Addressing challenges with AI in inclusive hiring

6 mins, 28 secs read time

We’re living in a new era. Over the past few months, there has been a 50% increase in candidate applications for opportunities at companies using Greenhouse over what we’ve seen in comparable months past. The pandemic has created massive unemployment in the US, and people are desperately competing for available roles.

Companies that are still hiring are trying to figure out how to sort through this high volume of applicants – often with reduced hiring teams. At the same time, as companies reflect on and commit to racial justice, they’re attempting to make these decisions not just quickly, but also fairly.

As a hiring software company, we recognize that we have a huge responsibility to help our customers navigate these new challenges and make the best, most fair and quick decisions possible.

The current role of AI in resume review

Many other HR tech providers wrestling with this new influx of applicants are inevitably turning to product innovations to solve them – including AI. And with all the leaps and bounds that many AI apps have made in recent years, this makes sense.

When it comes to AI in resume review, the concept is straightforward. The machine will look at all the resumes submitted, compare them to the job description, and then decide who should move to the next round and who shouldn’t. Many of the recent innovations in AI are anchored around matching and, at its core, screening candidates at the top of the funnel is a matching problem. There have been plenty of investments, acquisitions and, in true Silicon Valley form, grandiose promises surrounding AI in the hiring process. But, there are problems with this approach.

Challenges with AI in hiring

It’s inherently opaque

By using an algorithm that can’t be easily explained, AI tools remove agency from both candidates and companies – everyone is at the mercy of the technology. That’s why we’ve ended up with a recruiting content industry aimed at helping candidates “beat the ATS” – because no one, including the companies, actually understands how the decisions are made.

It amplifies existing bias

AI predictions can only be as good as the training set provided. So, for companies with existing issues around diversity, using their employees as the training set will often lock in their existing problems. If you use data from outside your employee base, it’s unlikely you can map that data to your employee base in a straightforward way. There are a few high profile examples of these algorithms going wrong, which should give us all pause.

How to address the problem

We turned to experts in the field to help us figure out the best solution. Their research shows that manually scoring applicants with simple, explainable rules can be just as accurate as using a trained model. And it’s a much more transparent process. It’s incredibly important for the candidate and the company to clearly understand how and why a decision is made. Because of this, we believe any solutions we implement at Greenhouse need to have two key characteristics:

1. Transparency. Companies should tell their candidates how they will be evaluated and what will be done with their data.
   
   
2. Explainability. Companies should be explicit about what they think is required for success in the job, design a process to elicit that data and then apply decisions consistently.

Here’s how you can commit to these guiding principles to make faster and fairer decisions.

Articulate your needs

This may sound obvious, but a lot of people skip this step. If you don’t know what you’re looking for, how will you know when you find it? Be clear about what you’re trying to achieve before opening the role – what are your company and team goals? This will help you make objective, explainable decisions.

Give an honest view of the job

Rather than shading the truth to make the job sound amazing, be honest about it – warts and all. You’re far more likely to attract candidates who will actually do the job, and reduce the number of people who may apply and back out. In addition, share details about the candidate evaluation process and how you are going to handle their data. This transparency can remove unfair advantages that in-the-know candidates – like applicants who have been referred and already know how the process is run – may have. It will also help you comply with ever-strengthening data privacy laws.

Add more useful questions to the job application

This will help you make more informed decisions earlier on. Candidates will submit any job application that takes a few seconds to fill out – a few extra thoughtful custom questions will guarantee investment and allow you to make smarter decisions. That being said, make sure the questions really elicit useful data and aren’t just needless torture.

Use data to make explainable, structured decisions

There are usually answers that guarantee the person won’t get the job (for example, missing licensing requirements) and others that will identify the person as someone you really want to speak to (for example, specific and applicable technical skills). Automating these decisions with specific qualifiers can save tons of time and is still easily explainable.

Evaluate with assessments

There’s a range of options on the market for evaluating candidates. While this may create a bit more work for the applicant, it also gives them agency in the process. They have a much better opportunity to distinguish themselves than by simply submitting their resume to a black box. Note that these assessments should follow the same principles above. They should be explainable and transparent, and specifically tied to job-relevant qualifications.

Train your recruiters to review resumes more efficiently

Many applicant tracking systems require several clicks to access a candidate or download their files, and then additional clicks to advance or reject the candidate or send an email. This obviously slows down the hiring process. In Greenhouse, users have the option to access all the information about a person in one place, make a decision in a single click and move on to the next person. This helps set recruiters up for success. Your hiring software should not be a barrier to efficiency when dealing with a high volume of candidates.

Classify and track your rejection reasons

Documenting rejection reasons helps to minimize opportunities for unconscious bias by ensuring mental shortcuts and assumptions don’t interfere with hiring decisions. Interviewers are forced to be aware of their reasoning in the moment, rather than following a gut feeling, which may stem from the fact they didn’t like the candidate's outfit. Tracking reasoning also allows you to report on it in the future and investigate potential issues.

Clearly communicate what happened with all candidates

Once you are tracking rejection reasons and making decisions in a structured way, you can easily set up email automation templates to follow up with candidates about their stage in the hiring process. This type of transparency allows applicants to feel respected and understand your reasoning – even if they didn’t get the job, you’re providing a positive candidate experience.

Following these actions won’t magically solve all the DE&I challenges that companies face with high-volume hiring, but it’s a great place to start. Companies need to commit to structured hiring, and continuously look at their data to make objective decisions. More importantly, we hope that teams continue to prioritize DE&I and understand that although they’re receiving an influx of applications, efficiency problems can be solved without having an adverse impact on candidates from underrepresented groups.

Thank you to Gary Davis, Greenhouse's Inclusion Product Strategist, for his contributions, insights and direction on this article.