True problem solving & the story of our technology competition
Artificial Intelligence and Machine Learning have become increasingly commonplace in recent years. A 2021 study by PWC found that 86% of respondents believed that Artificial Intelligence had become a mainstream technology. The study also found that 52% of those surveyed had accelerated their adoption as a result of the COVID-19 crisis.
While the potential impact of these technologies cannot be understated, organisations must give careful consideration to how these technologies align with their business objectives and whether or not an alternative approach can deliver the outcomes they need.
Background
In 2022, Analytics Engines began the delivery of a project for a US-based software company. The client organisation offers services that help enterprise organisations gain insight into their use of hardware, software, SaaS, and cloud assets, enabling them to optimise their IT spend and realize maximum ROI.
Their software platform is regarded as the world’s most complete repository of technology information. In response to an ambitious modernisation roadmap, the client was searching for a development partner to deliver internal microservices to replace a subset of repetitive, manual tasks which were part of their existing manual workflow.
The client was manually processing large volumes of data to identify what versions of software their customers were using. The client wanted to improve their efficiency and response time for their customers. Our team supplemented the internal capacity and enabled them to meet their objectives as a customer centred organisation.
The client believed that a Machine Learning based approach could help to address this issue. Based on prior experience however, our data scientists believed that an algorithmic, rules-based approach could deliver similar, if not better results.
Our team also believed that a rules-based approach would help to eliminate many of the issues associated with a Machine Learning based approach including increased costs, more difficult deployment, and a need for ongoing maintenance.
The solution
To discover which approach would be most effective in the delivery of the project, we conducted an experiment using a sample dataset provided to us by the client. We tasked both the Machine Learning Model and the rules-based Algorithmic Model with identifying and classifying the version numbers of the software contained within the sample dataset. From this test we discovered the following a rules-based approach was able to deliver a success rate of more than 99%, while a Machine Learning approach was only able to deliver a success rate of 70%
Based on the outcomes of this experiment, a solution was developed utilising a rules-based approach that combined data cleansing with automatic evidence matching. The algorithm is scalable while maintaining the high accuracy of the manual matching process.
The solution was able to automatically identify and classify software versions with a more than 99% accuracy. The solution has automated a significant portion of this project and has drastically reduced the need for manual processing.
This approach also provided the client with the flexibility to customise and target high-value products from select manufacturers. Many of these high-value products have particularly challenging naming patterns because they are released with many small variations in nomenclature that lead to different product or release IDs. With the algorithmic approach, the client can examine the data and add bespoke rules for these challenging high value products.
