Building a Data (Science) Driven Marketing Team

Marketers like to claim that we’re data driven, but are we… really?

Sure, we track our lead conversion funnel from acquisition through conversion to paid customers. We use campaign attribution to figure out where to invest our budget to generate the highest return. We personalize our emails, A|B test websites and landing pages, and create beautiful reports in Excel or Tableau proving our contribution to growth. Data certainly isn’t the problem. From Salesforce and Google Analytics to DMPs to enterprise-scale data warehouses and data lakes, marketers have more data than ever and plenty of ways to report on it.

The real challenge is still turning data into “oh shit, now what do I do now?” And because that’s hard, I think most of what we end up doing is making data-informed decisions — using data to reinforce our intuition and prior experience. Certainly a positive step forward from Mad Men marketing where an unknown half of marketing spend was wasted. But being informed by data is something very different from being driven by it.

If we’re honest, we’re missing the most important part of the data-driven marketing equation; the part that separates us from our own bias, uncovers the hidden insights not visible in pretty charts and graphs, and most importantly reveals the mathematical truth behind the decisions we’re making.

The data science.

Data science already sits at the center of many innovative organizations. For example, Alphabet created Google Brain, a research team dedicated to solving hard problems around deep learning and artificial intelligence. While mostly a research organization, some of the innovations from the Brain team have already made their way into consumer products like Google Photos. Tesla has analyzed over 100 million miles worth of autonomous driving data, using data science to build self-driving cars. Netflix uses data science to figure out what TV shows to produce.

Building Your Data Science Team

I see the data science team as an evolution of the marketing operations function, who are responsible for marketing technology, processes, and analytics. The head of marketing operations is often thought of as the Chief of Staff for CMOs, and is a natural home for data science efforts in marketing.

While there are lots of different titles for data science, there are two primary sets of skills you’ll to add to the marketing operations team:

1) Data scientists blend machine learning and business acumen to drive marketing teams to actionable learnings. They are responsible for formulating a growth hypothesis and then prototyping, validating, and deploying predictive solutions into the business. Common skills include expertise in machine learning and statistics; proficiency in predictive modeling tools like R, Python, and RapidMiner; data visualization and storytelling; and the ability to translate business requirements into testable hypothesis. Here’s how AirBNB hires data scientists.

2) Data engineers tackle the complexities of access to data, handling the infrastructure and tooling for the data scientist team. Marketers continue to drown in data, and this role helps ensure data scientists have access to the complete picture of a customer by connecting data across hundreds of structured and unstructured sources. Common skills include SQL/NoSQL database systems, data modeling and ETL tools, data warehouses, and Apache Spark.

There’s no doubt that data science roles are hard to find and expensive to hire. And the problem is only going to get worse in the next few years, as organizations scramble to recruit data science talent. McKinsey predicts that by 2018 demand for data scientists is projected to exceed supply by more than 50 percent.

Three Marketing Data Science Projects to Start Right Away

So you’ve decided to invest in building data science team — great decision! Here are the first three projects you should look at:

1. Use machine learning to score leads. Existing approaches to lead scoring are mostly based on gut and instinct, causing too many unqualified leads to pass through to sales and while better qualified leads sit untouched. A data scientist can combine data from all your sources — CRM, marketing automation system, product usage, social media — and then train a model to predict leads that turn into “Closed Won” opportunities with the highest frequency. If you are doing Account-based Marketing, you can score accounts in a similar way. Here’s where you can learn more on how we recently implemented predictive lead scoring at RapidMiner.

2. Prevent customer churn before it happens. Churn applies to every company with a recurring revenue stream. A data scientist can look at all the attributes of a customer (called features) and help identify the specific ones that predict someone is likely to churn. Then we can proactively target specific offers to the customer before they churn. Here’s a great article from David Skok on the financial impact of minimizing churn.

3. Use clustering to segment customers and personalize offers. Dominos Pizza uses over 85,000 structured and unstructured data sources to segment customers, allowing individual stores to tailor coupons.

So Can’t I Just Buy a Predictive Marketing Product?

Absolutely not. Data science is too important to be outsourced to a product. For example, at my previous company we deployed one of the most popular predictive lead scoring products and found that:

  • The data science was a black box. We knew which leads were better, but not why.
  • Every time we wanted to tweak the predictive lead scoring model, we had to wait for the vendor. Changes took months.
  • Each new use case for data science required purchasing a new product. For example, we wanted to look at building a predictive model to identify targets for our Account Based Marketing program, but it required purchasing a separate product.

But most of all, predictive marketing products can’t replace the curiosity and creativity of a skilled data scientist.

They provide the insight behind the action. They know how to formulate and prove a hypothesis. They understand the difference between correlation and causation, and how to prove it. Machines still need to be taught how to learn, and the black-box approach of predictive marketing products removes the humanity that is still required to deliver breakthrough data science.

(crossposted from

%d bloggers like this: