5 mins read
Recruitment has always been about maintaining a balance between efficiency and the human side of hiring. AI is changing that balance as it introduces new tools to manage workloads and handle candidate information. Simultaneously, recruiters and HR professionals have to be deliberate about when and how to use AI and when human involvement still takes precedence.
Here, we examine six ways AI should be leveraged in recruiting as well as six related missteps recruiters should do their best to recognize and avoid.
Do: Use AI to Screen Resumes for Relevant Criteria
It’s common for sought-after positions to create a lot of interest, leaving hiring managers to sift through hundreds of applications. This initial part of screening is tedious and error-prone when done by humans. Meanwhile, an AI can easily compare applicants and point out those that best meet your selection criteria.
Don’t: Let the AI decide who the best candidate is
It's important to emphasize the "your criteria" part. The AI should be given clear instructions. Maybe prioritize candidates with X years of industry experience or ones who hold a relevant industry certification.
Vaguely instructing an AI to find the best candidate risks discrimination through biases in its training data or biases connected to past selections. Ideally, the AI should only create a shortlist tailored to your criteria, while deeper selection and the ultimate decision rest entirely on you.
Do: Use AI to Improve Candidate Outreach
Candidate sourcing is repetitive and can be time-consuming, especially when you’re looking to fill a niche position. AI speeds up the discovery process by identifying prospective candidates, summarizing the publicly available professional information on them, and drafting an opening message.
Don’t: Let AI reach out autonomously
In the context of outreach, AI is best used as a research and screening tool. Recruiters can verify the information it presents and then determine the best way to pursue the lead, or if they should do so at all. Leaving actual outreach up to AI risks sending out artificially personalized messages and making broad claims about candidates’ professional interests and career moves with little to go on.
Do: Use AI for Repetitive Candidate Communication
Recruitment chatbots benefit both parties and can facilitate communication that doesn’t require complicated judgments. Candidates feel more at ease when they can immediately receive answers to simple yet important questions concerning application deadlines, upcoming interview logistics, working conditions, etc. Meanwhile, recruiters don’t get inundated with such questions and can devote more time to selection.
Don’t: Let the chatbot be candidates’ entire communication experience
Chatbots are genuinely helpful when they’re given a clearly defined area of responsibility. Stepping outside of it often results in misunderstanding questions and providing overly confident yet wrong answers. Chatbots should always connect candidates to a human when a question is complicated or ambiguous. The process should be straightforward and reasonably quick as well.
Do: Audit AI for Bias Before and After Deployment
An AI hiring system might produce materially different outcomes for different groups of people, even if conventional biases are removed. Other characteristics can stand in as proxies and lead to discrimination based on historical hiring data.
For example, a preference for candidates who live close to an office in Manhattan favors individuals of a higher socioeconomic status. Even a candidate’s name could be enough for a model to infer information about their ethnicity and gender, and add weight to it if historically successful candidates had the same name.
Don’t: Assume that vendors’ bias-free claims are enough
You have a responsibility to evaluate the model you’ll use, which also means ascertaining fairness by asking vendors some hard-hitting questions. What data was used in testing and training? Which groups were involved? What fairness metrics were used? Only once you get satisfactory answers should you proceed with procurement.
Do: Treat Candidate Data as Part of the AI workflow
Recruitment inevitably creates a lot of sensitive information linked to candidates, from resumes and screening results to interview notes and potentially background check information. Copying this into AI systems is remarkably easy and even necessary for the AI to be able to make more complete recommendations.
Don’t: Expose candidate information to AI indiscriminately
Information exposure should be minimized, so it’s a good practice to consider the type and sensitivity of information an AI needs and not feed it more. Routing AI traffic through an LLM gateway helps enforce this by masking or stripping personal data before it reaches a model. It’s also important to understand how AI providers handle supplied data. Is it processed and used as new training data? How is it stored, and who has access? What related notices does the jurisdiction in which recruitment takes place have? Answers to these questions should inform your judgment and data collection strategy.
Do: Keep Humans Meaningfully Involved in Hiring Decision-Making
Humans should be able to intervene whenever AI is making or influencing consequential hiring decisions. They should be able to question, reexamine, and overturn any decisions or recommendations an AI might have made. This becomes more important the farther the process progresses. Recruiters need a clear understanding of what information they need to examine and which decisions require their review.
Don’t: Automatically approve algorithm recommendations
The risk of overreliance on AI is creating a situation where human oversight exists nominally, but it’s just someone blindly agreeing with AI recommendations. The underlying evidence always needs to be scrutinized, and a recruiting professional needs to exercise their judgment when making the final selection, let alone extending an offer.
