Many companies are investing in predictive analytics models to better predict things like churn or fraud, or to identify better advertising and selling strategies. But before you can plug data into these models, you need to understand your current data.

Peter and I gave a talk on this topic at H2O World in November, which was a good fit, since H2O is creating an open source machine learning platform to help companies make smarter predictions.

In our talk, we focused on the key components you need to set yourself up for predictive analytics success. Here is an overview of the topics we covered.

What you need to understand your existing data

  • Create a team with a champion who can allocate resources and budget; the key stakeholders who need to give input on their requirements; and people with the technical background required to collect, store, and visualize the data
  • Understand what types of data you can collect, which tools are available, and how to create an effective data model
  • Understand how to address various challenges in data collection: accessing data stored in multiple locations, differentiating between duplicated data vs. new information, enriching data, and following privacy regulations
  • Figure out what your data is telling you by asking the right questions. What are users doing or not doing? How many users log in every day? How many people are converting to be paid users?
  • Agree on your most important business or product problems and what the right metrics are. Can you impact these metrics by understanding them and modifying your business or product?

What you need before running a predictive model

  • Understand what predictions you are trying to make. How will these predictions impact your business or product?
  • Understand the leading indicators that will influence your business or product. Be sure to define success — how will you know it’s working?
  • Understand who will see the results of this data. Is it for your internal team or external clients?

Additional Thoughts

Iterate, Iterate, Iterate. The analysis will never be perfect the first time. You’ll learn things and need to make adjustments, whether it’s changes to the initial data model or the predictive model. That’s ok and expected. After all, the whole point of analytics is to discover things that weren’t already obvious.

With great power comes great responsibility. Make sure to think through your product and brand implications. What impact will this have on society? What privacy should you protect?

As you can see, it’s a lot to think about. But once you have these components in place, you will be ready for success!

Next Steps

If you’d like to get more details on the areas we covered, you can watch our full presentation below:

 

And if you’re interested in learning how Keen can help you lay the groundwork for predictive analytics success, please reach out to us for a free data consultation or try Keen for free. We’d love to chat!