Predictive analytics isn’t just a buzzword or a far-fetched notion. It’s here now, and it’s giving us the ability to use past data to predict what may happen in the future.
Recently, INDUSTRY Product Interviews hosted a Q&A with Brian Brinkmann, our VP of Product at Logi Analytics, about how predictive analytics is shaping today’s products—and how product teams can leverage it to add value to their applications. Check out some highlights from his interview here.
INDUSTRY: What is predictive analytics?
Brian Brinkmann: If you look at traditional analytics, it takes historical data and tries to understand what happened and why. Predictive analytics still uses historical data, but it uses more sophisticated learning algorithms to determine what will happen in the future. For example:
- Will a customer churn?
- Is the machine going to fail?
- Am I going to default on a loan?
- And… what can I do about it?
How is predictive analytics relevant to product teams?
For product folks who are building applications, predictive gives you an avenue into insight you wouldn’t otherwise have. You have the ability to solve business problems that couldn’t be solved before. And that means you can ask customers a different level of questions and probe into their business challenges. From a product perspective, that makes your application far more valuable. And if you have a commercially available software product, you have an opportunity to increase revenue and add value to that product.
What benefits have you seen for customers who are using predictive analytics?
One of our customers, Group FiO, offers cloud-based applications for mid-sized retailers. By using predictive, they were able to look at customer data and determine what kind of marketing campaigns they could use to provide a sales uplift. They were hoping for 10 percent, but they were actually able to achieve a 20 percent uplift by creating very targeted marketing campaigns. That’s all done by looking at customer behavior.
We’ve also had companies who are suffering from churn, and by looking through their customer data, they’ve been able to predict why people are likely to defect and what campaigns they can offer to retain those customers. These are just two examples, but there are many more.
What challenges do companies initially encounter with predictive analytics?
The first challenge is about the business problem. We always ask customers, “What are you trying to solve?” Product people often forget that technology is an enabler. Predictive analytics is not valuable in and of itself; it’s valuable in how it’s used to solve a business problem. I recommend identifying a problem you have first. Then determine whether you have the data to solve that problem—and whether it’s clean data.
Then the next challenge is around expertise. You don’t always have the data scientists to help you. One reason we created Logi Predict is to help the product team build a model in a wizard-driven, no-code approach. And that way the developers can build the model, and it takes the burden off the data scientists.
A third challenge is actually getting the insights out to the people on the front line who need them. We recommend you do it in an embedded fashion through your existing application—so people can take action all in one place.
How does Logi in particular help prepare data for predictive analytics?
We do things inside our product, Logi Predict, to help with data problems. We run extensive programs with our customers to automate the cleanup of data issues. The most common problem is missing data values. We have a smart data cleanup that cleans the data and makes suggestions to fill in missing values, and that really helps smooth out the process. And all of this is completely tunable and controllable.
Is predictive analytics something you have to go all in on as an organization?
I actually recommend you start small. You don’t have to do this in production; you can do it in your development environment to proof it out. You absolutely don’t need to re-architect your systems and policies. In fact, it’s much simpler to build a business case if you run a pilot project with just a few people and show a prototype of what’s possible. Even then, you can roll it out to just a few customers and learn and enhance it before you roll out to the general audience. You don’t need to spend millions on investment.
How long does it take to deploy a predictive model and see results?
Once the predictive model is created, deploying it is really straightforward and fast. The second part of it is—how do you determine how well it’s working? We recommend that as you’re taking action, you also measure the results of those actions. Then you can use that information to refine the original model. You’re creating a virtuous cycle: you’re getting information, creating a predictive result, using and testing the result, and then improving the base model with the information you’re getting. This helps increase the accuracy of your outcomes and tunes the model to what’s happening in real life.