Data analytics guru Milena Arsova travels the globe championing ATM predictive maintenance
If you’re looking for your daily dose of inspiration, meet Milena Arsova, technology and innovation leader with IBM Technology Support Services.
Based in Bulgaria, Arsova is part of the team that started predictive maintenance at IBM. she travels across continents, advocating a data-based approach for banking technical support. Why? She’s personally invested in solving big ATM availability problems for banks through predictive maintenance, machine learning and insights.
What’s your personal connection to predictive maintenance?
I started my career in banking, working with statisticians and IT personnel. Then, in IT consulting, I built statistical models for credit risk management. At IBM, I held an analyst role before moving into management and tech leadership. Even in that role, I promoted the idea of machine learning and infusing insight into banking technology support processes.
ATM services support seemed like the perfect playing field because there’s a high volume of failures, there’s monitoring solutions for data collection and availability, and there’s a big problem to be solved — availability challenges. It feels like my own little startup within the corporation because I feel a personal connection to the idea; there’s satisfaction from having that idea realized and worked on, and even more so from it generating the benefits we expected and then some.
How have clients used predictive analytics in the last year, and what might they expect looking ahead in 2019?
There is a standard in the industry, which is generally reactive support, plus schedule-based preventative activities based on manufacturer recommendations or deterministic rules of thumb.
We’re saying this is great, but you can do better and here’s how.
Clients are using our support structures and predictive output. We pull the data through the cognitive engine and generate sets of predictions and a list of alerts about which ATMS are going to be at risk of failure. Our support people use the list to coordinate activities such as staffing, logistics and SLA exposure in advance, so we can smooth out our workload for efficiency and provide a competitive price back to our clients. At the same time, we’re not allowing uncontrolled outages because we’re capturing these failures in advance, and helping to improve the availability of installed bases.
Where does AI fit in?
[In this context], cognitive is a compilation of deep learning algorithms deployed for a particular purpose, and an implemented feedback loop that allows for continuous system learning. This is essentially how we built our infrastructure. Usually these solutions start with more rudimentary approaches, then we add complexity as data volumes increase and the problems we’re solving become multifaceted.
AI is more about simulating human interaction, so in this context it’s not something we’re targeting. For us, it’s more about getting to valuable insights well in advance.
What’s the net-net for clients exploring predictive maintenance?
For banks, the interesting part we’re proposing is a brand new model of proactive maintenance through technical support. We have the evidence and validation of our assumptions that this new data-driven and insight-driven methodology unlocks new opportunities for availability control.
From a model training perspective, we’ve been creating models and testing assumptions on more than 60,000 devices worldwide. We helped a client improve availability within the first two months of deployment, and helped another reduce service-related downtime by more than 3%.
For those considering a new support operational model to address availability challenges, I ask: Does it take longer for the bank to fix an ATM or for a customer to open an account at a new bank?