287. What if A.I. is Fairer Than Humans? The Truth About Bias in Recruitment, with Kate Young, Head of People Science at Sapia.ai

23:19Maybe start with episode 124, where you're going to hear how industry giants from Canva to YouTube define leadership. Listen to Billion Dollar Moves wherever you get your podcasts.

23:33Yeah, and again, as well as you say, in terms of fairness, that type of process allows for people to think and reflect, doesn't it? It doesn't put time pressure on it. which yeah I'm thinking my nephew is dyspraxic and he is incredibly intelligent and empathetic and talented but he you need to give him space to to communicate that so it's I can imagine that will work really well for neurodiverse candidates as well it really does and it's something I'm

24:01Kate Young:actually really proud of my own daughter was diagnosed as autistic last year so I think a lot about how she'll experience applying for roles and we track this data we're data obsessed as you might expect from an AI company. And we find our satisfaction rates for either neurodiverse candidates or candidates who identify as another, having another disability, typically comparable, maybe 0.1, smaller sometimes, just because hiring is always going to feel a little bit harder for neurodiverse candidates, but they're more or less comparable and really positive feedback themes. That's exceptional. Wow.

24:36Kate Young:It is. Wow. Okay, so I'm thinking now, because I can hear Al in my ear as like the business owner going, they ask this think about this I imagine he'd be going he'd be sitting there going wait a minute candidates all go through to an interview that they can type or speak into with no time limit how do I know that's them how do I know they've not got their mate or their parent filling it out how do you not know they're asking chat GPT how do you how do you check for the authenticity of who it is you're actually interviewing it's a really important question and it's a question that's not unique to an AI chat interview.

25:14Kate Young:We've had, I call it unprocted, unsupervised, hiring methods for years, personality tests, cognitive ability tests, those types of things that candidates will do with no supervision whatsoever. We've always just had to trust it's them. And unless we're going to proctor to supervise every candidate, which given the volume of applicants to jobs now is impossible, we have to accept I believe we have to we absolutely have to accept some level of risk like there are proctoring solutions out there where you switch on your camera and you're being kind of watched um it's not something we're willing to explore because we value that trust with the candidate and that inclusion and being filmed by camera is the opposite of an inclusive and comfortable experience so there's a little bit of a trade-off there so we're choosing to prioritize that kind of experience and trust over absolutely knowing it's not their mum doing it there's a lot of prompts and nudges in there around making sure it is them.

26:11Kate Young:So if we detect they're using AI-generated content, we stop them copying and pasting anyway. So if they're going to use ChatGPT, they're then going to type it all over. And that friction just will discourage a lot of people. And then if we detect they're doing it, they get a nudge like, look, we think you might be using AI content. Please don't do that. We need to see your responses and your truth. And if they do it again, they get prompted again. Typically, most people stop after that prompt. They don't do it. So, yeah, I can't do anything if they get their mum to do it. But by and large, because they enjoy the experience and they trust it, I think those rates are going to be pretty low compared to other solutions.

26:48Okay. So what happens next? I'm really enjoying this kind of going through the journey as a candidate and thinking as the client as well. So I've done this interview. What happens then with the data? What happens with the client? How do they get notified? What happens next?

27:06Kate Young:Oh, well, how nerdy do you want me to get? Very, please. Very nerdy. Okay. So in terms of data, so the data is their five chat interview responses. And this is really, really important because when we talk about AI in hiring, people will go to certain well-known lawsuits and they'll assume that, for example, my name, Kate, which is clearly a kind of female-leaning name, is going to go in there and it's going to tell the AI algorithm that I'm a woman and that's going to bias it. that is so we are super super clear that the data is those five responses um they're they're consumed um by a scoring model um so it's a scoring model you call it like a prompting framework and i kind of like to think of this as um something that's holding the hand of your of your large language models we use claude um all your listeners will have heard of chat gpt that's the most popular one claude's just like another one um it has the best privacy settings that's why we choose Claude and some other things.

28:01Kate Young:And it's holding the hand of Claude and saying, okay, Claude, we need to score this candidate's response. Imagine you are an expert occupational psychologist trained in all these principles. This is all the things I need you to consider. This is what you're scoring against. This is how you check you're getting it right. If you get it wrong, this is how you correct yourself. This is what counts as evidence. This is what doesn't count as evidence. And it'll go through that and it'll make you do it three times. And then it's going to give scores. So you get, I mentioned these competencies before.

28:28Kate Young:So you get a one to five score for each competency. And then they're pulled through into reports. So more importantly than score, and the score is what will be used to help effectively rank the candidates. So hiring managers can start to identify the ones they want to move through to whatever the next stage is, typically a telephone or a face-to-face interview. But what you're getting is one of those scores, a lot of explanation. So when we think about AI hiring and AI governance, explainability is the beginning, the middle and the end. Where you get a lot of fear in AI hiring is what we call black boxes.

28:57Kate Young:So data in, something comes out and no one understands how it got there. Of course, I'm not going to trust that. Why would I? Whereas with our model, the hiring manager will get a report that explains exactly how that four out of five or whatever's arrived at. So I'm giving Leanna four out of five for teamwork because she's told me an example of how she motivated a team when they're under pressure. And in question three, she told me about how she comforted a team member in a specific scenario and it will connect A to B repeatedly. So the hiring managers can have real real trust in that and also they can query it if they want to as well um you know because nothing's hidden and what's the threshold to pass as such in terms of the next stage is that individualized by the client or the role or is it kind of like a set benchmark we don't automate it pass fail um we give back a score to clients um and it's depending how many people they need to move through to the next stage we'll define where they tend to set their benchmark but typically because of the types of roles we're hiring for, the client's going to be applying a couple of other filters as well.

30:00Kate Young:So if we're thinking about retail, they're going to also just do some filtering for people who are available to work Saturdays and so on and so forth. They meet some other criteria. So they do some filtering and they're going to rank the candidates and they're going to work through those from top to bottom and start moving them through to interview. But the actual number they'll move through from kind of that first stage to second stage varies massively across client just depending on their hiring needs. So I'm also seeing at this point, hiring managers are quite happy because the first that I am, obviously I've been involved in the job analysis and checking those competencies.

30:30But the first that I am then, in terms of my time dealing with is candidates that have been screened, I've got lots of nice information for that I can filter for different criteria that I want. And then I go on to say the face to face, the telephone interview stages. Is that what you're hearing or are hiring managers a little bit distrustful of the AI that's happened and decisions that have been made without them?

30:55Kate Young:Well, hiring managers are not a homogenous group. It would make my job a lot easier if they were. By and large, hiring managers are delighted because their time to hire, which is what most of them are targeted on, is going down as much as 70 to 80 percent, especially for high volume clients. and they are finally released from the tyranny of CV screening, which nobody enjoys. It's a little bit of novelty when you first start. My first role was as a recruiter. I was a terrible addict because there was too much sales involved. And I used to find it really fun going through every CVs and going, oh, that person's put a picture on.

31:27Kate Young:That's ridiculous. But then the novelty kind of wears off because you're just trying to make decisions with rubbish data and repetitive data. So they're released from that. They're timed hire improves. They're quality of hire improves. But yeah, there are absolutely people who are reluctant to trust. And then that becomes a job for us to run the analysis, prove it's working and kind of hold the client's hand through that change management piece. And typically where we struggle most is where there's an easy alternative for the hiring manager. So again, in something like shops or retail, walk-ins where people walk in and say, I really want to work here.

32:03Kate Young:That's really compelling for a hiring manager. Like somebody's bothered to travel to the store, make themselves smart, walk up, shake their hand and say, I really want to work here. Can I interview? So that's usually our kind of big tension point is pulling people away from that and into the AI hiring tool. What's to stop a hiring manager filtering for, I don't only want to employ women. I only want to employ men. I don't want anybody who's neurodiverse. I don't want, you know, how do you, is there any way to prevent that? Or once that data is back to the client, it's theirs to do with what they see fit.

32:36Kate Young:So none of that data exists as filterable in our system. We collect, are you male, female, what's your ethnic group, only for anonymized background monitoring so we can check the system's fair. Ultimately, I can't stop a hiring manager if they've met two candidates and one's a man and one's a woman and they really wanted to hire a woman. I can't stop them making that decision and nor should I really because the decision, the tool doesn't exist to make the decision exist to provide the best possible information. So the decision sits with the hiring manager and that's another kind of really important principle around AI in hiring and accountability.

33:09Kate Young:You've got to be really clear what decisions sit with the humans, what decisions sit with the AI, what sit with me as the psychologist, what sit with the hiring manager as the person doing the job. and if I'm recruiting for a role and I know this is going to be a massive depends but let's say I'm hiring for for a role that typically in traditional recruitment might get 500 applications through this system how many candidates am I likely to pop out with it the end where then we'd take them to kind of second stage of that recruitment process oh that is massive it depends because I guess what I'm trying to get at is do I how is this in change what businesses or hiring managers might be used to am I getting far less candidates through am I getting maybe four or five am I getting hundreds that I need to to go through and I understand it can be depending on the role in the but I guess I'm running with these are a typical kind of outcome that hiring managers are going to get when they're faced with people who've gone through the that competency interview sure so if you want to typical outcome, we generally, we do sort of give a yes, maybe no label based, and that's purely based on score.

34:20Kate Young:And that is to help busy hiring managers if they're making decisions at pace. And if they've already established their trust in the tools, they're like, you know what, I know Sapia assesses well. I know the score's robust. I'm going to trust the tool. I'm going to interview the people Sapia recommends. And we'd usually look to be about 30 to 40 % yeses. So we'd be saying you should probably review these people now. A lot of hiring managers won't do that they'll they'll go top down they'll they'll start with the best schools and just work down um but yeah that's typically the kind of reduction rate and in terms of again going back to that candidate side of it and the experience will all candidates get notified if they're not successful how does that work in terms of the hiring manager's role versus sapia's role so that's not sapia's role because the decision it doesn't sit with us so um what sapia will do for candidates is everybody will get feedback.

35:11Kate Young:So it's a really nice, it's called My Insights. It's actually a lovely report because it's very kind of coach-based, very conversational, quite informal, and it aims to really give the candidate a sense of their strengths, but also a tip at the end. So, you know, when thinking about future job roles, maybe consider structuring your day or something like that. So there's never that ghosting. Sapia makes socks. They're lovely socks. They're very warm. If you see them at a conference, to do grab a pair and one of them just says no ghosting on the on the foot of it so they never get ghosted they always get feedback from a sapiens side but in terms of how clients choose to notify candidates that that very much sits with them um that said we're really passionate about candidate experience and we work with clients to set these experiences up so typically we will have that conversation with them how are you planning to communicate to candidates what's your messaging and offer them that support who are the clients that you're primarily working with at the moment are they larger businesses yeah so we we sell to enterprise um and the reason for that is um it is an ai hiring solution so it's kind of a governance and a setup layer um which a small business isn't going to have the architecture for yet maybe in the future when this is more kind of normalized um and we're able to kind of democratize the tech further um so for example we work with holland and barrett in the uk which we all know where we get our our vitamins and our tasty, healthy protein bars.

36:36Kate Young:They're a really good example. Also Costa Coffee. Do you know, I'm not a salesperson, so I always forget who you know I'm allowed to mention and not, but those are two very strong examples in the UK. And then within Australia, because I'm actually headquartered out of Australia, we have a very strong relationship with Kmart, which are their sort of main big kind of retail store. They were recently featured on the Josh Burson podcast talking about how they'd increased the efficiency of the experience with Sapia. And I guess what, I'll be honest, Kate, right? As you're going through the process, I'm loving it.

37:10It's everything that I was taught and everything that I've done with clients. And I'm also thinking, well, I'm redundant then because if AI can do all that, they're not going to need an occupational psychologist to do it. So now I'm thinking, is there a future where Sapia would maybe, I don't know, license this software to other practitioners to use with clients? if we get to the point where small businesses have that structure, what's the future with it in terms of its growth and application?

37:36Kate Young:I'm not sure we've been asked the licensing question. I can't see why we wouldn't. But I'd also want to say it's definitely not making us redundant. I did a workshop last week at the Division of Occupational Psychology Conference over in Cardiff. It's a beautiful city, wonderful conference. And the workshop explored what is our role as psychologists within the hiring system. And really the takeaway is we're just moving from individualized hands-on judgment and ad hoc review to a very sort of systemized role. So we're designing the architecture, the governance, the parameters that need to sit within a system to make sure it's working well and it's fair and it's unbiased.

38:14Kate Young:So it's like many things. Is AI going to make me redundant? Absolutely not. It's just going to make your job more interesting because you can do more of that elevated, interesting problem solving. What is the future? so I think for us what we want to do is take everything we're learning through chat interviews and all the intelligence we're building around competencies and use that to leverage it more from talent intelligence perspective so not just for hiring but also to support things like internal mobility so well you're kind of collecting all of this rich data up front about someone make it work for them as they travel through the organization when you're thinking about internal roles or perhaps coaching, so on and so forth.

39:00Kate Young:So doing more with the data we have because it is so rich. There might be someone listening who goes, but you're reducing somebody to a number. You're giving them a score. You're giving them a three or a four. Is that the right thing to do? Is that dehumanising somebody? It's a really fair question. And it might be a kind of overly pragmatic response, but I would always encourage anyone asking that question to say, what would you prefer? By which I mean, is it better we reduce someone to a CV that's screened in 20 seconds and you get a decision based on bad data, bored recruiter? Yes, they are given a score at the end of the interview and yes, it is used to make a recommendation as to whether they move forward but that score is based out of all of the robustness that I've just described, all of the great questions, all of the great principles.

39:51Kate Young:so ultimately to make a decision when you've got thousands of people in a hiring funnel you need a piece of data to make a decision on because there has never existed a time where all of those thousands of people would have got a lovely face-to-face interview that's just a kind of nirvana we like to imagine that was actually never real we used to post off cvs and hear nothing back and that's heartbreaking because you put in the envelope you put the stamp and go into the post box and you got nothing back so yes it's just it is a number essentially but it's a robust and fair number and I think in a big high volume situation that is the very best we can do for people.

40:24In terms of any small business listening who's like this sounds really cool but clearly it's not something that I can is accessible for me right now what advice would you give them in terms of making their their recruitment process more robust more fair?

40:39Kate Young:Well the best advice for small business is find a friendly psychologist in your network there's a lot of really great freelancers out there who will happily and the thing about psychologists I'm sure I'm sure you've experienced this Leanne we we just care so much about getting it right and being fair so most will just give 30 minutes an hour it's a bit like your free appointment the solicitor everyone always talks about you know it's kind of free most psychologists will give that little bit of advice and it's kind of the Pareto principle you can almost get 80 percent of the gain with with 20 % of the effort just by doing even rudimentary job analysis is simply prompting people sit down and go, what does this person actually do?

41:18Kate Young:You know, think, don't think about who you like, think about what they do. And then looking at your hiring process and going, are we giving the same interview to everyone in the same role? Or are we just going off the cuff? And have we got space to maybe introduce something a bit more structured here? And then also with the final decision making process, how are we making that? Who's in the room? and how, again, how are we making sure it isn't loudest person wins because we as psychologists call wash-ups. You get everyone in the room and you say, oh, I like this and I rated this and I did that.

41:46Kate Young:And actually what happens there is just that, you know, the loudest voice typically makes the decision. So there's some real low-hanging fruit that, you know, local-friendly psychologists can probably direct you to with the minimum effort. But failing that, go to sapia.ai. We've made lots of white papers with some really good advice on there. They're definitely worth reading. There's good analysis in there, but also some really nice kind of takeaway commentary. And there's lots of free webinars and resources around. Sapia do loads. I did one inclusive hiring the other week. And they're a good free resource just to start educating yourself a little bit.

42:22Say there's a business owner listening goes, this all sounds very nice, Kate and Leanne. And sure, I could get in touch and find a friendly psychologist. But you've already told me how this is really manual and really expensive and takes ages. And I don't have the time for that. I need to get somebody into my business quickly. What if I just send all the CVs, put them all into ChatGPT or Claude, ask them to pick out the one that matches the job description best and I'll interview that one and hire that one.

42:49Kate Young:Could I do that? I mean, I'd say you absolutely can do that, but the cost of doing that is so much more than taking a beat and thinking about putting a little bit more structure and science in your process because the cost of a bad hire, especially for a small business, when they're going to be a greater percentage of your workforce, is absolutely massive. So it is absolutely worth taking a beat, upskilling yourself a little bit, thinking about some structure, thinking about some basic points of measuring what matters and just interrogating your decision making process. So there's loads of easy how to guides and basics out there.

43:24Kate Young:Definitely go to Sapir, but there's also plenty on the rest of the Internet. it's not hard to start getting some of the basics right but it's when you kind of enter the volume of hiring space that the stakes then become really high and that's when something like Sapia is going to get involved. I think this is really cool and I think there's also organisations out there that are the HR based organisations, recruitment based organisations that might feel slightly threatened by this that are in high volume recruitment who might make arguments that human first is better, it's more personal, it's how does how do these things coexist in a world or do you think that high volume is gonna become ai first i don't think there are any high volume processes that are still human first no one's got the read absolutely no one has got the resources if they are they're the clients i've spoken to in the last year who say i my recruiters are spending hours hours hundreds of hours a week screening CVs, this has to stop, this has to stop.

44:24Kate Young:We can't afford it and it's not scalable. So I think, I kind of think in terms of putting some level of automation in whether it's AI or not, but some kind of standardization at the top of the funnel has kind of sailed, that ship has sailed. If the two can coexist, absolutely, and they absolutely should. I think AI in hiring is most effective at the start, at the top of the funnel to help you when you've got a lot of data to synthesize that data that's what's really doing is taking all the data and synthesizing it into something consumable for hiring managers and then at that point the humans come in and do those face-to-face interviews and you know that could be solutions from more traditional hr providers you know there are loads of wonderful providers out there who provide online role plays or other types of exercises and those things can absolutely coexist and also psychologists i love it when the next part of the hiring process is really different to the first part because the chances are it's going to measure different things as long as they're still relevant things and you can get more information on the candidate so everyone wins i have a very nerdy question but i love it so you mentioned before that you gave the model all of all the data from job description it popped out how many competence did you did you say it's 25 25 are you looking at maybe you already have are you looking at because i know i think places like shl have like the standard grade eight competencies are you looking at potentially I don't know using these 25 competencies in a way that could support smaller businesses for example is that what yeah tell me about that I'm interested it's a great question because the the comp frame the competency framework itself is a great resource and if you compare it to something like the great eight or corn fairies comp framework or any other really established ones you'll actually notice quite a lot of commonality and that's because those frameworks, e.g.

46:16Kate Young:the Great Eight, were also created in a really robust way. Now, they didn't use AI, they did it long before AI. So, they did a relatively similar process, but manually, so less data, more time, but it was not dissimilar. So, when you have robust psychologists working with similar data, you will get outputs that are somewhat overlapping. I think we could absolutely use that competency framework to support small businesses. So, one thing we've been experimenting with a little bit is just using the job and analyser tool to just help people define their face-to-face questions. They're not going to bring in a chat interview or AI, that's not going to work for them because of the AI governance piece and the scale.

46:53Kate Young:But you can absolutely use the tool just to support defining your face-to-face, your first stage interviews. And that's a massive time save potentially for small businesses. Is this public knowledge at the minute in terms of what these 25 are? Is it still kind of proprietary information where's that all standing um i'm pretty sure there must be a write-up of them on our website yes oh okay it should be up to name the document but i confess i can't but yes it's not a secret certainly if i'm listening as a potential client how do i look into if i could use this in my business definitely go to sapia.ai um all the resources are on there there should be a contact form on there um anyone listening who wants to talk about the science they're welcome to contact me either via LinkedIn.

47:36Kate Young:I'm Kate Young at Sapia on there. I'm not hard to find. I'm quite noisy on LinkedIn or Kate at Sapia.ai. But we love talking about this stuff and we're pretty open about it. So absolutely get in touch. Amazing. And if listeners take away just one thing in terms of hiring better, what should it be? I think I said it a few times, but it's measure what matters because unless you define what you actually need to be successful in that role, it doesn't matter if you choose the best tool, the worst tool, how you execute your process. You've undone yourself from the start. So even if you can just apply some very rudimentary job analysis, that's already going to get you a big win.

48:17That was Kate Young from sapia.ai. And I'll be honest, I did come into this conversation lightly and slightly sceptical about AI in hiring. I love AI, but hiring felt like the one area where humans should probably be involved in every single stage. But now I'm starting to think we've been doing it wrong for a very long time. I'm the same. But what I learned was that AI in this context is acting as an amplifier. So if the science and the governance is right, then AI is simply making the process faster, easier and fairer. And Kate's point was very valid. When done right, with the support of occupational psychologists, this actually makes for a much better candidate experience.

48:54And that's something I've been pushing for for a very long time. So these are the three takeaways for leaders thinking about how they hire. Lee. Number one, your hiring process is only as good as your job analysis. I'll say that again, it bears repeating. Your hiring process is only as good as your job analysis. Before you worry about which tool to use, AI or otherwise, ask yourself, do we actually know what we're hiring for? Not the job title, not the vague list of requirements on the job description. what does this person actually do all day and what does it take for them to do it well? If you can't answer that clearly, no tool in the world will save you.

49:35Yes, and lesson two, bad hiring isn't just expensive, it's unfair. Kate made this point that we've romanticised the old way. The reality was that CVs were screened in 20 seconds by someone who'd lost enthusiasm for the pile by number six. At least AI, when it's done properly, applies the same rigor to every single candidate. If you're a small business and full AI hiring isn't accessible yet, even a 30-minute conversation with a friendly occupational psychologist, hint, hint, my co-host is one of those, can get you 80 % of the gain for just 20 % of the effort. And lesson three, the loudest voice in the room is probably the least reliable.

50:13So whether it's a hiring manager who just knows who they want or someone dominating the debrief after an interview, Kate's research is pretty clear unstructured decisions default to bias you don't need an AI system to fix that you just need some structure some agreed criteria and a process that stops the loudest voice from winning by default you can find Kate and the team at sapia.ai there are brilliant white papers on there free webinars and a contact form if you want to explore whether it's right for your business Kate is also very active on LinkedIn search Kate Young at sapia and your finder all the links as always are in the show notes that's all for today so go and write down what your next hire actually needs to do not just be this is truth lies and work we'll see you next week