Red Hat's Heidi Lanford to share her formula for becoming model-driven

Heidi Lanford is known for her leadership in analytics and data science. From her early customer segmentation work with Harley-Davidson that guided an important marketing reorg to her campaign and forecasting initiatives for Nortel that doubled revenue in two key verticals, Heidi has steered some of the world’s largest brands in successfully tapping into their data to better compete.

Now as Vice President of Enterprise Data and Analytics at Red Hat, the world’s largest open source software company, Heidi is making her mark once again. In less than two years with Red Hat, she’s implemented the company’s first enterprise data science platform for collaboration and a framework for certifying enterprise data assets—both part of a comprehensive plan that will help Red Hat scale data science in every aspect of its business.

Her work offers valuable insights in how to become a model-driven company—which, as most data science leaders know, is a significant challenge despite widespread excitement for and investment in data science. In fact, a 2017 MIT Sloan study found that while 85% of companies it surveyed believed models would enable them to obtain (or sustain) a competitive advantage, only 5% were using models extensively in their business.

How is Red Hat establishing a truly enterprise data science capability to drive adoption? We sat down with Heidi in advance of her Rev 2 session this month and got a sneak peek into her formula for success, which is centered on three pillars: collaboration, community, and adoption. The following is an edited transcript of our conversation.

On the importance of collaboration

Putting in place smart data scientists with lots of data at their disposal and great tools will only get a company so far. Many companies are transforming the way they use data throughout all parts of their business by collaborating, and if we don’t adapt, we know that we’re only going to reach a certain amount of success from our data and analytics work. For example, when it comes to model development, the best data science outputs come from lots of collaboration between data scientists and the business community. So at the front-end, we’re insisting that data science and analytics communities explain the value for any modeling project they’re going to embark on in ways that the business can understand. And in order to make the investment in producing models or other analytic outputs, the business must articulate: How will the model change someone’s day job or improve the customer experience?

Data scientists must also really understand the business so they can see if model recommendations make sense from a business standpoint and they need to review initial model output with business teams to get their feedback. Often business teams can provide insight on whether some of the trends and patterns surfacing are correlated, but more importantly if there is a causation effect at play.

On building a community for data scientists

Our data scientists have had education and experiences in different environments, and we need to provide ways that they can tap into each other’s knowledge. As a result, we’ve created an environment we call DAVE—the data analytics virtualization environment—that gives data scientists a space to share ideas, modeling techniques, data sources, and their subject matter knowledge, and to ask questions about business challenges they’re trying to solve.

On thinking beyond trusted data to trusted models

While we know we’re working off of our trusted data sources, we also want a process to determine which are our most trusted models—for example, are they using a large enough sample size, are they employing the right methodologies, do they have the correct documentation, can the technology infrastructure support the model, when does the model need to be refreshed, and so on. There’s both a development component as well as an operational component as we expand our governance practice. We’re putting in place a data science council that will deliver peer review to ensure that models are built with sufficient rigor. We’re just in the early stages of testing the process and identifying council members, but we see this work as vital in establishing a clear path to bring models to production and building trust that will ultimately drive adoption.

When it comes to driving adoption among end users, we also need to ensure that we have people across our sales, product, customer experience, and other teams who are fluent in “data as a second language.” Our data literacy program and “do you speak data” campaign are helping us bring data science to every level of the organization.

You can hear Red Hat’s full story (and struggles) and gain advice for establishing a truly enterprise data science capability at Heidi’s session, From Experiment to Enterprise: Scaling Data Science in an Open Organization, on May 24 at Rev 2 in New York.