As I discussed in my earlier post, analytics or data science teams know that two key challenges for analytics projects are making sure you solve the real business problem (framing the problem) and making sure you can operationalize the result (deployment). In this second post I am going to talk about deployment.
Once you have a predictive analytic model, data mining output, a machine learning result or any other analytic output, the answers to 5 questions determine how well deployment is going to go:
- Does the model evaluate well against business objectives?
“The operation was a success but the patient died” is an old trope but for analytic teams a real one. Just because an analytic model is predictive or statistically strong, does not mean it will improve business results. Make sure the model evaluates in business terms not just analytic ones. - Who’s using the model, to decide what, where and how?
To deploy a model effectively you need to know who is going to use the model, the systems it is supposed to be embedded in, who has to believe it and how explicable it needs to be. Make sure the model you have matches its planned use. - What will it take to make the model actionable?
Many decisions involve rules and regulations as well as analytics. Making your analytic model actionable means understanding and managing the rest of the decision-making as well as the analytic. - What’s the plan for capturing and analyzing data about model effectiveness?
Most organizations are looking for an ROI from analytics so you’ll need to capture data about the analytic results, about how those analytic results changed decision-making and about how that decision-making improved results. What’s the plan? - When do we need to revisit the model to make sure it is still effective?
And lastly models age. So make sure you have a plan for updating and sustaining the model over time to ensure it keeps delivering value.
All these questions help ensure a successful deployment and all can be answered using a decision model. If you are trying to make sure that analytics are deployed, learn more about deployment in our webinar Analytics Teams: 5 Things You Need to Know Before You Deploy Your Model.