Q&A: Tom Davenport urges more clarity in data analytics

'The people who are good at analytics tend to be not very good at visual representation of them.'

Businesses may be seeking to compete on analytics, but it's often difficult for business decision-makers to get their heads around data.

Photo credit: TomDavenport.com

I recently had the opportunity to chat with Tom Davenport, visiting professor at Harvard University and co-author of the seminal work Competing on Analytics: The New Science of Winning, about the difficulties of converting to an analytics-driven culture. Davenport, who is also co-founder and research director of the International Institute for Analytics, and a senior advisor to Deloitte Analytics, is working on a new book, discussing on how analytics need to be better communicated to business decision-makers. He shared some of the thinking behind his forthcoming work:

Q: BI and analytics vendors have been coming out with all sorts of graphic tools -- dashboards, balanced scorecards and so on -- for years. Do we need more than a nice splashy presentation on the tool to communicate analytics?

TD: We’ve all grown up on pie charts and bar charts or whatever, but there are probably at least tens, if not hundreds of alternative approaches to visual analytics. Narratives are a pretty good way to convey information in the past, so maybe we should be converting our data and analysis into stories. People are starting to do that more. Most analysts were unfortunately not trained in how you communicate effectively about analytics, so we’ve got a long way to go in terms of doing a better job of that.

Q: More and more data is flowing through enterprises. Is it a challenge to get C-level executives interested in turning this data into analytics?

TD: Not for all applications. Because increasingly people are feeding data into computers and the results go into another computer, and the decisions are getting more automated. Any time you have a human involved, it's important to try to help them extricate the meaning of the data and analysis. And there a variety of ways to do that. Historically, we haven’t been too terribly good at it, the quantitative people among us.

Q: Aren't more analytics going on behind the scenes anyway?  Why is it important to increase understanding?

TD: We saw the challenge in financial services; you have situations like the flash crash, where there were all these automated trading things happening, and we had no idea why. The real challenge is going to be being able to trace the logic and how the algorithms work when things go wrong, so we can intervene and override. But there’s not much doubt both in terms of speed and also costs that we're going to see more and more automated decision-making.

Q: There will be a greater need for communicating analytics at higher levels in the organization, then, versus lower-level decisions which can be automated?

TD: I think that’s true. One CEO I was talking to called it "big swing decisions." And they don’t tend to be automated for a number of reasons. Number one, they don’t tend to be made all that often. They’re not made very often, and they’re really important. You have  to have somebody who's accountable for them. And somebody to point the finger at if things don’t go well.

And I think ego probably gets in the way. Merger and acquisition decisions are usually not very successful. Generally they’d probably do better with an automated decision process. CEOs will say "that’s why we get paid the big bucks, so we have to make that decision."

Q: What about trust in data? Is there a risk in relying too much on data that may not be entirely trustworthy?

TD: I’m always interested in what are the analytical traits of managers and executives, and one is to be somewhat cynical and suspicious. To paraphrase Ronald Reagan, "trust, but verify" your data and the quality of your data. The context matters. For example, a delivery company may move toward real-time dispatching, which relies on "prescriptive analytics," which tells drivers where to go.That creates a pretty high bar for the quality of data.

So data quality is not quite as important if you’re talking about looking backward ar data through a dashboard or something, which is "descriptive analytics." So if you move from descriptive to predictive to prescriptive, the ante gets raised for how accurate the data needs to be, and how much trust you can place in it.

Q: Does this call for a new role to handle analytic communications within organizations? Or should data scientists sharpen their communication skills?

TD: Unfortunately, I think the people who are good at analytics tend to be not very good at visual representation of them. For the most part, you do need a division of labor. They need a visual representation expert who sits down with the data scientist and takes the initially very unappealing output and says, "here's some ways we can make this much more appealing." I think the team-based approach is a lot more likely than finding a single individual that has all the necessary skills.

This post was originally published on Smartplanet.com