Problem
A utilities company was looking for ways to predict complications arising as a result of tickets given to individual engineers, using a variety of data from disparate sources.
The data was often incomplete and not standardised making it challenging for engineers to find important pieces of information before they started work. The company wanted to refine this process, and provide easy to consume warnings and information to engineers to reduce complications.
Solution
Large amounts of unstructured information on tickets contained in call centre recordings, historic incident reports, and equipment data was not being seen by engineers working on sites since the information was inaccessible and poorly organised.
Mind Foundry was able to ingest the disparate sources of information, and use them to engineer features predictive of likely damage resulting from the ticket.
Given the highly imbalanced nature of the data set, there was a high false positive rate, but by tuning the prediction threshold, the client was able to trade off precision with recall, and arrive at a strategy that significantly reduced overall risk by pre-emptively flagging at risk jobs.
