Data analytics has the power to enhance decision-making, identify new revenue streams, and drive competitive advantage. But no discipline is perfect, and faulty algorithms or other analytics mistakes can have costly impacts on reputation and revenue. 

In a recent CIO article, Thor Olavsrud profiled some famous analytics and AI disasters, including: 

Zillow’s Erroneous Machine Learning Algorithm 

Zillow unveiled its Zillow Offers program with the goal to make cash offers on properties the company would then quickly renovate and flip. Homes were identified via an estimate, or “Zestimate,” of their value, which Zillow derived from a machine learning algorithm. However, the algorithm had some accuracy issues, with CNN reporting a median error rate of 1.9 percent and as high as 6.9 percent for off-market homes. This caused the company to unintentionally purchase properties at higher prices than its estimate of future selling prices, resulting in a $304 million inventory write-down in the third quarter of 2021.

Amazon’s Biased Recruitment Tool 

We’ve written about this one before at the APEX of Innovation, but it bears additional exploration. In 2014, Amazon began work on an AI-powered recruitment tool to assist HR in screening and identifying the best applicants. It sounds good in theory, but the ML models were trained on ten years’ worth of prior resumes submitted to Amazon—most of which were from men. As a result, the system vastly favored male candidates, even going so far as to downgrade applicants from all-women colleges. The company tried to tweak the tool to avoid discrimination but ultimately determined it could not guarantee neutrality. In 2018, news broke that Amazon scrapped the project. 

Target’s Privacy-Invading Predictive Analytics Project 

In one of the most famous examples of predictive analytics gone wrong, in the early 2010s, Target unintentionally revealed a teenager’s pregnancy to her family. Target’s project began under the assumption that pregnancy radically changes consumer buying habits and that reaching these individuals with targeted offers could deepen customer relationships and build brand loyalty. Of course, being too targeted could come off as creepy, and the blowback from the accidental pregnancy reveal forced Target to rethink its approach. While the company didn’t abandon the predictive analytics project entirely, it began including non-pregnancy and baby items in its marketing material to make the campaign feel more random to the customer.

Olavsrud has more on these and other analytics and AI disasters in his CIO piece.