These days every data project is a data-science project — and business stakeholders must take an active role in data science to realize the expected value.
When most people hear data science, they think of a black box. The inner workings are complex and difficult to understand, but they know the output from it — effective ways of analyzing data to gain valuable business insights — is incredibly important.
I believe companies need to take a more scientific approach to innovation — whether it’s by improving customer service, cutting unnecessary costs or selling more products.
Many companies are trying to cultivate data science to gain a competitive advantage. Figuring out “how” is where they get tripped up. One of the simplest approaches to getting better at data science is by making data science approaches ubiquitous throughout the business, and by forming stronger partnerships between line-of-business people and data scientists. More on that later.
Let’s first look at how and why many companies are struggling with effective data science.
The challenges are around perception and behavior. Line-of-business people — business development, marketing, customer success, financial analysts and others — have become increasingly aware of the power of data science to create business value. Some have even implemented user-friendly dashboards and reports that offer sophisticated data insights that even someone with little or no data-science expertise can use. But many of these tools don’t incorporate real data science and aren’t a substitute for the expertise of a skilled, experienced data scientist who can solve complex problems. So for complex data-science requests, line-of-business people will approach data scientists about their problem, say something like “this is what will create value for me,” and leave it to the data science team to figure it out.
This more siloed approach is often a misuse of everyone’s time and may not generate the expected results. The data scientist will do their best to analyze the data using automation and machine learning. But being disconnected from the business decision-making means they often can’t do a good enough job of understanding and interpreting the data to get the insights they need. They’ll build amazing models that may not answer the right questions.
Organizations can break down these data silos in two ways. First, there should no longer be “data-science projects.” Every project and every app should have data science embedded in it. Companies that want to be data-driven and customer-centric have no other path forward. Second, data science shouldn’t be a secluded endeavor — there should be greater collaboration and partnership between the business and data scientists. Data science cannot be a black box.
How can line-of-business people and data scientists partner more effectively? There is a long-standing model available known as CRISP-DM, which stands for cross-industry process for data mining. This model was created in 1996, and I’ve found it to be very effective. There have been lots of updates and variations, but here are the key steps:
• Business understanding: Start by understanding your business’s challenges and what types of insights would provide benefits. Here is where a line-of-business person would give their data scientist a use case for analytics and its success criteria.
• Data understanding: Here the business people, data scientists and database administrator (DBA) should work together to identify the available data to support their use case, including the source of the data and if the data is complete and trustworthy.
• Data preparation: Now the DBA, with input and direction from the data scientist, extracts and structures the data that machine learning will evaluate in future steps.
• Modeling: The data scientist identifies and applies the right machine learning algorithms to the data.
• Evaluation: Business people and data scientists work together to look at the data-mining results and determine whether the model meets the business objectives. If the result are not acceptable, they return to the “business understanding” step and cycle through again.
• Deployment: Lastly, the business people work with IT and the DBA to determine a strategy for deploying the results. For example, they could integrate the model into a mobile app or a line-of-business application.
Line-of-business people and data scientists have much to gain by collaborating more closely on data science projects. Data is relatively easy to collect but harder to analyze. Right now, many line-of-business people are likely wishing they had better insights but don’t know where to begin. Data scientists may want to help but don’t have a grasp of the business problems.
By integrating data science more deeply into the business, and by developing a better working understanding of how data science works, including the CRISP-DM model, business people can be more effective partners and drive their data initiatives forward.