At New York Advertising Week yesterday, I attended “Ways to Simplify the Ecosystem”, a panel session hosted by Dana Hayes, Group VP and head of Global Partner Development at Acxiom. It was a lively and informative session that touched on the subject of how prepared marketers and their organizations are to extract useful customer insights from the volumes of data now available. In the Q&A session, a woman working on the creative side asked an excellent question. In summary, she said, data scientists are backwards looking and analytical; creatives are forward looking and visual. How do you get these two teams to work well together given they have trouble communicating with each other?
Jason Kodish, the Global Chief Data Scientist for Digitas/LBI, who has a 380-person data science team (ok, I’m jealous) answered by saying that they pair up data scientists and creatives to make them work closely together. That’s a good answer, but as a solution it provides a necessary but insufficient condition for creating a successful data-driven marketing organization.
So what I want to do now is share with you how we at Acxiom Research, a team that conceives and builds intelligent, data-driven products, do this. We are not a marketing organization in the mold of an agency, which has to figure out how to understand audiences and optimize marketing programs and campaigns. But we face very similar challenges when it comes to building products to sell to those organizations. Our products are both data-driven and AI-driven. Making them great requires product visionaries (visualizers) working with engineers and data scientists to build smart products.
First, let’s look at the team. The product manager is ostensibly the “face’ to the customer and produces the market requirements document (MRD) and product requirements document (PRD). The data scientist builds algorithms to deliver the intelligence that the PRD specifies. The UI designer delivers a user experience that delivers all the functionality and delights the customer. The engineers build the product that the UI designer, PRD, and algorithms make necessary.
In our world, you cannot be a product manager for data driven products unless you can understand how to work with and analyze data (e.g. patterns of product usage, log files). UI design for data-driven products requires special skills: the ability to analyze the large volume of data now available to inform the design, and a deeper understanding of human factors than those embedded in traditional software. Let’s face it, it’s a bit harder to create an engaging user experience when the machine can talk back to you. Engineering has to have exposure to data science toolkits and platforms in order to build systems that are optimized for intelligence. Data science and engineering have to become visual thinkers in order to creatively apply data and data-driven systems to build products that provide elegant customer experiences.
In other words, every member of our team needs to make data-driven decisions and have the skills to do so. Equally true, the left-brain (analytic types) need to have the ability to think visually to understand the viewpoints of and pressure points of the product people, while the right-brain types need to be trained, if not to like math, at least to be comfortable with it. This bridges the vocabulary, viewpoint, and communication gap between the very different personality types and allows them to work as an integrated high-performing team.
So here are the tools and techniques we use to create an integrated team that communicates easily and creates world-class data-driven products. Hope you find these useful.
Teaming. Like Jason, we put together teams that consist of a product manager, a UE designer, a data scientist and an engineer who work together to conceive, design, prove the business case for, and take to market a new product.
Product managers are trained in SQL. When we interview product managers, we test their SQL skills. Passing the test is not a requirement for hiring, but it quantifies their skill level with databases and data, if they have any (many don’t). Immediately upon hiring, those who need SQL training are put through a basic SQL class, either online or a classroom-based program.
New product managers and data scientists do deep dives into our data. For the first six months, our product managers and data scientists do basic projects/tasks that force them to learn every aspect of our data. Six months is a minimum timeframe to learn enough to become productive. Although we don’t put our designers through the same rigor, we do look for designers who have worked extensively with data-driven products and interfaces, allowing them to ‘cross the chasm’ with engineering and data science.
New product managers are trained in doing advanced VRDs and wireframes. Not all product managers are visual nor are they outward-focused. In our group, product management and product marketing are handled by one person, so by definition we hire only people who prefer to talk to customers, have market vision, and are good communicators. However, many of these folks are not visual thinkers. So we train them to develop VRDs and advanced wireframes according to Labs standards to make sure they are ambidextrous and can guide both design and technical teams.
Data scientists and engineers participate in the creation of personas and customer journeys. Creating personas and customer journeys is a visualization exercise, so it helps the technical types to get used to rummaging around in amorphous problems and helps develop their left brain.
Product managers, designers and engineers must take Stats 1 and 2. There are many online courses that provide a solid but condensed reintroduction to statistics. (We even allow team members to use Statistics for Dummies 1 and 2.) Basic statistics does three things:
- It reintroduces people to probabilistic thinking.
- When necessary, it removes the fear of doing statistics and shows people they can do math if they put their mind to it.
- It provides a baseline for taking Introduction to Statistical Learning
Everyone takes Stanford’s Introduction to Statistical Learning. This is probably the single best course available online to teach the semi-technical most of what they need to know to be a deeply data-driven thinker. What I learned from taking this program and from experience is that there are about 15-20 statistical tools that cover the majority of problems that anyone other than the deep data scientists need. These include concepts like the difference between variance and bias, multivariate linear regression, multivariate logistic regression, k-fold cross validation, and bootstrapping. This class covers them all. You do have to do a little coding in R, but it really isn’t required if you just want to learn the concepts at a high level. As an aside, I take this class repetitively every year to keep my skills fresh.
Engineers and data scientists must take Andy Ng’s Machine Learning Course on iTunes University. Andy’s original course was delivered via iTunes University and was deeply mathematical. He ‘dumbed-it-down’ substantially when he ported it to Coursera. So I make sure the engineers and data scientists use the iTunes version. If any product managers or designers want to take the class, I recommend they take it on Coursera.
Everyone is trained in Design Thinking and everyone participates in brainstorms. One of the luckiest facts of my life is that my advisor in graduate school was David Kelley, founder of both IDEO design and the Stanford Design School. David has been teaching design thinking, which is all the rage today, for over 40 years. I was an early protégé and adherent of the method, and it is one of the reasons that I have been so successful as a product manager over the years. So I take everyone on my team through design thinking exercises, and ultimately they must each lead one or more design thinking sessions to prove they have mastered the techniques involved.
We hold weekly reviews of academic research on topics applicable to our projects. Each project we undertake usually involves a new set of statistical methods, which we often discover through deep dives into academic research and patents. Product managers are required to read ANY technical research related to their products. If taking on academic literature is overwhelming, we walk them step-by-step through methods that allow them to read without becoming frustrated by the equations. Designers and product managers are required to participate in research reviews, to again expose them to and make them comfortable with deeply technical concepts and literature.
Product managers attend Duarte’s Visual Storytelling Class. Nancy Duarte has been one of the top graphics designers in Silicon Valley for almost 40 years. In that time, she has written extensively on and given numerous talks about visual storytelling. Her TED talk is one of the most popular for marketers. Nancy has created a two-day class on visual storytelling that costs $1,600 and teaches all her best practices. Every member of my team, especially the engineers and data scientists, go to this class in their first year.