Machine learning is one of those hot technology categories that has lots of business and technology executives scrambling to see how their organizations can get in on the action. Done right, machine learning can help you create more effective sales and marketing campaigns, improve financial models, more easily detect fraud, and enhance predictive maintenance of equipment—to name a few.
But machine learning can also go terribly wrong, making you regret that enthusiastic rush to adopt. Here are five ways machine learning can go wrong, based on the actual experience of real companies that have adopted it. They’ve shared their lessons so you can avoid the same failures.
Lesson 1: Incorrect assumptions throws machine learning off track
Project PSA, a US firm that designs and builds professional services automation software that helps consulting firms run their businesses, learned this lesson the hard way when it tried to use machine learning to forecast variances in staffing plans.
Because consulting firms are all about specialized and well-trained consultants and using their talents efficiently, firms often employ project managers to assess and forecast the staffing needs for their projects.
They then track the time consultants spent on each individual project to bill clients for that time. If the organization manages both activities in a single system such as a professional services automation tool, there are some distinct advantages such as being able to compare forecasted with actual hours, to see how good different project managers were with the accuracy of their planning.