Jay Greene interviews Barclays' CIO Richard Sherlund in the Wall Street Journal:
Companies that just had big data have been disappointed that it’s not creating value. But once you can find the correlations in the data through machine learning, you can then begin to have predictive capabilities. Because that now becomes a lead-generation system. Not a more static customer-relationship-management system. I’m now telling a salesperson, “Call that customer.”
As companies scale their use of big data, the move brings a lot of questions. Does it require new architecture? Does it require new platforms? The Wall Street Journal’s Jay Greene discussed the topic with Richard Sherlund, managing director and chairman, software investment banking, at Barclays. The discussion covered everything from going beyond big data that looks to the past to newer applications that will be able to learn and make predictions.
Edited excerpts follow.
MR. GREENE: You have this idea that we are at the precipice of a big change in the way systems architecture is set up and that many CIOs are unprepared.MR. SHERLUND: I think the Internet of Things has enabled industrial companies to start thinking about the need to move to big data.
Because in the past you really couldn’t aggregate big data. It was too expensive. But we have available to us in the cloud now, we can start to collect big data.
The next challenge is what do you do with that. Because companies that just had big data have been disappointed that it’s not creating value. But once you can find the correlations in the data through machine learning, you can then begin to have predictive capabilities.
MR. GREENE: What is making CIOs unprepared for this change?
MR. SHERLUND: I don’t know that CIOs are building the next generation of applications. What they’re doing is building apps where they have some domain expertise. We’re seeing this everywhere.
Data in the airMR. GREENE: What’s fueling this platform are Internet of Things devices, right?
MR. SHERLUND: The new Airbus has 70,000 sensors. None of that data comes down to earth. That stays on the airplane. That has to be processed locally. There are security reasons that you don’t want that data going back and forth.
If you’re running GE’s jet engines, a lot of the correlation’s going to be done in the cloud on Amazon or Microsoft. But some of that data’s going to be processed on the edge [of the data network].
MR. GREENE: You talk about how this data is about systems of prediction essentially. So walk me through a little bit of that. Are we talking about these sensors predicting on an airplane when an engine’s about to fail or when it needs maintenance?
MR. SHERLUND: First of all, you have to have the data scientists to optimize for machine learning. Because for an aircraft engine you may be looking at [performance data such as] temperature, vibration, pressures, whatever.
There was an example of a carrier over in Europe. And they were trying to figure out, “Well, why is the engine wearing out faster than machine-learning predictions?” And it turns out they were short-haul flights. So you’re at lower elevations—wear and tear with more friction.
You have to have some domain expertise to be intelligent as a data scientist in looking at that industry. Every industry has its quirks.
There are probably 100 or more [machine-learning companies] out there. They’re just so small, you know? You’ve got half a dozen or a dozen data scientists that get together, and [the companies are] for sale now at $100 to $200 million. You buy it, and then they leave, you’re kind of hosed.
You wish there were a couple big machine-learning companies out there. As an investment banker, I’d be all over them now because there’s such high demand for them.
The progress at handMR. GREENE: Are there companies that you think are further along in this architecture, are doing it well?
MR. SHERLUND: I think there are industries that are further along. I think the industrial sector is kind of hyperfocused on this right now.
I think health care, there’s just enormous predictive value in the health-care field. You get physicals. And you can do correlations and realize kind of predictive things. Hospitals are huge in terms of the kinds of systems that need to be tied together. So, I mean, virtually every industry will be significantly impacted by this.
One of the things that’s exciting to me is at the highest level, which is the applications level. So if I know when a machine’s going to fail, let’s say it’s a jet engine. You know, a 95% probability in 200 hours that it’s going to fail. If I’m a software company in the business of providing a sales-force automation system, I want that data to be surfaced to the salesperson [so that he or she can attempt to make a sales call to that jet-engine customer about the need to repair or replace the engine].
Because that now becomes a lead-generation system. Not a more static customer-relationship-management system. I’m now telling a salesperson, “Call that customer.”
And because I integrate with the data, I know about the product. I can say, “This is the part. This is how old it is. Should be repaired or replaced. Call the customer now and notify them.”