Data science is all the rage these days. It seems like it’s hard to walk down the street without someone mentioning how important data science is and how they plan to apply data science to big data to create predictive models and AI. The details don’t seem to matter; what matters is that WE’RE DOING DATA SCIENCE. So it goes.
Despite the ubiquity with which it’s mentioned, I’ve noticed that many marketers remain unclear about what data science can do for them specifically. How can it help you with your day job? How can it make you run more successful campaigns? How can more effective measures of success be developed? As with most difficult problems, it’s much easier to talk about what needs to be done than it is to figure out how to go about actually doing it. Of course, the “how” can be specific to your business context. Fortunately, there are models of thinking in data science that can assist.
Definition of Data Science
Let’s start with what I mean by data science. At its core, data science is the discipline of learning from data. Learnings can manifest themselves in many forms. When people speak of data science they often think about machine learning and predictive models, but perhaps just as important are analytical insights that provoke discussion and inform strategy. In either case, the goal is to learn from data and draw conclusions in a systematic way.
In marketing, we often speak of using analytical insights and predictive models to inform more effective outreach strategies. For example, we can build models to predict consumer churn, forecast revenue for the next quarter based on year-to-date results, and conduct website A/B tests to estimate the impact on traffic and conversion rates.
So how can you, as a marketer, use your data more effectively? How can you better leverage modern data science techniques and strategies? One way is to engage in what I call the Data Science Feedback Loop.
Data Science Feedback Loop
1. Define the Problem: Always begin by stating the business problem you are trying to solve and how solving the problem will impact the business. I cannot overemphasize the criticality of this step. It’s very tempting to begin work prematurely without fully understanding the end goal. It’s also very easy to confuse busyness with progress.
Marketing and data science should work together as partners to define the problem. Marketers understand their customers and data scientists understand the data. Take advantage of both areas of expertise when defining the scope of a problem. Prior to fingers hitting keyboards, marketing can—and should—ask data science for an analysis plan based on the shared understanding of the problem. This helps avoid the “illusion of agreement” that often plagues analytical projects.
Can this analysis plan change? Sure. The point of data analysis is to gain insight that you did not have before, and so as new insights are discovered the original analysis plan may have to change to accommodate these new insights. Analytical plans shouldn’t be carved in stone; they should be sketched in pencil.
2. Understand the Data: Once the business problem is defined, the question then becomes: what data do we have that can help us solve this problem? With marketers and data scientists working together, this question becomes easier to answer.
Data scientists should have a deep understanding of the “what” and “how” about data: what data is available and how to best view and summarize the data. Marketers, with a deeper understanding of the business context of the customers they are trying to market to, often have a more intuitive understanding of the “why”—the meaning—of the data. If the goal is to truly understand the data, both perspectives need to cross-pollinate. This cross-pollination allows data scientists to better learn the meaning of the data and for marketers to better understand what is possible with the data that is available.
3. Analyze and Model: Whether the goal is to provide insights via analysis or to build a predictive model, the first step should always be to design an analytical plan. Data scientists should think through what they’re going to do, how they’re going to do it, and why they should be doing it. Ultimately, what is possible is dictated by what the data says, so this plan may need to be revised or even completely discarded as new information comes to light. However, proceeding without a plan often leads to inefficiency and a lack of clarity of purpose.
There are many opinions about how to best go about analyzing data and build machine learning models. I won’t go into much depth here, but here are some principles I’ve found useful in my career:
- Work hard to be lazy: Do some detective work to see if there is previous work you can repurpose.
- Use proxy metrics: Try to use metrics that are not exactly what you have in mind but are closely-related and more accessible.
- Stay in the upper right quadrant of high-value, low-effort: Focus first on the parts of the analysis that have the most value and are the easiest to complete.
- Decide if a machine learning model is necessary: Building and maintaining machine learning models is not trivial. If you can satisfy your minimum requirements without building machine learning models then you should.
- Build as simple a machine learning model as possible: Simple models are easier to maintain and understand. Have a bias towards simplicity, even if it does sacrifice model accuracy.
- Brace yourself for failure: Not all modeling efforts are successful! You will fail sooner or later. Focus on getting to failure faster.
4. Communicate Results: The key here is to avoid “big bang” communication where intermediate results and progress aren’t communicated until final deadlines approach. Data science is an iterative process, and it follows that communication of progress or roadblocks should be iterative as well.
Marketers should maintain an expectation that they can get updates on the status of analytical projects at any given time. The mechanics of how these updates are communicated is ultimately up to the teams involved, but having regular conversations, in addition to any task/project management tools that your teams may use, is always a good idea. Keep these conversations focused on the original intent of an analysis and try and avoid diversions. I’ve noticed that there is a strong tendency for people to ask interesting, but ultimately tangential, questions about intermediate data analyses performed. The more people that are in the room and the more complex the analysis, the stronger this tendency becomes. Be cognizant of this and try to redirect the conversation back to what matters most.
Done right, communication between the data science and marketing teams will either provide solutions to the business problem or suggest refinements of the original business problem definition. In the former case, we’ve achieved business value and celebration is in order. In the latter case, we’ve still achieved business value but now we’re back at the beginning of the data science feedback loop and can begin the process again.
Begin with the End in Mind
I’ve talked about one way of thinking about how marketers and data scientists can work together to solve business problems. Regardless of the approach, in the end what matters is accelerating the pace at which business value is achieved.
Framing collaboration between marketers and data scientists as a feedback loop can help you focus on ways to shorten the amount of time it takes your organization to cycle through that feedback loop. This leads to more efficient teams, better insights, and faster results.
Stay tuned for a Jornaya report on this topic.
Chris Snyder is Manager of Data Science at Jornaya.