Imagine your company’s best onsite service techs. Picture how busy they are rushing from site to site, noting symptoms, prescribing fixes and installing replacement parts. Every day they are adding to a catalog of causes and effects, symptoms and solutions, fringe anomalies and their root causes. Every day they are on the job they become more effective and more valuable.
Now think of your most junior tech. Skilled? Yes. But that technician just doesn’t have the database of experience to pull from, and it will take years to amass the skills of your most valuable techs.
What if you could have that experienced tech at every one of your ATMs 24/7 – 365? What if every time a new symptom and solution popped up the information passed instantly to every single clone of your best technician? And what if the diagnosis of the problem was happening instantly without a service call even happening? Now, one more, what if the knowledge of that best technician was shared across your entire company from service managers making hiring decisions to purchasing teams deciding what replacement parts to stock? This is what Artificial Intelligence is bringing to the ATM market today.
The Future is Here
Voltonix has worked with service departments across all kinds of electromechanical equipment manufacturers for years, and this is a pivotal moment. We are all standing on the line of the foundational shift from reactive to proactive service. That is, Artificial Intelligence is changing the world around us. It’s not changing only video games and chess matches but also the way we do work, and the way enterprises make decisions.
AI is an elastic term and means different things to different people. Some more well known examples include Log File or Call Analysis, IoT Data, Digital Twins and Rich Sensor Data. It is the last one – Rich Sensor Data – that is both cost effective and transformational to First Line Service Organizations. The data provides real-time actionable insights with anomaly detections, and it learns and understands what is important as time goes on. It is also retrofittable and allows for cross platform implementation.
Voltonix is partnering with Sigsense Technologies on a wall-to-wall AI branch solution that uses Rich Sensor Data and requires only one sensor between the wall and the plug for each ATM, Cash Recycler, etc. An FLS partner recently told Voltonix:
“This is fundamentally going to change the DNA of our business.”
The solution eliminates mistakes in the diagnostic process and dramatically shortens it. Technicians would have the right part to fix the problem the first time on site. Real first call resolution is within reach.
Getting Specific with the NCR SDM
Inside the ATM ecosystem the NCR SDM cartridge serves as a specific use case and illustrates the agility of the solution. The SDM cartridge holds three bands that peel currency from the deposit. It relies on friction and tension and is generally a filthy component that can often jam bringing the ATM down. So, how does a remote computer with the assistance of only one external sensor alert a service organization to a jam in that specific section of the ATM? Or better yet, how can we tell that the cartridge is wearing before it actually fails and causes downtime?
The sensor provides a window into the machine’s entire electronic power draw. Through this we can see exactly which components are being activated and the sequence in which they occur. Sigsense’s Deep Learning algorithm takes that Rich Sensor Data and understands what a good deposit looks like versus a failed one. It is intelligent enough to correlate the jam signal with previous observed signals to alert technicians to the jam’s exact location. Rather than a technician responding to a general “deposit jam service call” the tech is dispatched to address the exact problem the first time out.
Through cumbersome log file analysis some Service Orgs perhaps can tell where the jam is, but the Deep Learning differentiator comes via correlations over time. The Algorithm understands and tracks when the SDM fails to strip the bill two times but then succeeds; maybe tomorrow the deposits fails three times before a successful sort.
In a log file system these ultimately successful deposits probably do not generate alerts. But with a Deep Learning application we can tell that the unit is entering a failure state by the increase in the frequency of almost failed deposits. That is, a non-catastrophic event can be flagged as an indicator of a future failure and thus a service call. Pattern matching over time empowers the unit to make predictive assumptions based on previous observations.
AI On the NCR S2
The same observations are simultaneously done with an S2 dispenser. Failed picks are easy detections as are the alerts, but the AI digs deeper. Typically, the S2 does not suddenly just stop pulling bills; it goes through iterations of increased rejects over time. Maybe it is a vacuum leak, or maybe it is bad currency. Each potential scenario looks slightly different to the algorithm. These observations are pattern matched to deliver prescriptive alerts for preventative maintenance. With enough data, we can even estimate time to failure.
The good news is to roll this out you do not need deep OEM pockets or specialized humans. In many cases, you do not even need a year’s worth of carefully organized log data. Deep learning, sensor-based implementations learn in real time, and the return on investment scales with the amount of data observed.
A number of companies are patching data solutions together to try to get ahead of the curve, but they are often cost prohibitive and lack the granularity and nuance that deep learning and Rich Sensor Data bring. Your data should be working for you already, and with this approach, it will work for your entire organization and not only your most valuable service technician.