Machine Learning Times
Machine Learning Times
EXCLUSIVE HIGHLIGHTS
SHARE THIS:

3 weeks ago
Wise Practitioner – Predictive Analytics Interview Series: Richard Boire at Boire Analytics

 

By: Steven Ramirez, Conference Chair, Predictive Analytics World for Industry 4.0

In anticipation of his upcoming presentation at Predictive Analytics World for Industry 4.0 Livestream, May 24-28, 2021, we asked Richard Boire, President at Boire Analytics, a few questions about their deployment of predictive analytics. Catch a glimpse of his presentation, Developing Data Science Solutions in Industry/Operations: A not so easy process, and see what’s in store at the PAW Industry 4.0 conference.

Q: In your work with predictive analytics, what behavior or outcome do your models predict?

A: Over the course of my career, where I have had the good fortune of working with so many great individuals across many different industries, our  models  have been varied ranging from the simple to the complex. My career began  in the marketing area developing simple acquisition response models but then evolved into producing  solutions that predict credit loss. Overtime, the demand for these skills evolved into non-consumer behaviour type models  such as developing models to improve the operational procedures of a certain business process or the ability to predict call center volumes that are at the frontlines of customer service.

Q: How does predictive analytics deliver value at your organization – what is one specific way in which it actively drives decisions or operations?

A:  Clearly, our biggest success is when we have tangibly been able to explain the value of a predictive model by looking at incremental ROI that  is used for targeting purposes.

We have also used  predictive analytics for non-targetting purposes such as models to improve business processes. In these instances, we indirectly impact ROI as we identify specific processes/activities  areas that hinder positive outcomes. Another example was the development of predictive  models that would be used to staff the necessary human resources to service increased demands for services.

Retention Models have also been extensively used by us  where we identify high-value customers that are most likely to become inactive. With this information, marketers can develop programs to better service the high value high risk customers. Yet, our analysis has also found that the marketing impact of retention is highest around the mid-scoring modelled deciles indicating that marketing has no impact in improving retention on the highest-risk customers nor on the lowest scoring customers.  Hence the need  for net lift models.

Q: Can you describe a quantitative result, such as the predictive lift of your model or the ROI of an analytics initiative?

A: It is not unusual, where our models are used for targeting purposes, to observe results of at least 50% lift from the average and if the target group is highly targeted, we typically see 100% improvement. As mentioned before, if our models are used for targeting purposes, we can determine an optimum cutoff  based on a desired ROI level. In our predictive models, we have developed cutoff strategies based on ROI in both the marketing as well in the area of risk. Our risk models have encompassed both credit card risk as well as insurance claim risk.

Q: What surprising discovery or insight have you unearthed in your data?

A: In my experience and what I present at conferences, seminars, as well as teaching at colleges/universities, the data clearly matters more than the math in building solutions. Not that math is unimportant, it is. But the math procedures are readily available for one to use especially in the open-source environment of today. The real value is how the data scientist “works” the data so that he or she develops the optimum inputs that can be used by the algorithms.

For example, just today  we developed a model  using a rating score as the dependant or target variable. The  model  was really weak and clearly sub-optimal despite all the different math routines that we tried.  Upon further investigation, we discovered that the scores were really clustered around two values. So we created a binary variable around these two values where  1 represented the occurrence of either of these two values and 0 for all outcomes. Voila, once we did this simple transformation, our  model performance as evaluated by the AUC curve and deciles charts improved dramatically and equally important makes sense to the business.

Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World.

A: It is my hope that attendees appreciate the importance of data science evolving more towards solving problems and less about the tech. We all know that we need to understand tech as many of us require it in our day to day roles as data scientists. But the world is moving quickly and the tech is becoming more automated and easier to use by larger business audiences. The use of AI is just accelerating this trend. But the real demand will always be for problem solvers as it is these type of jobs that will never be automated. It is my belief that the data science discipline needs to shift  its emphasis towards individuals that can embrace tech but more importantly have those deep lateral mental problem skills which cannot be automated by any machine.

—————————–

Don’t miss Richard’s presentation, Developing Data Science Solutions in Industry/Operations: A not so easy process at PAW Industry 4.0 on Thursday, May 27, 2021 from 11:30 AM to 12:15 PM. Click here to register for attendance.

By: Steven Ramirez, Conference Chair, Predictive Analytics World for Industry 4.0

One thought on “Wise Practitioner – Predictive Analytics Interview Series: Richard Boire at Boire Analytics

  1. Pingback: Wise Practitioner – Predictive Analytics Interview Series: Richard Boire at Boire Analytics « Machine Learning Times – The Predictive Analytics Times – IoT – Internet of Things

Leave a Reply