The game is afoot – and this time around, it pits marketing automation tools against data science driven data products. Specialized digital marketing service providers against dyed-in-the-wool data geeks. Campaign management tools versus correlation driven, random forest, gradient boosting, data products.
Marketing automation tools have, in their short history, focused on delivering specialized services (serving up display advertising, managed paid search ads et al). But over the last few years, these tools have broadened their reach and gained access to considerable data – emerging beyond their specific channel of interest. Tools such as Tag Management Systems, with their data layer, seek to become the data-broker of record. Offerings such as Data Management Platforms (DMP) integrate deeply into re-targeting platforms as well as with first party data and deliver fine-tuned audience segmentation and other behavior profiles in near real-time which can be acted upon immediately to serve better content and deliver richer engagements with the visitor.
Data Scientists in the digital realm have often prospered by:
- A cross-channel online focus – bringing together online data that, until now, was siloed and typical marketing automation tools had no visibility to
- Combining online and offline (company owned) data sources to generate insights that an online-only focus will not be able to deliver
- Adding third-party data (frequently purchased) to further embellish information about their customers and using the same for yielding further insights
The new wave of marketing automation tools are beginning to step into the first of these areas – by beginning to combine data across multiple online channels (website, mobile, social media, email et al). Offerings such as Audience Stream from Tealium are an example of such a product.
The marketing automation tools are also beginning to tap into the third space – bringing in third-party data to combine with company owned data – though the scope is presently limited to third-party online behavior of the visitors the company is interested in. This is the specialization that DMP from companies such as BlueKai offers.
How long before these marketing automation tools provide a method for pulling in offline data to significantly improve the analysis they can provide? And once the data is available, several push-button models can be added on top to deliver interesting insights. Some variation of this is beginning to happen already and by 2020, we may well be dealing with marketing automation tools that cover multi-channel online data as well as online + offline data – with all of it leveraged by push-button models that yield interesting new insights.
So then, where to for the Data Scientist of the Digital proclivity?
For now, there is plenty of gap still remaining that the home-grown (or consultant-driven) Data Science team can deliver insights through the gaps identified above. But then, the team has to continue to expand focus – both onto new data sources (in the world of IoT), in discovering new relationships (graph like approaches to discover and ask more interesting questions), processing larger amounts of data to generate new insights (using new statistical approaches as Math begins to catch up with Big Data), and really spending more time answering the frequently asked, but seldom answered, questions (“what is the customer’s intent when they arrive at a site”, “how best to move the customer along the conversion funnel”).
The age of marketing automation has arrived – coinciding with the increased burden on digital marketing to generate revenue producing leads. The scope and reach of marketing automation will continue to grow and will continuously challenge data science teams to move up the value-chain and deliver deeper insights and focus on answering the harder questions for the marketer.
Good piece, Sri. But that future has already arrived — there are many marketing automation tools that do all three things you list. Some are primarily for lead scoring (Mintigo, Lattice Engines, Infer, Fliptop), some for customer success (Totango, Optimove, Gainsight, ServiceSource), some for personalization (AgilOne, BlueConic, Monetate, Aginity), and some for audience management (X+1, IgnitionOne, Lytics.io). They all combine data from multiple online sources, offline sources, and (in many cases) third party sources, and most incorporate predictive modeling and decisioning. We call them “customer data platforms” but the name matters less than the fact they are giving marketers new capabilities without the need for extensive support by data scientists. As you point out, this should free the data scientists to do more challenging work. So everybody wins.
Thanks for your comment David. Indeed, much progress has already been made and are limited perhaps only by how much data they can get their hands on (overcoming years of data silos built into existing architectures).
That is an interesting question you ask in this article.
Just one question. You write:
“using new statistical approaches as Math begins to catch up with Big Data”
What do you mean exactly by that? What new math or statistical tool are you talking about?
Hi Thomas
Some of the math is not entirely new (Lasso and Least Angle Regression – they date back to 1996; or greater use of topology based algorithms; or even graphs for that matter).
As you know, with Big Data, not only are sample sizes growing – creating spurious correlations – but also number of features that may be interesting is also rapidly growing. Techniques are being developed to deal with this growing complexity.
Btw..there was an article in Wall Street Journal last year that captured some of this at a very high level:
http://online.wsj.com/articles/SB10001424127887323452204578288264046780392
I agree with David that good tools free your best people to do more interesting work – and that is especially true with Marketing Automation. Sure they come with “built-in” lead scoring and segmentation, but they certainly don’t tell you where those breakpoints are and how to best utilize the information. In fact, I will argue that it creates more demand for data scientists because the implementation of their research is easier to incorporate into business practices. So please, hold your ground and pull up a chair to the marketing strategy table.
Very good point about applicability of models to existing business process Sheryl. Third party pre-built models have certain limitations, including – not necesarily being ready to use in an existing business process, the data used in building the models may not line up very well with the data that is relevant to the company, the feature set used to build the models may not cover the requisite ground and so forth. These gaps are what consulting companies then thrive upon – since they will build you models starting from scratch using data and features specific to the company, and frequently in tune with existing processes.
This is the “buy versus build” discussion, but now with data products. Some hybrid approach will win out in the end.