- 7. © 2014 SiriusDecisions. All Rights Reserved 7 SiriusDecisions, Kerry Cunningham • Research Director, SiriusDecisions • Lead Development & Management • 15+ years in b-to-b demand generation and lead management • VP Operations for b-to-b teleservices organization • Research methods and analytics • 5 years social science research • Organizational behavior • Employee selection science • Propensity modeling people • Behavioral economics • Personality correlates of well-being, Book Chapter, Summer 2014
- 8. © 2014 SiriusDecisions. All Rights Reserved 88 Peering behind the curtain…
- 9. What We’ll Cover Demystifying Predictive… • Where lead scoring has come from and is going • What predictive (anything) really means • 4 key considerations for planning a predictive lead scoring program • 4 factors for making good predictions © 2013 SiriusDecisions. All Rights Reserved 9
- 10. © 2014 SiriusDecisions. All Rights Reserved 10 Predictive Lead Scoring Assumptions
- 11. © 2014 SiriusDecisions. All Rights Reserved 11 Predictive Lead Scoring Assumptions • Not all leads convert
- 12. © 2014 SiriusDecisions. All Rights Reserved 12 Predictive Lead Scoring Assumptions • Lead Scoring is an attempt to predict which will • Not all leads convert
- 13. © 2014 SiriusDecisions. All Rights Reserved 13 Predictive Lead Scoring Assumptions • Lead Scoring is an attempt to predict which will • Somewhere in the world are clues as to which leads are most likely to convert (big data) • Not all leads convert
- 14. © 2014 SiriusDecisions. All Rights Reserved 14 Predictive Lead Scoring Assumptions • Lead Scoring is an attempt to predict which will • Somewhere in the world are clues as to which leads are most likely to convert (big data) • Theoretically, it is possible to know and account for all of those clues • Not all leads convert
- 15. © 2014 SiriusDecisions. All Rights Reserved 15 Predictive Lead Scoring Assumptions • Lead Scoring is an attempt to predict which will • Somewhere in the world are clues as to which leads are most likely to convert (big data) • Theoretically, it is possible to know and account for all of those clues • Practically, it is not possible to account for 100% of the clues • Not all leads convert
- 16. © 2014 SiriusDecisions. All Rights Reserved 16 Predictive Lead Scoring Assumptions • Lead Scoring is an attempt to predict which will • Somewhere in the world are clues as to which leads are most likely to convert (big data) • Theoretically, it is possible to know and account for all of those clues • Practically, it is not possible to account for 100% of the clues • The more we can account for, the better we can predict whether any given lead will convert • Not all leads convert
- 17. © 2014 SiriusDecisions. All Rights Reserved 17 Predictive Lead Scoring Assumptions • Lead Scoring is an attempt to predict which will • Somewhere in the world are clues as to which leads are most likely to convert (big data) • Theoretically, it is possible to know and account for all of those clues • Practically, it is not possible to account for 100% of the clues • The more we can account for, the better we can predict whether any given lead will convert • Current lead scoring probably doesn’t account for as much as we might hope • Not all leads convert
- 18. Traditional Lead Scoring What’s the problem?
- 19. © 2014 SiriusDecisions. All Rights Reserved 19 The Problem With Current Lead Scoring Implicit Explicit Current lead scoring fosters this view of the world…
- 20. © 2014 SiriusDecisions. All Rights Reserved 20 The Problem With Current Lead Scoring Implicit Explicit Behavior - Hiring - Expansion - New products - Social media - Communities Fit - C-level attitudes - Tech Ecosystem - Financial Health - Competition - Positioning When reality looks a lot more like this…
- 21. © 2014 SiriusDecisions. All Rights Reserved 21 What We Are Trying To Do When We Predict To better understand that, just look a little further down the waterfall from where current lead scoring occurs
- 22. © 2014 SiriusDecisions. All Rights Reserved 22 The Best-in-Class B-to-B Scenario For most, there’s a substantial drop-off between TQL/TGL and SQL qualification… Conversion % AQL > TQL 66.6%
- 23. © 2014 SiriusDecisions. All Rights Reserved 23 The Best-in-Class B-to-B Scenario For most, there’s a substantial drop-off between TQL/TGL and SQL qualification… Conversion % AQL > TQL 66.6% TQL > SQL 48.8%
- 24. © 2014 SiriusDecisions. All Rights Reserved 24 The Best-in-Class B-to-B Scenario For most, there’s a substantial drop-off between TQL/TGL and SQL qualification… Conversion % AQL > TQL 66.6% TQL > SQL 48.8% Conversion from AQL to SQL = 32.6%
- 25. © 2014 SiriusDecisions. All Rights Reserved 25 Downstream People and Processes Today, most of that qualification involves teleprospecting and sales calls
- 26. © 2014 SiriusDecisions. All Rights Reserved 26 Downstream People and Processes § Call decision makers § Ask key qualifying questions
- 27. © 2014 SiriusDecisions. All Rights Reserved 27 Downstream People and Processes • Expensive • Slow • Limited to stock on hand • Very high propensity
- 28. © 2014 SiriusDecisions. All Rights Reserved 28 The Future of B-toB Lead Development Find clues that exist out in the world, which reliably point to qualifying criteria you would ask the decision- maker if you could get him/ her on the phone? The Role of Data Science
- 29. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 29 The Promise of Big Data/ Predictive Lead Scoring Cheap/ Fast Shallow Qualification Deep Qualification Slow/ Expensive Before 1. List purchase and selection unsophisticated The promise of predictive technology is to perform deep qualification of accounts and opportunities through data science and automation.
- 30. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 30 The Promise of Big Data/ Predictive Lead Scoring Cheap/ Fast Shallow Qualification Deep Qualification Slow/ Expensive Before 1. List purchase and selection much more sophisticated 2. Technology does more deep qualification With MAP The promise of predictive technology is to perform deep qualification of accounts and opportunities through data science and automation.
- 31. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 31 The Promise of Big Data/ Predictive Lead Scoring Cheap/ Fast Shallow Qualification Deep Qualification Slow/ Expensive Before 3. And sales becomes more efficient 1. List purchase and selection much more sophisticated 2. Technology does more deep qualification With MAPPredictive The promise of predictive technology is to perform deep qualification of accounts and opportunities through data science and automation.
- 32. The Nature of Predictions Correlation and Regression Without Math
- 33. © 2014 SiriusDecisions. All Rights Reserved 33 The Nature of Prediction Propensity to… Likelihood of… Predict-o-meter Inquiries Sales Qualified Leads
- 34. © 2014 SiriusDecisions. All Rights Reserved 34 The Nature of Prediction Propensity to… Likelihood of… Predict-o-meter Inquiries Sales Qualified Automation Qualified (AQLs)
- 35. © 2014 SiriusDecisions. All Rights Reserved 35 The Nature of Prediction Propensity to… Likelihood of… Predict-o-meter Factors Not Accounted For
- 36. © 2014 SiriusDecisions. All Rights Reserved 36 The Nature of Prediction Propensity to… Likelihood of… Predict-o-meter Factors Not Accounted For Other scorable factors
- 37. © 2014 SiriusDecisions. All Rights Reserved 37 Making Predictions: Correlation and Regression Visual regression model: Predicting size from clothes Knowing a person’s sport coat size, can you predict the size of the person?
- 38. © 2014 SiriusDecisions. All Rights Reserved 38 Making Predictions: Correlation and Regression Visual regression model: Predicting size from clothes Does this new data improve your prediction?
- 39. © 2014 SiriusDecisions. All Rights Reserved 39 Making Predictions: Correlation and Regression Jacket and shoe size are correlated… knowing one helps predict the other Visual regression model: Predicting size from clothes
- 40. © 2014 SiriusDecisions. All Rights Reserved 40 Making Predictions: Correlation and Regression Visual regression model: Predicting size from clothes +
- 41. © 2014 SiriusDecisions. All Rights Reserved 41 Improving predictions – Regression Modeling Factor 1 Factor 3 Visual Regression Model Factor 2
- 42. © 2014 SiriusDecisions. All Rights Reserved 42 Improving predictions – Regression Modeling Factor 1 Factor 3 Visual Regression Model Factor 2
- 43. © 2014 SiriusDecisions. All Rights Reserved 43 “Error” in Predictions/ Noise In the Data
- 44. © 2014 SiriusDecisions. All Rights Reserved 44 “Error” in Predictions/ Noise In the Data Suit too smallSuit too big
- 45. © 2014 SiriusDecisions. All Rights Reserved 45 Predictions Buildings Built Employees Visual Regression Model: Predicting Construction Management Deal Size Const. Workers + + = Predict-o-meter Guessing Perfect Prediction
- 46. © 2014 SiriusDecisions. All Rights Reserved 46 Predictions Buildings Built Employees Visual Regression Model: Predicting Construction Management Deal Size Const. Workers + + = Predict-o-meter Guessing Perfect Prediction
- 47. © 2014 SiriusDecisions. All Rights Reserved 47 Predictions = Recent HiresSeries C HR Leader + + Visual Regression Model: Predicting HR Mgt SaaS Sales Predict-o-meter Guessing Perfect Prediction
- 48. © 2014 SiriusDecisions. All Rights Reserved 48 Predictions = Recent Hires Series C HR Leader + + Visual Regression Model: Predicting HR Mgt SaaS Sales Predict-o-meter Guessing Perfect Prediction
- 49. © 2014 SiriusDecisions. All Rights Reserved 49 Predictions ?
- 50. © 2014 SiriusDecisions. All Rights Reserved 50 Predictions ?
- 51. © 2014 SiriusDecisions. All Rights Reserved 51 Predictions In reality, there are often numerous predictors that go into a predictive model + = + + + + Predict-o-meter Guessing Perfect Prediction
- 52. © 2014 SiriusDecisions. All Rights Reserved 52 Predictions In reality, there are often numerous predictors that go into a predictive model = 1.15 * 1.05 * 1.2 * 3.3 * 12.75 * 1.75 * % Lift++ + + + Predict-o-meter Guessing Perfect Prediction
- 53. Predictive Lead Scoring Considerations 53
- 54. © 2014 SiriusDecisions. All Rights Reserved 54 Building A Model Use Case Starting Point Entity Predicted Source of Predictors Model
- 55. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 55 Use Cases Find new businesses that have a high propensity to buy from me Among many use cases for predictive lead scoring, finding new leads, scoring known leads, and scoring existing customers are the top three.
- 56. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 56 Use Cases Find new businesses that have a high propensity to buy from me Score and prioritize businesses already in my database on their propensity to buy from me Among many use cases for predictive lead scoring, finding new leads, scoring known leads, and scoring existing customers are the top three.
- 57. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 57 Use Cases Find new businesses that have a high propensity to buy from me Score and prioritize businesses already in my database on their propensity to buy from me Score and prioritize existing customers for their propensity to buy other products and services we sell Among many use cases for predictive lead scoring, finding new leads, scoring known leads, and scoring existing customers are the top three.
- 58. © 2014 SiriusDecisions. All Rights Reserved 58 Starting Point Historical Data Became Customers Didn’t Become Customers Prospects that: • bought or not • convert or not • respond or not Data that clearly distinguishes the two groups
- 59. © 2014 SiriusDecisions. All Rights Reserved 59 Starting Point No Historical Data Fit the profile Don’t fit Prospects that: • Have a business problem • the motivation and resources to solve it Data that clearly distinguishes the two groups
- 60. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 60 Entity Predicted Need to find best contacts within target accounts? Predictive lead scoring can reach much deeper into a contact’s world to determine who is most likely to be involved in a buying cycle. Job Role Common Titles Company
- 61. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 61 Entity Predicted Need to find best contacts within target accounts? Predictive lead scoring can reach much deeper into a contact’s world to determine who is most likely to be involved in a buying cycle. Company Hiring Tech Ecosystem Prof. Communities Job Role Common Titles Content Engagement Social Media Interaction MAP PLS
- 62. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 62 Entity Predicted Modern data science can also reach deeply into online digital artifacts to unearth evidence of business problems and buying initiatives. Need to identify best company targets within large addressable universe?
- 63. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 63 Entity Predicted Modern data science can also reach deeply into online digital artifacts to unearth evidence of business problems and buying initiatives. • Corporate websites • Press releases • Job postings • Application signatures
- 64. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 64 Source of Predictors What is likely to be most predictive may be at the contact or the account level, and gleaning information from both is normally important. Top down Bottom up Some PLS providers collect and analyze data on contacts in order to predict what businesses are doing Some providers focus primarily on business level indicators to determine where the opportunities are
- 65. SiriusPerspective: © 2014 SiriusDecisions. All Rights Reserved 65 Source of Predictors What is likely to be most predictive may be at the contact or the account level, and gleaning information from both is normally important. Top down Bottom up The best models typically include both prospect and account level predictors
- 66. On Predictive… 4 Important things to know to make good predictions
- 67. © 2014 SiriusDecisions. All Rights Reserved 67 Conditions For Good Predictions Past behavior >> Future performance!
- 68. © 2014 SiriusDecisions. All Rights Reserved 68 Conditions For Good Predictions Past behavior >> Future performance! High-frequency, habitual situations and people are more predictable than rare ones
- 69. © 2014 SiriusDecisions. All Rights Reserved 69 Conditions For Good Predictions Past behavior >> Future performance! High-frequency, habitual situations and people are more predictable than rare ones Larger data sets enable more reliable predictions Stories are dangerous!
- 70. © 2014 SiriusDecisions. All Rights Reserved 70 Conditions For Good Predictions Past behavior >> Future performance! Larger data sets enable more reliable predictions Stories are dangerous! Predictions work best over short time intervals Tomorrow’s prediction is more accurate than the one for next week High-frequency, habitual situations and people are more predictable than rare ones
- 71. © 2014 SiriusDecisions. All Rights Reserved 71 Conditions For Good Predictions Past behavior >> Future performance! The anticipated situation must be essentially the same as the past situation Larger data sets enable more reliable predictions Stories are dangerous! Predictions work best over short time intervals Tomorrow’s prediction is more accurate than the one for next week High-frequency, habitual situations and people are more predictable than rare ones
- 72. © 2014 SiriusDecisions. All Rights Reserved 72 Making Predictions Count Insights “Something I don’t know.” Does the model prioritize or find prospects based on criteria a sales rep cannot readily acquire him- or herself?
- 73. © 2014 SiriusDecisions. All Rights Reserved 73 Making Predictions Count Insights “Something I don’t know.” Does the model prioritize or find prospects based on criteria a sales rep cannot readily acquire him- or herself? Compared to selecting prospects based on current methods, what improved conversion (lift) does the model provide? W.A.R. Wins above replacement player
- 74. © 2014 SiriusDecisions. All Rights Reserved 74 Making Predictions Count Insights “Something I don’t know.” Does the model prioritize or find prospects based on criteria a sales rep cannot readily acquire him- or herself? The promise of big data is finding important clues about which prospects will buy. The danger is that many variables are related but make little difference to that prediction Broad v Big Data Compared to selecting prospects based on current methods, what improved conversion (lift) does the model provide? W.A.R. Wins above replacement player
- 75. © 2014 SiriusDecisions. All Rights Reserved 75 Terminology “Big” Data? Big in what way? Big Data - Buzz word. Doesn’t mean anything in particular or officially. When “big” = Volume: many measures, records, repetitions, etc. When “big” = Breadth: lots of new and interesting things measured
- 76. © 2014 SiriusDecisions. All Rights Reserved 76 Terminology Machine Learning Machine Learning- Buzz word. Many propensity modelers and predictive lead scoring vendors use the term In general, it refers to the automation of the process of incorporating feedback loops within analytic algorithms. It does not refer to something special about the statistical procedures themselves.
- 77. © 2014 SiriusDecisions. All Rights Reserved 77 Key Take- aways • Marketing automation provided a great step forward in lead qualification • Current lead scoring does not account for enough of the variance in lead conversion • Modern data science can generate proxies for questions your best salesperson would ask prospects if he/she could reach them all • It is possible to model contacts, accounts and even existing customers • Marketers should understand key considerations for making good predictions
- 78. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Replace Traditional Scoring with Predictive Scoring?
- 79. © 2014 Mintigo. All Rights Reserved. www.mintigo.com It depends….
- 80. © 2014 Mintigo. All Rights Reserved. www.mintigo.com CASE STUDY #1
- 81. © 2014 Mintigo. All Rights Reserved. www.mintigo.com • B2B SaaS Core Product: VisitorTrack • Global clientele across various industries such as tech, manufacturing, HR, & retail • Lots of leads, no scoring system previously
- 82. © 2014 Mintigo. All Rights Reserved. www.mintigo.com No Scoring To Predictive Scoring For Fit • A: Great fit! Both company & prospects match netFactor’s CustomerDNATM • B: Company fit, but prospect doesn’t match buyer profile • C: Company does not match CustomerDNA • D: Low quality data (i.e., bad emails)
- 83. © 2014 Mintigo. All Rights Reserved. www.mintigo.com CASE STUDY #2
- 84. © 2014 Mintigo. All Rights Reserved. www.mintigo.com 12+ products
- 85. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Already Have A Multi-Product Lead Scoring Explicit-Demo/Firmographic • Contact data • Job title • Industry • Custom fields Implicit-Behavioral • Web visits • Email engagement • Content downloads • Webinar reg/attendance • Trial downloads/activations • Product usage • Form completions
- 86. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Great Rates, but wait… 0.05 % 0.14 % 0.81 % 2.15 % 0.00% 0.50% 1.00% 1.50% 2.00% 2.50% Sales Promo CR by Lead Score Great conversion rates, but: • Limited to track-able implicit behavior and explicit form completions • Scoring data = time to collect, build, maintain • We are only human!
- 87. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Predictive Score Identifies Target & Cross- Sell Opportunities In Real Time Test Ops Dev 42 82 19 24 11 95 77 79 35 6
- 88. © 2014 Mintigo. All Rights Reserved. www.mintigo.com CASE STUDY #3
- 89. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Traditional Demo/Firmographic Scoring Mintigo’s Sweet Spot: – Job Titles: • Demand Gen, Marketing Operations • General Marketing Management – Company Size of 1,000 employees and above – Users of Eloqua, Marketo and/or Salesforce.com – High Tech vertical, companies such as:
- 90. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Expanding Into Financial Services
- 91. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Traditional Scoring based on: – Job Titles: • Demand Gen, Marketing Operations • General Marketing Management – Company Size of 1,000 employees and above – Users of Eloqua, Marketo and/or Salesforce.com – Industry = Financial Services Predictive Scoring based on: Traditional firmo/demographic score to determine fit for new market, Predictive score to determine propensity to buy
- 92. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Predictive Score identifies propensity to buy Traditional Score shows fit based on demo/firmographic data
- 93. © 2014 Mintigo. All Rights Reserved. www.mintigo.com HOW MINTIGO WORKS Quick Overview
- 94. © 2014 Mintigo. All Rights Reserved. www.mintigo.com How Mintigo Predictive Lead Scoring Works
- 95. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Getting Started with Mintigo Marke8ng Need Assessment/ Data Discovery Build Predic8ve Model Predict / Score Leads and select Marke8ng Indicators Score Mystery File Validate model by iden8fying % of Closed Won Opportuni8es in scored Mystery file Set up real-‐8me lead scoring and data append 1 2 3 4 5 ~2 weeks 6
- 96. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Q&A
- 97. © 2014 Mintigo. All Rights Reserved. www.mintigo.com Host: Tony Yang Director of Demand Gen Mintigo @tones810 Guest Presenter: Kerry Cunningham Research Director SiriusDecisions THANK YOU!