I’ve always been a big fan of the American television show MythBusters – where scientific methods are used to prove or bust all sorts of myths and urban legends. One of the myths that I’d love to see Adam, Jamie and the team tackle is this: “you can’t do analytics on the mainframe.” While it probably wouldn’t make for a very entertaining show, this is one piece of conventional wisdom in serious need of debunking – so I guess I’m going to have to give it a go myself!
This is the sort of topic that is going to take multiple “episodes” to cover properly, so in this post I’ll begin by simply setting up the business dynamics that are forcing analytics deeply into operational business processes; I’ll save the actual myth busting for my next blog post, where I’ll take on the IT argument that mainframes and business analytics don’t mix.
From my point of view, the most exciting advancement in business technology is the use of analytics techniques to actually optimize the decision making process. One of my favorite thought leaders in this space is James Taylor, CEO of Decision Management Solutions. I highly recommend you read James’ book, Decision Management Systems: A Practical Guide to Using Business Rules and Predictive Analytics.
James has identified four broad areas where decision optimization can have a real impact on both top-line growth and bottom-line savings. Let’s take a quick look:
Managing Risk
Calculating risk is a huge concern in industries that issue credit (what is the likelihood that the applicant will miss a payment in the future?) or insure assets (what is the likelihood of a loss claim on this asset?). When risk analysis is deeply ingrained in your core business processes, you gain the competitive advantage of being able to differentiate risk at a very granular level and offer your clients more personalized pricing models.
Reducing Fraud
Any entity that processes a payment of any kind must be able to protect against fraudulent requests. It’s critical that fraud be detected before the payment is made, because recouping bad payments is both costly and difficult. Pre-payment fraud detection requires sophisticated analytic models that are tightly integrated with payment systems and capable of flagging problems without adversely effecting legitimate payments.
Targeting and Retaining Customers
Every customer interaction is both an opportunity to provide excellent service and a vehicle for improving business results. Deep knowledge of your customers, coupled with systems capable of optimizing huge numbers of micro-decisions, can lead to greater loyalty, more effective campaigns and increased spend.
Focusing Limited Resources Where They Will be Most Effective
If you have constrained physical and/or human resources at your disposal to solve complex logistical problems, optimizing the deployment of these resources is crucial to achieving business results, and a bad decision can be very costly and create satisfaction issues. Think of tasks such as managing cash flow (banking, insurance); ensuring public safety (government); managing airport resources and trucking schedules (travel and transportation); optimizing inventory (retail); and the like. Real-time analytics can help you make the right decisions and react to changing conditions that are outside of your control.
So what does any of this have to do with the mainframe?
Each of these examples represents a class of business optimization opportunity that must be tied directly into the day-to-day execution of standard business processes; this is not the traditional view of performing analytics after the fact to enhance human decision-making. This is real-time analytics performed against live – not warehoused – data.
Where do you run your business operations, and where do you keep your operational data? If you are predominantly a mainframe shop, and the decisions that you want to optimize are based on the processes and data that are on the mainframe, doesn’t it make sense that it would be simpler, easier and more effective to bring the analytics to the data? Why would you not want to do analytics on the mainframe?
Oh yes, your IT people have told you that it’s not possible; that’s why. In true cliffhanger fashion I’m going to stop the program here, let you think a bit about how real-time, operational analytics can help you improve your business results and leave the actual myth busting to Episode II. Stay tuned!
Paul DiMarzio has over 25 years of experience with IBM focused on bringing new and emerging technologies to the mainframe. He is currently part of the System z Growth business line, with specific focus on cross-industry business analytics offerings and the mainframe strategy for the insurance industry. You can reach Paul on Twitter: @PaulD360
To effectively compete in today’s changing world, it is essential that companies leverage innovative technology to differentiate from competitors. Learn how you can do that and more in the Smarter Computing Analyst Paper from Hurwitz and Associates.
Thanks for your post. I look forward to reading your further thoughts on these topics. Will you be borrowing from actuarial science and using some of their stochastic methods? Sometimes multiple risk models may relate, and sometimes they will give different results. How a business makes a decision in the face of competing risk analyses, and what composite risk it ends up taking, is an interesting consideration, both from a decision making perspective, as well as from the actual metrics that composite the individual risk approaches. Fascinating and important – thanks for your work in this area.
Hi Dilip and thanks for your comment! Indeed the techniques you cite are important to the world of business analytics. But this layer is common across all platforms and embedded in the higher-level software offerings (such as SPSS and Cognos products from IBM). The algorithms and techniques used to gain actionable insights are consistent regardless of where they are deployed.
I’m going to be digging into it from a different perspective – sourcing the data that is used for this analysis. The old adage “garbage in, garbage out” has never been more true! As businesses begin to rely on analytics to drive decisions they need to be very cognizant of where the data is coming from, how fresh it is, and how they can keep it safe and unified. This is the level at which infrastructure matters and is where I will be focusing my attention.
The rest, I leave to the rocket scientists
Paul, I’m looking forward to future episodes of the story. But let me share a real-life example with you. From 1989-90, I was one of the leaders of a team that was trying complete the final test for a new generation of IBM mainframe systems. The final test of these systems was a long, arduous affair running longer than a year. We had an extensive database of internal problem reports, many thousands of them, that tracked the workflow of each problem through its lifecycle, including initial discovery, investigation, fix preparation, installation, and verification. As we were part of IBM’s mainframe development laboratory, naturally this database, which was accessed in real time by hundreds of people, ran on an IBM mainframe. (In fact at times it was running on the internal beta version of the new system!).
On a project of this size and importance to IBM, managing the flow of these problems was a critical management issue. So I built a Lotus 1-2-3 worksheet that extracted the data from the problem database and ran an extensive analysis of the problem workflow, highlight the overall trends in problem discovery and resolution and pinpointing the parts of the organization that were running behind schedule in responding. Essentially, what we had was a problem management dashboard with the ability to explore details. And here’s the interesting part: this Lotus 1-2-3 worksheet actually ran on the mainframe itself. That’s right, we had a mainframe version of 1-2-3. Today, of course, we would use a modern descriptive analytics tool from our Cognos family. But back then, Lotus 1-2-3 was the tool. And having the tool run on the same system where the original data resided greatly simplified the task, to the point where I could, from a spreadsheet interface, produce the dashboard. I still have a sample report somewhere as a souvenir from the day.
Thanks for the comment, Ken. This is a great illustration of the theme I’m going to try and build in this series of blogs – that there are great benefits to performing your analysis as close to the root data as possible. You brought your spreadsheet to the data and had positive results; many others would have copied the data down to a workstation and performed the analysis there. Why? A perception that it was cheaper to do so; the analysis tools were not available on the mainframe; etc etc.
Well, all that has changed. Hopefully I’ll be able to provide enough examples and proof points to show that in the current business environment, and with advances that have been made on the mainframe, the best course of action is to do what you did – bring the processing to the data instead of copying the data to some other platform for processing.