Big Data in the Operational World Requires a Different Mind-Set: A Spotlight Q&A with James Taylor
Originally published November 25, 2013
This BeyeNETWORK spotlight features Claudia Imhoff’s interview with James Taylor, CEO of Decision Management Solutions. James explains that the approach to operational analytics in these days of big data requires an approach that starts with the business problem.Right now you’re really focused on operational BI and operational analytics, and I love that topic. But it has changed recently because now we have big data in our operational world. How do you apply big data to operational decision making?
James Taylor: The first thing to remember is that there is a tendency to think that if I’m going to do big data analytics, I have to buy new hardware and do all this new stuff—and therefore I have to apply it to big strategic questions: How do I see my business in the next 20 years? What are the questions I can’t think of? Data exploration, which I sometimes jokingly refer to as “letting my data scientists cavort through fields of data.” Instead ask “What are my current operational problems and how could I use big data to solve them?”
The first exercise in using big data in operational analytics is to actually decide to do so because it is those operational decisions, by and large, that generate big data. It’s the operational interactions or the high-volume interactions you have with customers on your website, with devices or with call centers. The things you do often is where big data comes from. So it seems to make perfect sense to say if it’s where I’m generating the data, why wouldn’t I use the data to improve the way I do things at the point where I generate that data?
It’s also a little bit of a challenge because if you’re looking at operational analytics, you’re often looking at predictive analytics. You’re looking at embedded analytics. A lot of the big data infrastructure tends to be more focused on reporting, querying, visualization and some of the analytics that are perhaps less suited to this operational environment. So it’s also about looking at your technology platform and saying, “Do I have an environment that will let me bring these big data sources and my structured data into the same place, and be able to build analytics against them both.” And say, “I have an analytic, and I have a way of doing this operational decision. Is there big data that would help me refine that, make it more accurate and enhance the predictive power.” I think that’s a little bit of a different mind-set about using big data than the trendy view of it which is to buy a separate big data architecture and let people go off and play and figure out how we can stop being a paper goods manufacturer and be the next Google. I’d rather have them focus on how they can get better at what they do today.
I agree with you. One of the things, though, when people think operational analytics is that they think customer analytics because that is the lifeblood of most companies. Their questions are typically: Who are my customers, what should I be selling them right now, how are they behaving and so on. How do you see this whole big data operational analytics stuff affecting the traditional CRM mantra of the 360-degree view of the customer?
James Taylor: Customer analytics is clearly the primary use case for advanced analytics. I’ve done some surveys, and it comes out again and again. Focusing on your current customers is what really counts. The thing about big data – particularly once you take big data, the new data sources that companies have inside, the new data sources they can buy, and the increasing prevalence of data in the cloud that’s available to enhance your data – is that you have to stop thinking about the 360-degree view. That’s because if you don’t do anything until you have a 360-degree view, you’ll never do anything because there’s always another data source. There’s always more data, and you can get so bogged down thinking that you haven’t finished integrating all your data sources and that you haven’t got all of your data cleaned up.
I don’t have my ducks in a row so I can’t move.
James Taylor: Exactly. And you’re never going to have all your ducks in a row because there is always a new duck. You have to start thinking about what problems you’re trying to solve. For example, if I need to increase the number of customers who have more than one of my products because customers with two products are more profitable and easier to retain, how do I do that? Well, I have to get my call centers to do a better job of cross-selling a second product when people call. That’s what I have to improve.
Starting with that problem, I can work backwards to determine what kind of analytic would help them do that? If that’s the type of analytic I need, what data might help me do that? Now that data might be inside my firewall, but it might not be. It might be data I own, it might be data I can buy, or it might be data that’s only available in an audio stream or video stream or in unstructured text. Then I have a big data problem, but I have it to solve a particular business problem. I think you have to change that mind-set. There are far too many companies that are bogged down trying to integrate, clean and organize all the data before they do anything with it. The problem with this is in the world of big data, you are never, ever going to finish.
So the idea is to get something going and focus on the business problem.
James Taylor: Absolutely. I like to misquote Stephen Covey and say, “You have to begin with the decision in mind.” What decision am I trying to influence? What analytic would help me influence that in a positive direction. Of course, you have to know what a better decision looks like – that’s a separate problem. But then you work backwards and figure out what data you might be able to use. It also tends to make you think outside your warehouse a little bit more. If I start with my data warehouse I tend to think I’ve got this data, how can I analyze it? What can I do with it? But if you start with the problem and start pushing back, you might say, “Gosh, if I just knew xyz, that would really help. None of my data is going to tell me that, but you know there’s this service that sells data – and if I had that data, maybe I would be able to find out.”
Let’s drill into that a little bit more because obviously what you’re talking about does have an impact on the traditional BI architectures. What do you see going on there?
James Taylor: To really focus on these operational analytics, you’re typically pushing them way out to the very front lines of your organization so that means automated environments, mobile apps, websites and self-service applications – and to people who are never going to be effective consumers of BI tools. They work in the call center or they work in the branch or they drive the trucks. To get analytics to improve their behavior, it has to be embedded in their systems. So that means you have to move from reporting and visualization to data mining and predictive analytics – things that can be turned into algorithms, real pieces of analytic insight that you can embed in a system so it becomes an analytic system. Its behavior is analytic. That’s a big stretch for a lot of companies. They have to move up that curve. And even companies that do BI and advanced analytics, they often have one group that does BI in IT, and then the business has a group of people doing predictive analytics. And they have to find ways to share more infrastructure and resolve some of these differences. Because, in the end, it’s a spectrum that they need to support.
The other thing I think it does is put pressure on what you think of as the data you need to manage. Because if you talk to predictive analytics people – data scientists and data miners – they want the history. They want to know when values changed. They want to be able to collapse things on the timeline so instead of it being an absolute date, it’s the number of days before you quit as a customer, normalized for every customer who ever quit. All that temporal data management and not overwriting fields with new values so that you can see what the value was on a certain date – all of that ripples through into the warehouse and the infrastructure because if the warehouse only has summary data, if it only has current data, if it doesn’t have the history, if it doesn’t have that temporal axis, if everything has been summarized by day or by week, then the analytics team won’t use it. They’ll go suck call detail records out of the switches and out of the log files.
They’ll go off on their own.
James Taylor: Yes, they’ll go off and do their own thing. And all that data quality work you did, and all that clean-up work you did will be essentially subverted because you’ve cleaned it too much for them. So I think it’s forcing people to say, “I have to think about what it means to have a data warehouse.” It’s not a summary anymore. It’s a collection of history that I have to make sure is visible, not only so I can run nice reports against it to show my business owners, but also so I can drive deep predictive analytics models from the same source. Otherwise, it’s just never going to scale to the hundreds and thousands of predictive models that companies are going to need in the future.
What is the number one thing that someone should do to get started in this?
James Taylor: It’s to look at key metrics that they have as a business and say, “What day-to-day decisions that we make hundreds or thousands or millions of times affect these metrics.” Those decisions are not made by executives. They’re not made by managers. They’re not made by analysts. They’re made by people in the call center. They’re made by the website. They’re made by the automated environment. Focus on those and determine what could be done analytically to do a better job with those decisions because so many of them are being made that they don’t have to be all that much better to start making a big difference.
That’s a great suggestion.
James Taylor: We have a bank we work with. They’re a big bank, but they’re not a mega bank. They add up all their customer transactions through online banking, the website, the branches and the ATM. They have 140 million interactions a month. How much better do you need to make each interaction – let’s say cross-selling – before that starts to add up to be a lot of money? Well, not very much. So you don’t need hugely sophisticated analytics. You don’t need the data to be perfect because you get this huge multiplicative effect because you’re focused on these decisions. That’s one of the big things I like to say - don’t think that because analytics is expensive, that big data is hard, that you have to apply it to big strategic decisions. Apply it instead to the decisions you make all the time.
Yes, and every one of them can be made a little better and a little faster, and it adds up.
James Taylor: It adds up fast.
James, thank you so much for providing this insight into operational analytics.
Recent articles by Claudia Imhoff
Copyright 2004 — 2020. Powell Media, LLC. All rights reserved.
BeyeNETWORK™ is a trademark of Powell Media, LLC