by CIO Executive Council

3 CIOs Reveal How They Got Started With Predictive Analytics

Nov 18, 20134 mins
Big DataBusiness IntelligenceCIO

Want to s쳮d with predictive analytics? These CIOs say it takes a lot of front-end data work and angst about cultural change.

If you want to learn how to succeed with predictive analytics at your business, these three CIOs offer hands-on adivce. In short, they say it takes a lot of front-end data work and be prepared for angst about cultural change.

Expect Culture Shock

Chris Coye, Senior Vice President & CIO, Disney ABC Television Group: We’ve implemented three predictive analytics tools this year: One analyzes what-if ad sales scenarios, another is a promotional media-optimization tool, and a third will help our executives decide which pilots to pick up. We created a small data analytics team in IT, but the models are built by Disney’s revenue sciences group.

The biggest technical challenge was getting the right source data. We have multiple divisions, and that data had to be standardized. We built our own extract, transform and load tool, but we’re migrating to a commercial tool to speed the process.

Culturally, these tools have caused a lot of angst. Research doesn’t want sales to see its data too early; sales doesn’t want finance to see its data too early. Information is now available earlier than people are comfortable with; everyone wants to maintain control over the narrative describing their results.

It’s a big change driven by our CFO and CTO, with the expectation that these tools will enable better decisions. Determining whether to sell an ad or use that time slot for a show promo used to be based on gut feel; now the tool predicts what will drive more revenue–selling that ad or getting more viewers to watch tomorrow night’s episode of Revenge.

Start Small and Build Enthusiasm

James Clent, CIO, United Orthopedic Group: Our market–orthopedic braces–is a limited one, so customer retention is key. Our first attempt at predictive analytics was to identify our silent defectors: customers who don’t say they’re unhappy but whose orders trickle down or stop altogether. We hypothesized that we could rescue those customers if we could identify them before they left. We built a tool that, based on order patterns, could predict when a customer’s next order ought to occur. If that time came and went, a salesperson would give them a call.

We started with a pilot, which reduced silent defections by 50 percent, and used what we learned to deploy the system nationally. During the pilot, we developed customer trend lines that allowed sales staff to see order volumes and deviations in addition to next-order predictions, but it was too complex.

It’s important to keep it simple, get the organization used to working with predictive analysis, and build enthusiasm over time. The easily overwhelmed will get on board, and the advanced users will ask for more. Then you can have follow-up projects, like the sales budgeting and analysis tool we’re building. It becomes more than a fad; it’s something that lasts.

It’s Not About the Technology

Ed Brandman, CIO, Kohlberg Kravis Roberts: Predictive analytics is one more tool that helps us in our rigorous process for making private equity investment decisions and evaluating the performance of companies we own. We spent a lot of time figuring out how to collect all the data we need. We get financial and operational data every month from our 80 portfolio companies, each of which captures its data a bit differently. It’s a massive amount of information that can provide a lot of predictive insight ahead of what is publicly available–if you can normalize it. After two years, we have some pretty solid data. It took a lot of legwork, and we’re still constantly tweaking it.

We overlay that with publicly available trend data from third parties like Standard and Poor’s or BCA Research. We can look across our companies to understand which debt is fixed versus floating rate and, based on larger interest rate trends, find the most opportune moments for financing, for example. Our portfolio management committee can determine whether to be more or less aggressive about expanding a business or managing costs.

We’ve developed the tools ourselves. We’re not looking at massive time series data that would take us down a Hadoop path. Too many companies get caught up in all this new technology, but it’s not about the technology. More data–and more frequent data–is not always better. It’s about getting the right data at the right time. Unless you can turn that data into an actionable event, it’s a waste of time and money.

Follow everything from on Twitter @CIOonline, Facebook, Google + and LinkedIn.