AI-Proof Your Hiring: Tactics to Humanise Recruiting
Four practical tactics to treat job applicants with empathy.
If you’re short on time, read the 30 second version of this post.
AI is making recruiting harder. You may have hoped it would automate the search for your next star employee. Instead, candidates and employers are locked in a game of one-upmanship, with "efficient screening" tools on one end and “automated applications” on the other.
We’re missing the opportunity to use AI to elevate people, transforming the experience altogether. Easier said than done, right? But I believe it’s possible, and I’ll outline four practical ways for founders and hiring managers to do so in this post.
Before we get into those possibilities, here are some trends I already see playing out:
Smaller teams: The rise of smaller teams due to the end of the ZIRP1 era and an AI-assisted workforce. As others have predicted, we may be close to the first one-person unicorn.2
More competition: With smaller teams, competition for jobs will inevitably increase. Standing out as an applicant will be more challenging than ever. We’re already seeing this play out in 20243.
Signal gets harder to discern: How much signal you could gather from a resume and cover letter is debatable. What happens when AI gives everyone a perfect cover letter and a highly polished resume? These signals will be less useful even for those who believe in them. We’ll need to rethink how we gather signals from job applicants, and candidates will need to get more creative.
The signal point is one I hear a lot from founders. Given the current state of the job market, the issue isn’t finding candidates; it’s finding high-quality ones from the pile of applicants. See below from the FT.
We’ve seen a really rapid shift in employer sentiment. Two years ago, their biggest complain was volume, or lack of,” Andrew Flowers, director at Recruitonomics, says. Now many say they are overwhelmed with applications — but still struggling to find quality candidates among the deluge.3
There are ways AI can help, but it isn’t as simple as relying on an algorithm to pick keywords or having an AI agent run screening calls. What to do instead? Let’s start by looking at where traditional methods and AI fall short before exploring some practical tips that founders can use to navigate hiring in the age of AI (if you’re short on time, skip to 3. practical tips below).
1. Where Traditional Methods Fall Short
“Submit a resume and a cover letter” is how most companies ask candidates to apply for a job. Whilst it’s the best out of the available options, it’s limited in what it can tell you. Constrained to a resume and cover letter, many use prestige associated with previous employers and/or alma mater, a signal that I’ve personally found bears no correlation to a candidate’s performance.
Referrals are common, but also flawed. Candidates are incentivised to submit referees that provide biased references (most do), and even in the case of back-door references, how much can you trust that the referrer is giving an honest evaluation? A personal example: I had an ex-employer carry out a backdoor reference at the offer stage with someone I had never worked with before (even though we were at the same company). The referrer gave me a positive reference, and I got the job. Still, the employer never really validated whether the reference was legit; and, to be clear, it wasn’t.
In many cases, interview processes can be designed to assess how well a candidate can perform in the interview process rather than on the job itself. Google’s brain teaser questions are an often cited example; eventually zero correlation was found between a candidate’s ability to answer these questions and on-the-job performance4.
The old ways are overly rigid and not focused enough on allowing candidates to showcase their skills based on what they’ll need to do on the job. But it’s the only signal employers have had to evaluate which candidates are worth progressing.
2. The AI-Hiring Paradox: More Tech, Less Human?
A lot of effort has been spent by recruiting-focused startups on keyword matching, from a job description to a cover letter and/or resume. The thinking is that a perfect combination of keywords will predict a match between a job and a candidate. I see this as an impossible challenge, as many job descriptions and resumes/cover letters are inaccurate. I have yet to see any of these products succeed, and I doubt we will. The inputs would need to be quality-controlled for accuracy - impractical to say the least.
I also dislike the incentives this creates. If candidates need to hit specific keywords to make it past an algorithm: Is this a signal that they will be good at their job or that they’ve learned to game the process? A hacker news post discusses how candidates are using “white-fonting” to make it past a resume filter with text that’s only legible to an AI, for example. I suspect we’ll see more of these in the near future.
Finally, there’s the “AI agent replaces the recruiter” theory. There are many reasons why I don’t see this playing out, mostly, human psychology. Will the best candidates really want to spend their time talking to an AI agent instead of a human being? When one of these candidates is assessing multiple job offers, will they go with the company that built rapport with them throughout the process or the one that treated them as a number, a task to automate?
3. Practical tips: Using AI to humanise recruiting
Instead of forcing AI into existing hiring processes, we can refine and reimagine what’s possible while treating candidates empathetically. It can be used to:
Evaluate how a candidate thinks.
Identify outlier candidates.
Provide an interviewee your full attention.
Create an inspiring job description (hear me out on this one).
3.1. Evaluate how a candidate thinks
Allow candidates to use AI in their application and interview process. I’ve noticed a few companies “forbidding” candidates from using AI, and it’s something I strongly disagree with:
They will use AI on the job, why not see how they use it? It’s not the output but the candidate’s ability to use the AI to progress their reasoning that matters.
You can quickly assess where the candidate’s “quality bar” is. If a piece of content sounds like an LLM wrote it, that’s a good signal that a candidate isn’t worth progressing.
If the candidate is proficient at using LLMs, it proves they’re curious enough to experiment and keep their toolbox updated.
I’d take this a step further and recommend a session where you get the candidate to prompt an LLM in their interviews. You can learn a lot about how a candidate thinks based on how they prompt, and having an output they can provide feedback on is a great way of doing that. Remember it’s not the output that matters, it’s how they think that does.
3.2. Identify outlier candidates
Give candidates an optional way of showing you how good they are at their job, in their application. One that isn’t constrained to output an AI can necessarily filter. Think about what you are good at evaluating that an algorithm might not be. Let candidates share a portfolio, feedback on your product or anything that humanises them to more than just a name and a few rigid documents (you could even let them show you how they use AI in practice).
The candidates who do go the extra mile will likely be the ones who are most motivated anyway. It might not be something an algorithm can help you screen, but it’s something that you can evaluate for originality and taste, which seems to be in vogue right now anyway5.
3.3. Provide an interviewee your full attention
If you’re doing one-on-one interviews, as the interviewer you’re juggling a couple of things. Asking questions, listening to the answers as well as writing down notes. Depending on how well you multitask, it can lead to the candidate feeling like they’re not getting your full attention. I’d previously have to apologise to candidates at the beginning of the call, assuring them that I was taking notes rather than being distracted elsewhere. That was before Metaview.
Metaview has been one of those magical products which feels like it benefits everyone in an interview. It’s an AI transcription service which takes notes, and is built specifically with the recruiting use case in mind. The only additional task is ensuring candidates are comfortable with a transcription service at the beginning of the call. Once the interview is finalised you can “generate AI notes” that allow you to add commentary before sharing with others. This allows you to give the candidate your undivided attention during a call. Magic!
3.4. Writing a job description
I know what you may be thinking. Fully outsourcing this task to an LLM sounds too easy and that’s because it is. Use it to get to a first version that you can share with the team for unfiltered feedback. To outline what it is that you’re looking for and to create multiple drafts in quick succession. All of these are things which lead to a better output, usually in less time.
Here are some specific steps I’d recommend
Scope definition: Before prompting an LLM, ensure you have clarity on what the role will entail (responsibilities, skills, etc). Share it with the team/ hiring panel for feedback, and gather ideal LinkedIn profiles. This usually leads to a few rounds of iteration and refinement on who the target candidate is.
Provide context and guardrails: This step is about preparing the LLM. Feed in the notes from your scope definition step and explain your company’s tone of voice. Provide example job descriptions you admire and detailed instructions on what you consider a good job description (here’s a resource I usually refer to).
Iterate: You will likely dislike the LLM’s first output. That’s ok, you now have something to iterate on. Provide specific feedback on what you dislike. You’ll likely go through a few iterations, some edits coming from the LLM, others from yourself. This back-and-forth is valuable, as it forces you to refine the job description.
Peer review: Share your draft with your network to identify any remaining blind spots or biases. Use this feedback to provide an FAQ section where you can pre-empt questions candidates may have. Feed this back to the LLM and evaluate whether you have any remaining blind spots.
This is the power of AI. With the right prompt, feedback, and guardrails, you can produce a job description that’s a step above what you could have done before. This is potentially the first impression a candidate will get of your company, so it’s worth spending the time in ensuring it’s a good one. You don’t get a second chance to create a first impression.
Conclusion
Notice how all of the above recommendations are designed to treat candidates with empathy. In an age where many will take the “efficiency” route, the companies that spend time treating candidates with empathy are likely to attract the best talent.
These next few years are arguably the most exciting time to be in and around startups. As the early adopters of new tools and ways of working, those of us working in them will get a “front-row seat” to this new world. But there are still many open questions and uncertainties about what this means for the world of work and especially for how we hire.
We’ve started this newsletter to directly uncover insights from conversations with startup founders who can share insights into how they are hiring, what candidates are doing to stand out in job applications, and how they predict these next few years will play out, both in terms of recruiting and the wider world of work. We’ll focus not just on the tools they’re using but also on how they are allowing candidates to provide a more authentic, human version of themselves.
To say “the only constant is change” is an understatement, the pace of change will only accelerate. Through conversations with founders we’ll help others connect the dots. If you are a founder that’s hiring and would like to have your role covered in this newsletter, alongside your thoughts on hiring, please do reach out.
https://en.wikipedia.org/wiki/Zero_interest-rate_policy#:~:text=Zero%20interest%2Drate%20policy%20(ZIRP,amid%20the%20COVID%2D19%20pandemic.
https://medium.com/pi-labs-notes/sam-altmans-one-person-unicorn-and-the-future-of-the-office-8ca607b63012
https://www.ft.com/content/1429fcb2-e0ef-4e47-b2b8-8bd225ac2fe2?accessToken=zwAGInZpsTtgkc8UKfyy4O9OR9OyuIvSJawv4g.MEYCIQCKURRXuo1SYFGdV_9mfIaaAs0b5zHrjVRxyepyoOsVAgIhANidhgZ7DfDr4mLDQEqYPZeJNFdqcd_Th1dHHCoW3S7P&sharetype=gift&token=b907f04a-03f4-4515-bd24-a2344591d601
https://www.journeyfront.com/blog/googles-interview-questions-were-all-wrong.-how-are-yours-doing