Artificial Intelligence (AI) is here to stay and its impact will be felt across society, including in all areas of human performance. I recently attended an HR technology event where AI recruiting software was demonstrated. A number of these tools compared hiring profiles, resumes, and other data, and used machine learning to match people to positions. I asked the question “where do the hiring criteria come from?” and the universal response was – “we rely on the manager to define that”.

In our work developing profiles of critical job capabilities, we find managers often lack a strategic view of the roles they are managing. This prompted me to think more broadly about underlying issues in using automated tools and artificial intelligence in recruiting.

Though recruiting technology is mirroring the trend toward the adoption of AI more broadly, there is much we don’t yet know about how AI can help in recruiting. What key considerations do we already know about employing AI in the recruiting process, and how might they influence how we use it in the future?

Cautions Around Incorporating AI in Recruiting

As you explore using more AI in recruiting, consider this —

Efficiency is no substitute for quality
We know that the cost of a bad hire is huge – from low productivity, to direct costs of replacement hiring, to loss of team cohesion. In its early stages, AI has been used primarily as a resume filter – this is all about efficiency. But filtering out candidates based on limited criteria or a basic algorithm will result in good candidates being ignored. Some filtering may be necessary, but getting to a shortlist based on limited criteria means putting efficiency before quality. Better to invest in a robust hiring process that puts quality first.

Key success factors are often the hardest to measure
Recruiting technology usually uses job-specific skills and knowledge as a key candidate filter. This works if you have a large candidate pool of people who all have the critical capabilities you need, but this is rarely the case. For example, that type of skills matching is very common when recruiting computer programmers, but the best programmers are great at visualizing complex systems and learning new technologies quickly. Therefore, whether a candidate is a fast learner and can handle complexity is more important than his or her expertise in today’s leading technologies – and is difficult to infer from lists of skills and knowledge.

Bias always exists
Computer algorithms hold the promise of eliminating certain kinds of bias, but algorithms reflect the biases of the people who designed them and the data upon which they are based. Here are three truths about high performance:

  • There are many ways to be successful
    A one size fits all recruiting model won’t get it done. Looking at a current successful employee is just a snapshot of one way to excel in the job. Think about different ways employees could succeed in the role and the range of different strengths that could make some successful. Considering too narrow a range as a first pass will surely limit the pool of capable people.
  • Capabilities reside in teams, not just individuals
    We do not work alone. Frequently it’s the team that needs the capabilities, not any one individual. Over time, biasing requirements towards closely matching those of existing team members leads to less diverse opinions and expertise. Think about the extent to which you need a wide variety of knowledge and skills vs. deep expertise in a few. If you are doing automated candidate matching, make sure that the algorithms that score different factors take this into account.
  • Current capabilities may not be critical to future performance
    Past performance may be a good indicator of future performance for a specific job, but if the job changes, it may require capabilities not well represented by your current employees. A focus on perfor