Fixing Things Faster — How Machine Learning can help automate
and speed up decision making and meet SLA’s.
The Magic Bullet: Machine Learning?
Before beginning, I should first make clear that Machine Learning is no magic bullet. It is not some mystical powder you can sprinkle on to your business’ challenges and watch in delight as they disappear before your very eyes. On the contrary, it’s actually very important to understand that Machine Learning, while versatile and wide ranging, is something that still requires skill and a logical approach to do well. At Somerford, we like to avoid hype and the use of buzzwords, especially when it comes to Machine Learning. Instead, we favour being honest, straightforward, and professional. So there will be no hype nor hyperbole in this tale of applied Machine Learning…
Now, without further ado, let us begin the story of where one careful application of machine learning techniques was able to simplify and streamline ticket management, helping to ensure SLA breaches were reduced, and costs reduced with them.
Customer SLA Problem
The customer in question approached Somerford with a query. They used a ticketing system for their customers to raise tickets whenever they were experiencing an issue. For each of their customers, there was an SLA that they were beholden to. There was, however, one small problem. The ticket submission system required the end user to classify what they thought the category of the problem was. Most of the time, this would be done accurately. Though every now and then, a customer would incorrectly label a support ticket. This meant that while the issue may have been something that sat within the remit of the Windows Active Directory team to rectify, instead it was incorrectly sent to the Network team. As you will no doubt know, when it comes to SLAs the clock starts ticking as soon as that ticket is first submitted. This mislabelling of tickets meant that from time to time, tickets were not getting to the correct team fast enough for resolution within the SLA. This was not good. It was a cost to the business. So how do you solve something like that? You could hire additional staff to manually read and assign tickets to departments, but that’s slow, manual, labour intensive, and expensive. Businesses need to move faster, to react quicker, and to do so at lower cost.
The Machine Learning Solution
Enter, stage left Machine Learning. Somerford’s customer had already begun to explore a Machine Learning solution to identify “misclassified” tickets, however they had faced difficulties in so much as it was a complicated solution using Python and Tensorflow, and it proved less than simple to maintain. The training burden to ensure enough competent and experienced developers were available to maintain the solution was higher than was ideal. Instead, they wanted a simpler solution. It was fortunate then, that they had widespread adoption of Splunk within their organisation, and a large corpus of experienced Splunk users. This allowed for the exploration of a simpler solution using the Splunk Machine Learning Toolkit.
After receiving a set of labelled sample data from the customer, engineers at Somerford were able to build a machine learning model which used textual analysis to “read” the freetext portion of a support ticket, and to decide what category it should be. Surprisingly even to us, it achieved an accuracy of 98%. This allowed for Splunk to do the hard work for our customer, and to alert them in a timely manner when the category selected didn’t seem to match what the freetext portion of the ticket suggested. Any such tickets could then be flagged for quick manual review, meaning far fewer tickets would ultimately find themselves going outside of SLA. The beauty of this solution though really lay in its simplicity. From nearly 100 lines of Python, we went to seven lines of SPL. Nobody needed to learn new coding skills, and there were no awkward dependencies to maintain. Simple, elegant, with no additional training burden, and on a platform that the users were all already familiar with. The great power of the Machine Learning Toolkit is how it lowers the bar to entry into using ML for business. If you can already use Splunk, then it’ll take less than a day to start using the Machine Learning Toolkit, and it won’t cost a penny too. Speak to us if you’d like to know more about how you can adopt ML, and what it might mean to your business.
