close icon
close icon

Search our site

Computer says no: The limits of AI for recruitment

Talent Attraction

Share this post:

Can ‘clever machines’ really pick the right person for the job?

Much has been made of AI as a panacea for our times. In theory, cognitive technology can make complex decisions much faster than humans, harnessing countless data points to improve the quality of those decisions. Its performance should also improve over time, as the results are fed back into the system in a continual feedback loop.

The benefits for recruitment at first seem obvious.

Recruiters looking to fill highly competitive positions, and those involved in volume recruitment, are faced with the task of sorting through hundreds of CVs. Busy recruiters have famously been said to spend on average 5-7 seconds reviewing a CV, using rudimentary red flags such as an ‘unprofessional’ email address or a single typo to discard candidates. This can leave the nagging suspicion that a potentially great individual has been passed by arbitrarily. But what if this initial cull could be handed to an AI that reviews every aspect of every CV in a fraction of the time?

This approach is exactly what tech giant Amazon recently trialled in its Edinburgh office. The team created an AI to select the most promising candidates from a stack of job applications, and trained it using over 10 years of historic data. However, the AI quickly learned the biases inherent in the past data and started to mark down all female candidates. All CVs that included the word “women’s” as in e.g. “women’s chess club champion” were penalised, as were two all-women’s colleges. The team attempted to fix the bias, but eventually the project was abandoned and its candidate rankings discarded.

The human touch still wins out for talent attraction

The AI’s downfall stemmed from the fact it was configured to replicate what came before, without the ability to critically analyse the data or creatively envision a different but preferable outcome. While computers can become adept at data analysis based on existing models, they neither understand the meaning of what they’re doing, nor are they able to step outside the parameters of the task to see the bigger picture.

Recruiters should take heart, though – this means that talent attraction will continue to be a fundamentally human business; we are not all about to be replaced by robots! What’s more, there is another way to address the challenge of improving the quality of your candidate funnel.

Here at Creed, we’ve used audience profiling on a number of projects to pinpoint the precise demographics and locations of our clients’ target audiences. This information is used to devise messaging that resonates highly with the desired audience, and optimised media campaigns that target them via the right channels, at the right times and places.

For examples of how audience profiling delivers better hiring results, check out our Sky AdSmart campaign for Vodafone, or our peak season campaign for Hermes.