March 2015 Machine Learning Meets Marketing Frost & Sullivan Analysis by Sandy Borthick Big Data & Analytics (BDA) Volume 3, Number 3
Machine Learning Meets Marketing Table of Contents Executive Summary... 4 Introduction... 5 Machine Learning for Lead Scoring... 5 Four Startups That Provide ML-Based Lead Scoring... 7 Mintigo Identifies CustomerDNA... 7 Radius Intelligence Targets Enterprises Selling to SMBs... 8 Infer Challenges Enterprises to Compare Predictive Modeling Results... 9 Fliptop Takes a Consultative Approach... 11 New Lead Sources, Old Software, and Vendors of Record... 12 A Simple Solution from PerfectLeads... 13 How LeadiD Addresses Lead Quality Issues... 13 Repurposing Existing Software, Waiting on Vendors of Record... 13 Salesforce Has the IP and the AppExchange... 14 Adobe Enhancements Are Most Relevant in B2C... 15 Oracle Gears Up through Acquisitions... 15 How Ready Are Buyers and Sellers for ML-based Lead Scoring?... 16 The Last Word... 20 List of Exhibits Exhibit 1: Generating a Machine Learning Model (Classifier)... 6 Exhibit 2: Mintigo Process... 8 Exhibit 3: Radius Segment-Building Screen... 9 Exhibit 4: Infer External Signals Screen... 10 Exhibit 5: What Fliptop Signal Data Looks Like in Marketo... 12 Exhibit 6: External data sources are not very popular with large enterprises... 17 BDA 3-03, March 2015 Frost & Sullivan, 2015 Page 3
Machine Learning Meets Marketing Executive Summary Machine learning is the latest in a series of data-driven technology developments that are disrupting and transforming the Customer Experience, Marketing and Sales Analytics category of the Big Data and analytics (BDA) market. Stratecast Frost & Sullivan has identified more than vendors who supply solutions in this category. 1 Competition will require each of them to develop or partner to deliver machine learning (ML) capabilities for lead scoring. The basic idea is that ML algorithms, in the right hands and with the proper data, can enable moreinformed gathering and evaluation (scoring) of marketing leads. The higher-level value proposition is that, if businesses apply machine learning in this way, they will be able to adjust their sales and marketing efforts to address customers and prospects with the highest propensity to purchase. Machine learning algorithms have been used by academic and scientific researchers for decades to discover patterns in new data based on previously processed datasets. Now, vendors are commercializing these algorithms in cloud-based applications that combine ML with additional functions, new data sources, and user-friendly interfaces. Marketing departments can use these new solutions, which are essentially ML applications that have been trained with data on existing customers, to score sales leads based on their propensity to buy. The variety of ways in which MLbased lead scoring solutions are coming to market means that there truly is an option to satisfy every level of budget, analytic skill and marketing automation maturity. This report explains why these new solutions represent a major improvement over existing marketing automation measurements, how they work, and how different vendors are exposing machine learning capabilities in their lead scoring solutions. This report should be of interest to buyers, sellers, and current users of marketing automation (MA) and customer relationship management (CRM) solutions. 1 See Stratecast report: BDA 2-11, The 2014 Big Data & Analytics Vendor Directory, published October 2014. For a free copy of the directory, contact your account executive or e-mail inquiries@stratecast.com. BDA 3-03, March 2015 Frost & Sullivan, 2015 Page 4
Introduction 2 Whether one characterizes the marketing and sales process as a journey, a funnel, or a pipeline, there is no getting around the fact that this process is now generating more granular information than marketing and sales departments know how to handle. This is a function of the explosion of available data, and it reflects the continuing migration of business activities into the online world. The marketing department has been working to keep pace with this growing informational challenge for years, as email, Web sites, and social networking have gradually supplanted printed press releases, direct mail, catalogs and magazine advertisements. It was hardly necessary for marketing to prove that email outreach was cheaper and faster than postal mail, or that an engaging Web site and a responsive social presence have become basic business necessities. Activating these new channels of communication with customers and prospects has also enabled new forms of measurement, which marketing and upper management hoped would demonstrate their value. Marketing departments began by tracking click-through rates on Web sites and email campaigns. Many have since moved on, tagging and tracking their Web site visitors, counting tweets and likes in social media, and mapping the relationships among influential social network participants. Linking these data to composite descriptions of customers (personas) and conversion processes (funnels or pipelines) gave marketing a way to contextualize the available information, and to track the interactions that are intended to move (nurture) prospects toward the goal (purchase). Unfortunately, the effort that marketing has put into measuring these activities has produced little more than increasingly detailed portrayals of the process, from which only a disappointingly small effect (sales lift) could be attributed to digital marketing automation. In other words, although digital marketing practices are now widespread, and a variety of detailed metrics are being used to quantify their component activities, only incremental contributions to presales funnels and customer journeys have been documented. Implementers have been unable to clearly demonstrate that marketing automation delivers a substantial, quantifiable sales boost. This is the basic challenge that predictive lead scoring, based on machine learning, is poised to address. Machine Learning for Lead Scoring The rationale for predictive lead scoring based on machine learning (ML) takes only a moment to sink in, because it is both logical and straightforward: Rather than use digital marketing automation to push tailored content and online experiences to all comers, hoping that the recipients can be nurtured into moving through the conversion funnel, why not start from the other end of the 2 In preparing this report, Frost & Sullivan conducted interviews with representatives of the following companies: Fliptop Jessica Cross, Director of Marketing Infer Jamie Grenney, VP of Marketing Mintigo Atul Kumar, Chief Product Officer; Blake Tablak, VP of Sales; and Tony Yang, Director of Demand Generation PerfectLeads Brian Roizen, Chief Technology Officer and Chief Data Scientist Radius Intelligence Peter Tait, VP Marketing; and John Hurley, Director of Product Marketing Please note that the insights and opinions expressed in this assessment are those of Frost & Sullivan and have been developed through the Frost & Sullivan research and analysis process. These expressed insights and opinions do not necessarily reflect the views of the company executives interviewed. BDA 3-03, March 2015 Frost & Sullivan, 2015 Page 5
process and work backward? First, identify the characteristics of actual customers, then use those characteristics to sort and score a prospect list. This process reveals which prospects most closely resemble the best existing customers, and are thus the most likely to convert. Marketing and sales efforts can then be redirected to focus on those prospects, which should produce not only a tangible increase in revenue, but also a happier, healthier alignment between marketing and sales. Machine learning algorithms number in the dozens, and a full discussion of their characteristics and applications is beyond the scope of this report. The reader is directed to the excellent (and free) 2010 ebook, Types of Machine Learning Algorithms, New Advances in Machine Learning, from which Exhibit 1 is taken. 3 In the case of predictive lead scoring, the Problem shown at the top of Exhibit 1 is to develop a model, or classifier, that can be used to evaluate the prospect list. The next step, Identification of Data, typically refers to the customer or sales database, which many vendors can ingest directly from spreadsheets (as.csv or.xls files), and from marketing automation or CRM systems (e.g., Eloqua, Marketo, Salesforce). Some vendors offer additional connectors, or will write or partner to provide them, so that data can be ingested from other sources. Exhibit 1: Generating a Machine Learning Model (Classifier) Source: Taiwo Oladipupo Ayodele (2010). Types of Machine Learning Algorithms, New Advances in Machine Learning, Yagang Zhang (Ed.), ISBN: 978-953-307-034-6 3 Taiwo Oladipupo Ayodele (2010). Types of Machine Learning Algorithms, New Advances in Machine Learning, Yagang Zhang (Ed.), ISBN: 978-953-307-034-6, InTech, Available, at no cost, online from: http://www.intechopen.com/books/newadvances-in-machine-learning/types-of-machine-learning-algorithms BDA 3-03, March 2015 Frost & Sullivan, 2015 Page 6