This post was originally part of a TrustRadius blog, “Predictive for ABM: Interview with J.J. Kardwell, President and Co-Founder of EverString”.
Foundation: The Potential of Applied Data Science
Before predictive marketing SaaS companies emerged, there were consulting firms that saw the potential of using applied data science to tap into and tie together internal data sources, giving marketers insights they wouldn’t otherwise have had. In a simple sense, for a marketer who is interested in doing segmentation, the application of data science enables segmentation on tens of thousands of variables instead of just a few. The earliest custom-built predictive deployments by consulting firms cost millions of dollars, and were only affordable for Fortune 500 companies. Human beings with PhDs built predictive models over the course of a month or more, and then the client could filter data on new and existing customers through that model. This can be immensely helpful, even if you are just using the historical data a company already has. For example, by looking at simple things like past buying behavior and usage data, the customer could recognize early warning signals for churn. As an example, if a customer’s logins and spend are going down by 25% every year, then there is likely a problem.
There are now companies that self-identify as “predictive marketing” or “predictive analytics for marketing and sales”, and which have been around for almost a decade, but those are businesses which started as consulting companies. When pure SaaS companies, like EverString, entered the market a few years ago, these older predictive consulting firms tried to reposition as software companies. Most of the SaaS companies in this market, which were born as pure SaaS and pure data science companies, started between 2010-2013. These are the companies that are now out front in defining the market.
Key Technology Transformations and the Emergence of SaaS
Several key transformations at the technology level (independent of any individual company) made it possible for pure SaaS companies to emerge in this category.
One factor was the emergence of cost-effective on-demand cloud computing (from companies like SoftLayer, AWS, Rackspace, and VMware vCloud Air). This meant startups could leverage that physical infrastructure to get started much more cost-effectively. In our business, if you’re doing predictive right, and you’re doing it at massive scale, the infrastructure strain becomes significant: crawling for data, high-speed compute, data storage, etc. During our seed round in 2012 we raised $1.7 million (we have since raised an additional $77 million). A decade ago, a new predictive company would have spent most of that money buying servers, without accomplishing much in terms of building the product or the business. In that regard, cloud hosting and infrastructure-as-a-service were foundational in the development of predictive marketing SaaS offerings.
The second foundational element is parallel computing; MapReduce and the emergence of Hadoop as a mature technology layer. That maturation of Hadoop happened in large part through the emergence of businesses like Hortonworks and Cloudera, which made Hadoop technologies more usable in an enterprise-grade setting.
The third major contributor to the emergence of this category is innovation in data science methods, and particularly in the areas where EverString has been leading the industry in innovation around automated feature selection and automated model building. What currently takes a consulting company one to three months to do, can now be done in a matter of hours by a fully automated platform. Turning data science into software is one of the most difficult things to do in our market.
With these developments, a market that had initially been consulting-driven is giving way to true SaaS. I have seen this trend replicated across many industries–early market entrants with consulting or “technology-enabled services” business models almost invariably get displaced by pure software solutions.
From Predictive Lead Scoring to Full-Lifecycle Predictive Marketing
What emerged from the early days of predictive was a category that, a year and a half ago, still self-identified as “predictive lead scoring.” EverString really started our business 15-18 months ago; before that we were in R&D and platform development mode. Until February 2014, our entire team was just 12 data scientists and developers.
We’ve set out to change the way people think about this space. “Predictive lead scoring” is a narrow and terrible moniker for this category. Although some companies still use that term, our view is: if you believe in predictive—if you think applied data science works—why would you only use it inside your existing pipeline? Why would you limit the application of predictive to only within your CRM or marketing automation system, solely to reprioritize existing leads? Why not apply it outside of your pipeline as well, to find completely new prospects? The second problem with the notion of “predictive lead scoring”, is that you should not just focus on the “lead” (i.e., contact or person). The best-in-class predictive deployments start by evaluating companies (i.e., accounts). Focusing on the company before the person enables you to extend predictive outside of your existing pipeline, to find new companies that look and act like your best customers.
To put this in perspective, in the U.S. there are approximately six million companies that are large enough to have a website. Typically, the customers we talk with have approximately 50,000 to 100,000 prospects in their CRM and/or marketing automation. What about the 5.95 million companies that are not in your database? Why not use applied data science to have a view on them as well? This type of proactive “predictive demand generation” is more difficult to do than simply reprioritizing your existing pipeline. Simply repackaging existing data about your contacts and their transaction histories from your marketing automation and CRM systems is not enough anymore. We collect massive amounts of external data, which enables our customers to have a point-of-view on every single potential prospect in their addressable market, regardless of whether that prospect is already in their pipeline. You shouldn’t be marketing to every company, but you should have a point-of-view on every one—then you can make an informed decision about which companies and people are worth marketing to, advertising to, and selling to.
The Predictive Space Today: SaaS vs. Consulting
Taking a broad look at this market and its evolution, there is a clear distinction between SaaS and consulting businesses. It’s very easy to tell them apart. Don’t ask the vendors, ask their customers: “How long did it take to build and deliver the predictive model?” If the answer is 30 days, or 6-8 weeks, it is painfully obvious that humans are building the model manually. That is not a software company, as a software company should be able to build a model in hours rather than weeks or months. Regardless of positioning or glossy messaging on websites, the time to build a model and fully implement is one of the best ways to cut through the spin and test if a company is actually trying to sell you software or consulting.
There is also an important distinction between companies that take a comprehensive view of predictive’s potential, versus those that only deliver lead scoring, whether because of technology limitations or a narrow belief. There is a very small universe of people doing anything else beyond lead scoring with predictive.