Through a series of exploratory machine-learning collaborations with Microsoft, we’re reimagining fundamental business processes and streamlining traditional pain points.
Katherine Lin is, to put it simply and in the vernacular of her native Boston, wicked smart. A Harvard-trained data scientist and start-up founder who now works for Microsoft, Lin is focused on efforts to develop cloud machine learning services and tools that can help organizations transform data into business intelligence with Cortana Analytics Suite. Our data analytics and new technology incubation team engaged in collaborative conversations with her on a recent project that tapped into her passion for applying machine learning to new fields of study.
We’re working to crack the code on a retail banking conundrum: How can we harness all the information flowing from hundreds or thousands of ATMs in a vast, sometimes international financial organization’s network? And how can we act on that Big Data to reduce failures, cut maintenance costs, monitor (and perhaps improve) ATM health and even access a more holistic view of consumer behavior? Through a customized approach that pulls clear, actionable insights from complex retail banking data.
An ATM Health Score
When’s the last time you visited an ATM for fun?
Diebold employees notwithstanding, most of us only use an ATM when we need something: we have to withdraw cash, make a deposit or get a quick read on our accounts. So when we get there, we expect the machine to work without fail.
Our engineers are working on solutions that deliver optimal uptime, and we’re always driving toward 100% availability. That means monitoring and predicting failures before they occur, so we can preemptively address the problem. We build this “health score” by harnessing the power of machine learning algorithms, through the Cortana Analytics Suite, Azure Machine Learning and the power of the cloud.
How to Spot an Algorithm
We’ve been using algorithms since we were kids. Algorithms put the guiderails on number-crunching, in a way. They’re defined methods of solving problems, used in everything from multiplication to computer science. Today, data scientists can actually “train” machine-learning algorithms to understand data sets, discover patterns and identify areas of interest that no one ever noticed before.
Katherine works with our team to apply these machine-learning algorithms to incredibly complex, rich data gathered from modules and sensors on each self-service terminal in a network. Consider this: We can gather more than 250 data points from a single module on an ATM — and that ATM might have 10-15 modules operating at a given time. That means there are potentially thousands of data points we can pull into advanced algorithms to create real-time fleet-visualization tools for financial institutions.
No human could ever crunch all those numbers and find meaningful patterns or offer actionable, predictive analytics. But a computer can — if we teach it the right algorithms. These machine-learning programs literally evolve over time. As they receive more and more data, they get better and better at making predictions. And eventually, they can craft an ATM health score that predicts and explains some interesting things:
- For example, the probability that an ATM will fail in the next 1,000 note transfers.
- Or the reason why 9% of a bank’s service calls result in a “no problem found” resolution description.
- Or the overall health score rankings for a network of thousands of ATMs spread across the country.
Big Data in Action
Our work with Katherine and the Cortana Analytics Suite aggregates ATM data in a way that few other organizations can do. Because Diebold is a services, software, hardware and security provider, we have access to a much larger picture of a retail bank’s networks than almost anyone else. Paired with the storage and compute capabilities provided by Microsoft, we’re positioned to apply health scores to networks at a global level. Today, we’re in the pilot stage of implementation with three major banks internationally, working to understand their ATM networks at a deeper level.
Our collaboration with Microsoft is just one way we’re working on new solutions that analyze and shape retail banking data. It’s through multi-partner collaborations like these that we can figure out the right questions to ask, in ways that are fundamentally different than they’ve ever been asked before.
Interested in finding out more about how machine learning and Diebold subject matter expertise could help rewrite the business rules and drive efficiencies at your organization? Let’s talk.