Any new predictive analytics project requires development and engineering resources—which starts with getting support from management. And one of the first questions executives will ask is: How accurate is the model?
The accuracy question often leads to a trap. Companies wait for the best accuracy they can get before going live. This waiting game could delay your predictive analytics project indefinitely—and it’s completely unnecessary. In fact, you should go live with less than 100 percent accuracy.
How to Assess the Accuracy of Your Predictive Model
When talking about accuracy, you can’t compare your current predictive model to what it could be with more fine-tuning. It’s important to look at the accuracy of your model in comparison to what you do without the model.
For many companies, they make decisions based primarily on gut feelings. This may sound random but it’s not necessarily a bad practice. The key is to measure the accuracy of that gut feeling. If your predictive model preforms much better than your guesstimates, you know it will be worth moving forward deploying your predictive analytics. Over time, you can continue to tweak the model to improve accuracy even more.
If we wanted to compare apple to apples, we could train a model to answer a single predictive question—then compare that with the answer based on gut feeling. Are they the same? Which is more accurate? Compare the two answers with the actual outcome.
Say you are trying to predict which customers will churn. Maybe your starting accuracy is 65 percent. That may not sound great, but it could be an order of magnitude better than the current guesses you are making. In that case, you should take this model live. Over time, you can work toward improving your accuracy incrementally in future updates.
Figure out how to solve common challenges of predictive analytics, including how to distribute the information to the right people and incorporate workflows inside the application so your users can act based on the insights. Listen to the data and continue to improve the model, your distribution strategy, and the actions your users can take.
Don’t Wait for the Stars to Align!
Reasonable accuracy does not mean perfect accuracy—and a reasonably accurate predictive model may be worlds better than what you currently have in place. Don’t wait for perfection. Once you deploy your predictive analytics, the feedback from end users will give you a baseline so you can continue to adjust and improve the model.
Once you’ve started demonstrating the return on investment (ROI) of your predictive analytics, you can add more data and incorporate new insights into other parts of your business workflows. Every small success will give your organization significant boosts in competitive differentiation and revenue drivers.