Taking a chance on smaller sessions at World of Watson 2016

Independent Consultant

I discovered three excellent sessions today, each of which revealed insights into ways of building models that offer tremendous return on investment. IBM Insight at World of Watson 2016 is bigger than last year’s conference, offering even more sessions. More than once, in fact, I’ve been turned away at the door from an already filled session. So I have a new game plan. I’ve moved overviews of popular technologies to the back burner—after all, I can learn about these online, including by using resources such as Big Data University. The truly big sessions I can even locate on IBM GO for later viewing. Instead, I’m moving client case studies and other smaller sessions to the front of the line, sometimes even dropping in on sessions that I haven’t researched in advance.

Modeling success from utility poles to toner

The first session that impressed me today was a high-level introduction to the Internet of Things (IoT) and weather data a year after IBM’s acquisition of The Weather Company. As with all the talks I attended today, its subject was modeling in all its seemingly mundane operational details. But done well, predictive modeling reimagines those reputedly pedestrian decisions in ways that can be worth millions—even tens of millions—of dollars.

In this session, Joe Sullivan of The Weather Company (now an IBM company) and Jack Lynch of IBM described how IBM uses weather data to assist clients in the utilities space. Those attending for the technology details heard plenty of them, but I was all ears for the modeling strategy—the description of how data science detectives solve real-world problems. One example in particular fascinated me: If you want to proactively tackle a million utility poles and a $25 million capital budget, then you should keep your eye on those utility poles that have been exposed to freezing temperatures and high winds. Because these are your more vulnerable assets, you’ll need to adjust their placement in the maintenance queue.

The second talk described ways of saving money spent on copy toner. That might not sound sexy, but a savings of $750 million certainly does—and that’s the amount wasted each year, industrywide, on toner. With that figure nailing us to our seats, Scott Hornbuckle of the Photizogroup dove into why the waste occurred, then laid out a plan for tackling the problem.

When consulting, I’m forever telling clients that insights don’t help if you don’t both deploy a model and also build an intervention strategy around that deployment—and Scott provided a great example of just this. Persuasion and education won’t always change human behavior—and they frequently don’t. Rather, the solution was as simple as carefully scheduling toner to arrive just in time, preempting the tendency toward premature replacement that costs some large corporate clients as much as $20 million in wasted toner.

Reimagining business with predictive analytics third talk featured a team from QueBIT, an award-winning IBM Partner that had developed a model together with NAPA. Having been a part of the QueBIT analytics team some years ago, I eagerly anticipated this session, which described how the team helped NAPA stock an optimal mix of replacement car parts. Such an initiative might seem mundane at first, but big money was at stake.

I’m always curious about the hard work that goes into science, and this was another great example, for much of that work took the form of data preparation. What’s more, in addition to drawing on NAPA’s internal data, the QueBIT team looked farther afield to climate data and data about local car ownership—and all that data also had to be prepped. During this session, James Saronson and the QueBIT team described how the team created a model that built unique parts assortments involving 6,000 stores and 500,000 SKUs. Those 3 billion little weekly predictions ultimately enhanced accuracy by more than 37 percent, drawing on a 2.7 TB data mart.

But data doesn’t stand still, so each week, the model is rebuilt. About 80 percent of the time required for this lengthy process is ETL-related; only about 20 percent involves modeling algorithms. As our session continued, Grant Remington of NAPA Automotive Parts Group NA described the results of the implementation: After an initial 61-store rollout, the positive feedback from within NAPA was immediate. One manager reported a 25 percent sales increase during the first month. There had been no new customers, no change in product line and no change in the team—the only difference was the presence of the model.

If you’ll be attending IBM Insight at World of Watson 2016, then be sure to check in with the IBM Big Data & Analytics Hub for daily blogging about IBM Insight at World of Watson 2016, and remember to follow @KMcCormickBlog for live tweeting from Las Vegas. Also, to learn more about predictive analytics and statistical analysis, be sure to check out our informational IBM Analytics resource page. If that piques your interest, then be sure to check out my books, the IBM SPSS Modeler Cookbook and SPSS Statistics for Dummies, for more information about big data analytics.

Stay tuned for more from IBM Insight at World of Watson 2016. I hope to see you here!