Data analytics: start in the middle

Quick insights can be gleaned from incomplete data

Published In: Intelligent Utility March / April 2012

Share/Save  

LLOYD TOKERUD, SENIOR MANAGER OF ANALYTICS FOR First Choice Power, a competitive retailer in Texas, was hired from PepsiCo to lead the utility's analytics group. I spoke with him recently about issues in data analytics.

First Choice Power has been in business since January 2002, when deregulation in Texas took effect. The retailer was acquired by Direct Energy last November.

Perhaps Tokerud's most intriguing role is to work with "business clients" across the organization to understand their challenges and to develop data reporting and analytics to swiftly address those challenges.

Data challenges
"The unofficial goal of our team," Tokerud told me, "is to allow the organization to make decisions faster than our competition."

I asked for examples of the data challenges for a competitive energy retailer in Texas.

"The biggest question we asked first was around customer profitability," Tokerud said, "and trying to understand which of our customers, among our different enrollment processes, were more profitable than others. We bring customers into the organization in different ways, some are more costly, some less costly. Some have higher, ongoing operational costs. So we wanted to understand who in our customer base is really returning the best profit to us so we could better understand and market to those people."

An article, "Big Data, Analytics and the Path from Insights to Value," by Steve LaValle et al., published in the MIT Sloan Management Review (Winter 2011), suggests focusing first on a single major question to apply analytics for early results and, thus, organizational acceptance and change.

 "One of the functions of a good analytics team is the ability to work with the executive team to understand and clarify what exactly is `at task,'" Tokerud told me. "More importantly, you need to eliminate a lot of `noise' in order to get insights quickly.

Steps to quick insights
"The first step is to define what you're after and what you're not after," Tokerud said. "Which groups of customers are most relevant? Which ones do we not want to worry about for now? What's the correct equation for profitability? Answer those questions first.

"The second step is to assess the current data `landscape' or sources so that I understand what I'm being asked to measure and deliver insights on," Tokerud continued. "What components of that are readily available now? And what are the tangled messes of information that we may never be able to get an answer on? Then draft a framework that will get you where you need to be quickly, with an acceptable level of error.

"One of the points in that MIT article is to get to a workable solution as fast as possible, instead of getting hung up on defining absolute perfection before you can start reporting anything," Tokerud said. "You may be missing key pieces of data. You have to make assumptions, fill that information in and keep rolling. The concept is that you `start in the middle.' There's an 80/20 rule. You can get most of the way there. Don't get hung up on getting 99 percent accuracy for that data set."

The third step, Tokerud continued, is that once you have preliminary insights, socialize that information back to your audience. Share your findings with your in-house client to determine whether you're headed in the right direction. This helps you, but also more importantly gets directional results back to the leadership team which may influence decisions or change behavior.

The final step is sharpening your findings and putting insights into action so they become a new metric or key performance indicator to manage against. That completes the cycle and enables real-time management of information.

Centralizing analytics
Tokerud said that data analytics are likely to become a centralized function. A centralized team enables uniformity, puts everyone on the same page and produces the proverbial "single version of the truth."

"The challenge with a centralized model is that you're `it,'" Tokerud said. "Suddenly, everyone is coming to you to answer some basic questions about their bailiwick. It's time consuming. That makes how you approach your task more critical. You need to prioritize, tackle the most valuable questions first."

Thus, Tokerud has established two prioritization streams: one is a queue of requests from each business unit, which is dealt with on certain days of the week; the second is a separate queue from the executive leadership, addressed on other days of the week.

Analytics, in Tokerud's view, are heading toward real-time "decisioning."

"That's the stuff that companies need," he said, "because if you can answer stuff on the fly, then you can move in the marketplace faster than your competition."

Are analytics likely to be handled in-house or outsourced?

"In a lot of ways, First Choice Power follows an outsourcing model," Tokerud said. "One company runs our call center, another runs our billing system. We've contracted with an outside company to develop analytical tools for us.

"But the core competency of analytics should not be outsourced," he emphasized. "When it's done right it's more of a partnership and organizational ability. You want analysts who can talk to the business users, understand their problems and meet them halfway on how data can help answer those questions."

Defining the limits
Finally, as data analytics become the newly revered tool, what are its limits?

"It's more a question of pitfalls," he said. "There is a startling number of ways in which analytics can be less effective than it could be.

"Your approach, how you set up things organizationally, is really foundational to the success of your analytics efforts," he continued. "A common pitfall is to approach it as a reporting exercise, instead of an analytical exercise. If you have it roll up to a functional department that's misaligned with the objectives of the leadership, that's another potential problem. If you don't have the right sponsors or participants, that can be a problem."

The ideal data analyst would combine roughly equal competencies in business and IT.

"We need people who understand both sides and can communicate with leadership," Tokerud concluded. "How it's set up and who's brought in to run it are important decisions."

 

Related Topics