AI: the 4th Generation of Process Automation in ITSM (Part 2)

Over the next few days we will be publishing a series of short articles written by our VP of Products, Peter Schneider. This is Part 2 of the series, we will be adding links to the bottom of the page as the other parts are released.

Predictive Maintenance

Predictive maintenance is the art of learning when a component is about to fail. Any failure correctly predicted and corrected before it occurs means potentially hundreds of less alarms, tens of less incidents, and at least one fewer emergency change in IT Operations Management. Predictive maintenance is superior compared to preventive maintenance because it actually learns the average, expected failure time based on real usage and environmental factors. Costs can be saved compared to preventive maintenance due to utilization of equipment for its true expected lifetime and nothing shorter. Predictive maintenance is in particular beneficial for IT assets with mechanical parts such as disk drives or electrochemical components such as batteries, which are likely to be the focus for innovations in the near future.

Machine-learning is increasingly applied for predictive maintenance in order to estimate from historic records of similar equipment in similar conditions when a failure is likely to occur. Current machine learning algorithms applied for IT equipment such as infrastructure elements and workstations are still of trivial nature using a limited set of input sensors, however improvements in IoT enable better learning mechanism with a wider measurement of external factors over the next years to come.

Like predictive maintenance for elevators, electrical motor drives, and ship engines, predictive maintenance for IT assets such as IT infrastructure elements and workstations is likely to be implemented most successfully by the manufacturers of these components. Access to sensors assisting the predictive maintenance and the ability to combine such in a cost-efficient manufacturing process will help predictive maintenance to become mainstream business. It is likely that predictive maintenance values of IT infrastructure components can be monitored from central tools such as Prometheus or PRTG in the future. Discovery tools already in use for device and software asset management are good candidates for predictive maintenance of workstations.

Event correlation tools may provide a second level of predictive maintenance analysis computing the vast amount of utilization and sensor information provided from IT assets, especially for such elements without much own computing or memory capacity to perform predictive maintenance locally, in order to predict failures.

ITSM tools serve as single point of record for all warnings from IT assets, monitoring tools, and event correlation tools. ITSM tools will also provide supportive, third-level predictive maintenance by correlating multiple events through machine learning algorithms analyzing events from multiple input channels based on historical data of similar combinations of events.

Identity Theft Protection

Artificial Intelligence can also be used to protect enterprise data and prevent identity theft. Already few months of historical data will be enough to profile typical behavior of users. Once unusual behavior is recognized, corrective measures can be triggered.

Every human being has some routines that one follows at the workplace. While some people might have a high degree of doing things always in the same order, at the same time, and from the same location, other’s behavior might be less regular. But even that can be measured over time and used to identify that the user is likely to be the one she or he claims to be.

There are two major applications in user behavior profiling for ITSM tools:

  • AI-enriched user profiling for multi-factor authentication
  • AI-based user profiling of typical ITSM tool usage

AI-enriched user profiling applies machine learning algorithms to identity during the sign-on procedure to the ITSM tool whether time, location, device, and browser are representative of the typical behavioral score of a user. A support person might typically sign-in from a fixed workstation from the location of the office around 08.00 in the morning. Now, if that same user signs in with another device from another location at 03.00 in the night, something might be wrong. The multi-factor authentication might prevent access to the ITSM tool until the morning, notify the 24/7 service desk, or enforce a two-factor authentication based on the user’s mobile phone. Because the profiling is dynamic and continuous learning, it will learn if the behavior changes due to changes in the user’s lifestyle, job change, or relocation. A positive event such a successful two-factor authentication might trigger that the ITSM learns a new pattern for the user.

Even more powerful is identity theft recognition based on ITSM tool usage pattern. This method is harder to imitate, however also more complicated to implement because it needs to have access to all actions of the user in the recent past. In the same way that ITSM tools will be able to predict the next action the user is likely to take they will also be able to recognize when users display unusual pattern for some time. Let’s assume an administrator typically creates new reports and modifies the attributes of the solution during daytime. Let’s further assume the same user is starting to delete large amounts of data during night time. The system might recognize such a change in behavior and starts taking measures to prevent damage by temporarily restricting access rights or creating automatically a security incident.

While profiling is generally a sensitive topic to many users of software applications, especially due to privacy as well due to illegal performance measurement concerns, under the umbrella of enterprise security using anonymized data, it is likely that we will see more and more implementations of such functionality in the next decade.

We will also be hosting a webinar on Tuesday the 27th of November, where Peter will be discussing the topic of AI with one of our key partners, You can find out more about the webinar and register from the link below.

 View the webinar recording

Other articles in this series

Overview, Part 1, Part 3, Part 4

Peter Schneider

Written by Peter Schneider

I am Chief Product Officer @Efecte. Responsible for product management, product marketing and product strategy.