Episode Transcript
[00:00:00] Good morning everybody. Today I want to talk about the wild world of AI fueled recruiting. So this is a huge topic for many reasons. The biggest reason is that recruiting is a massive part of our lives. 20 to 30% of Americans change jobs every year, even during the worst recessions. So the talent acquisition or recruiting process is a huge industry. It's always going on. Look at how many ads you find on the Internet for indeed, you can just sense there's a massive market here for young people. Of course they're looking for jobs quite frequently early in their career. And then the rest of us are job seeking periodically as we get older or change companies, or our companies go out of business or we get laid off. And on the HR side there's a big arms race of tools and systems to help find people, screen people, select people, interview people, qualify people, and, and hire them. In fact, we've done a lot of benchmarking on talent acquisition over the years. And the cost of hiring one person could be in the hundreds of dollars for a very high volume, low skilled job to five to $10,000 or more for a highly skilled person. In fact, if you want to go hire a really senior software engineer or a leader, you're going to have to pay almost a third to a half of their first year salary to get an executive recruiter to help you. So, and if you hire the wrong person, not only have you wasted a lot of time and money on the hiring process, but now you have somebody in the company that's not working out and you have to figure out what to do about that. So it's a big, high risk, high stakes, high cost endeavor. Big companies have very large talent acquisition teams. They have hundreds of recruiters, many of them are on contract. And the HR profession, a very high percentage of it is recruiting type of roles, including just people who schedule interviews and do background checking and all those kinds of things. Now AI has been around a long time in recruiting. It started in the days of Monster.com, which is now part of Raidency by the way, which was one of the early tools for what we call job boards. A job board was a place you could post a job so people could find it. It used to be impossible to find open positions because there was no Internet. You could, you had to call people or send them letters and apply for a job. And there were job ads in the newspaper and in magazines. And so these companies like career builder and Monster.com moved the job advertising space from print to online. Then a enormous amount of innovation entered the Market with search tools and more and more intelligence to help job seekers and job hiring managers to make match their candidates to their positions. And this is sort of where LinkedIn entered the market. LinkedIn was originally a social network and they didn't actually know how to monetize it at all. It was, it was early days in the Internet and Reid Hoffman and a couple of guys got together and built this network. And it was very useful because it was unlike Facebook, where you actually could have a reasonable facsimile of your resume online, people could find you. And then they realized that if they monetized jobs, they can make billions of dollars and a very high percentage, certainly way more than half of LinkedIn's revenue is from recruiting. Okay, fast forward another couple of decades and I think one of the, and by the way, the other part of the market is there's an assessment industry around this of some highly validated and some just hacked up tools to try to assess whether you as a candidate fit a job, including job previews, job testing, simulations, technical testing, and other screening tools to try to figure out if you fit. I've done a lot of research on that over the years. It's often called pre hire assessment. And the really good companies that do this are very sophisticated. SHL is one and there's a lot of others. And they literally study the performance and characteristics of the best workers. And then they go back and they find a test or they build a test to determine who fits those criteria. Okay, so you would think AI would be a big part of this, and it is, but it's been very wacky. I think the first thing that happened was the applicant tracking software companies. These are the systems that store incoming job applications, built matching tools that would score your resume against a job description. And it was mostly just word matching. You know, did the words seem to match and did what you said on your resume seem to match what was stated in the job? And did you have the right educational background or licensing background or other things?
[00:04:43] And that was great. You know, ATSs all had them and there was a parsing engine that people used to talk a lot about that went through and found the words and characterized them for you. And those were tools that were somewhat public sourced and they weren't really AI, although they might have been. I'm not sure what they were, but they used pre AI. Bert, it was called technology to do that. And then the company that I think kind of changed a lot of it was a company called Eightfold. Eightfold was a very pioneering Young Co. Maybe 10, 15 years ago that said we're going to take all of the profiles of individuals on the Internet, we're going to scrape it from lots of sources, we're going to anonymize it, and we're going to study the skills and technical information that we, that we can assess that describes the people against the jobs so that we could infer skills. And what Eightfold promoted was what we named talent Intelligence, which was looking at a large population of people and determining using AI, what jobs they would be most suitable for. And it was very, very groundbreaking. That idea was much bigger than the parsing that was done in atss. And Eightfold took off and was very successful in its early days. It was a little bit expensive and a little bit hard to understand, but a lot of people bought it and still do. And the AI under the covers of Eightfold was proprietary. There was no off the shelf LLMs at all. So nobody really knew how it worked. But There were some PhD papers on it and a lot of other companies started to do this. Isims I talked to a lot of vendors in the early days who were trying to copy what Eightfold was doing. And it was useful for finding people who were very unique and were hard to find. It was useful for screening, it was useful for determining the capabilities and skills you already had in the company for finding internal candidates, because you could take all of the employees in your company and you could put them into Eightfold and you can still do this and it'll show you career paths and opportunities for people in the company. So it had many, many applications. And as soon as they did that and the skills idea picked up speed, there were a whole bunch of other vendors that tried to copy it, and these ended up becoming skills technology companies. Skyhive, there was Lightcast, which was originally mz, there was Tech Wolf and many, many others. All of gloat fuel 50. All of the talent management companies realized they needed some kind of skills inference technology, and they either built it or bought it. I think this is one of the reasons that Workday acquired, Hired, Score and this scoring stuff got very sophisticated, but nobody knew exactly how it worked, how or whether it was biased. And a lot of articles and legal opinions were written. And then the New York State and Illinois and a few other places around the world passed laws that said that you have to prove that your scoring or inference technology is not biased. And that's yet to be proven in a lot of cases because there's two lawsuits out there, there's one against Eightfold and one against workday that are potentially big validating or trying to prove that these are biased systems. And then along came ChatGPT.
[00:07:56] Now AI is a commodity, everybody has it, anybody can use it. And we of course have been heavy into this and we found very quickly that you could take a body of resumes or profiles of employees and you could assess their fit against the job using any of the LLMs very, very well. Because the LLMs are word matching systems and they can match words and phrases extremely well and in a very sophisticated way. And you can train them by telling them what kinds of words or characteristics or experience you're looking for. So all of a sudden this very expensive stuff that was kind of hard to get seemed like was available off the shelf. And within a year or two hundreds of tools and websites and startups started to offer AI based recruiting and matching systems for companies which of course affected the job seeking population. So all the job seekers realized that they were getting scored by AI and some intelligent young people built the opposite. They built tools that allow you as a job seeker to take your background, bio, resume history, job experience, et cetera and stick it into their system and they will fix it all up for you and send it to hundreds of these recruiting systems. So now we have AI creating resumes and producing and sending them out and AI consuming them. So we have slop talking to slop. And I don't think it's completely slop, but it's pretty sloppy because the recruiters and most of you in town acquisition know this constantly tell us that they're getting so flooded with fraudulent resumes that they really can't sort through it all. We just posted a job on our, on LinkedIn for a sales job and we got, you know, a lot of candidates, the AI and LinkedIn summarizes the candidates one at a time. And when you read the AI summaries they're meaningless, they don't tell you anything. Now I'm not saying LinkedIn doesn't know what they're doing, but the AI in general isn't very good at really decoding complex multi step issues in a resume. I mean theoretically what Eightfold was trying to do, and I think they do it, but most of these other tools don't. You would look at a resume, you would say this guy worked at Google in 2015 or 2010 in this particular group. Who else worked in that group? What was the success of their projects? How, how well did their trajectory go in their careers? What technologies did they use? And you could Learn all sorts of things about this person because of where he or she worked and who he worked with and who's he's connected to and stuff these off the shelf tools don't do any of that. Seekout was another one that was very successful and still is at this. So anyway, so we've got the war for sourcing and selection and screening on the employer side and then the war for job seeking on the candidate side.
[00:10:45] And I don't think anybody's happy. The New York Times loves to publish articles about this, how hard it is for young people to find jobs. And if you really think as a job seeker you're going to find a job by submitting your resume to one of these job sites, you're crazy, because that's not really the way it works. You still have to talk to somebody in the company. And then this huge article came out in the Wall Street Journal last weekend that the North Koreans have managed to put together a massive army of fraudulent software engineers who apply to these jobs, do video interviews, accept the jobs, do some of the coding, take the money and give it back to the North Korean government. And these banks and insurance companies and other people hire these engineers, never knowing that they're not even American citizens and that they're not even the person they claim to be. So it's a pretty messed up space. And it's even worse than that. This weekend I read a very detailed article which is worth reading. I'll link to it by some scientists, computer scientists, who looked at hundreds and hundreds of resumes and found that if your job search tool is using ChatGPT and you use ChatGPT to mess up or fix up or spruce up your resume, your search tool is 50 to 75% more likely to accept you as a candidate than someone who did it by hand or someone who used Claude or Gemini or another tool.
[00:12:06] So the AI off the shelf recruiting tools or selection tools are biased in favor of resumes that were written using their technology and their language model, because apparently they recognize the formations and the embeddings of the words that come from their own technology. So that's really bizarre because it means that whatever you buy as a company to select or source candidates is biased almost by design. And then you have this issue, which nobody likes to talk about, that the AI itself really isn't intelligent anyway. All it's doing is matching words. So if the person looking for a job is not a good writer or doesn't like to write or doesn't think about using AI, they're going to have a very hard time getting a job because the AI won't even see them in a sense where they might be the best candidate of all, because many jobs don't require writing, you know, they're, they require other skills. So why do we have to be a guru of AI to find a job when the job itself doesn't even use it? So it's a big chaotic mess. Now there's some great companies in this market, HireVue, Maki People, Workday Paradox, Smart Recruiters, now owned by SAP. I mean, there's some really sophisticated software companies here and it makes sense that there are because it's a giant market and a very, very important market. And the technology that's used for recruiting is also used for matching and sourcing and succession management and career development and development planning for employees in the company. One of the big trends over the last decade or two has been reskilling or talent mobility or talent marketplaces of internal candidates where if you work for Amazon and Amazon wants to keep you, they will train you to become a software engineer or a security engineer or something else, because they want to be good for you and good for the economy. So they want to know what your skills are and what your aptitudes are most likely to be that uses the same technology.
[00:14:11] However, if you've done recruiting or looked for a job, which I think everybody who listens to this podcast has, you know that the real information or intelligence about a job and a company and a manager and a role is very subtle. I mean, I used to when I was younger, I haven't applied for a job a long time. You would get on a plane or get in a car or get on the phone and you would visit a company and you'd stand in the waiting room and you'd be let in and you'd have a face to face interview. And all of that time you were picking up signals about the company from the way people walked around and how happy they were and what was going on inside the company and who you met and how they treated you. And you don't get any of that through this AI stuff. Zero. So there's no subtle clues, there's no human connections.
[00:15:02] If you do a video interviewing, presumably the video could help you as a recruiter assess a candidate's ability to speak clearly and communicate and perhaps their honesty and. But the other way around, you don't see much. So we've kind of lost the human connection between candidate and job seek and employer. So the AI industry is Very early on this. Now I don't know where it's going to go because the skills assessment part of AI is very, very good. We've done a lot of experiments with this and Galileo is very good at it because it's loaded with a massive skills library from Lightcast and it knows what the words and the skills are. For example, one of the bodies of knowledge in Galileo is the skills, the official skills library for airlines. So it knows in its LLM and its training what the skills are for a flight attendant or a baggage handler or front desk employee at an airline. And so if you applied for a job and you asked Galileo to assess you, it would tell the recruiter a lot of specifics about what might be missing in this person's resume and then generate a behavioral interview for the recruiter to ask questions, to try to figure out whether that stuff is true or not. And these AI tools are very good at building behavioral interviews, coming up with good questions and even analyzing the answers. But if they really are biased to answers that come from their technology, you know, we've got work to do. Where is this all going to go? Well, I can't really tell, to be honest. I would have thought by now that it would have been better because the next big thing that everybody's been doing is building AI interviewers voice and animated characters or animated videos that interview you as if it's a human being. And even the most sophisticated ones, like from companies like Eightfold or Hiredvue or, you know, some of the very advanced ones are mostly only used for screening, not really final assessment. And it's not clear to me if job candidates want to be interviewed by a video interviewer or not. I mean, I know that one of the benefits of online AI based interviews is that the job seeker can apply in the middle of the night or off shift or over the weekends and they don't have to wait for a scheduled interview, that's a huge benefit, huge advantage. But if the quality of the output isn't good and you never hear back, or you can't call back and say, why didn't I get hired? Because there's nobody to call, you know, it may not turn out to be as useful as we hope.
[00:17:42] And there's this never ending quest to make this process better.
[00:17:47] And what we're learning, and what we probably learned this long ago, is that the process that a company uses to hire varies depending on the level or type of job that the company's hiring for. If it's a relatively low skilled entry Level or front office or back office job that we call those Type 1, Type 2. You can go through a pretty quick process. Can the person lift enough weight? Can they show up on time? Can they meet the needs of the shift? Do they have the basic skills to perform the tasks? And then we can hire them and you know, if they don't work out, it's not a huge cost because it's a relatively low pay job. You get up into more highly technical jobs, certified skills, nurses, electricians, contractors, people that have to be licensed or have very specific skills. Now you have to test them. Now you have to validate that the skills are real and you don't want to lose them because they're hard to find and hard to hire. And then you go into really those, we call those type 3, type 4. And then you go to professional jobs like doctors, nurses, licensed airline pilots. Those are very, or even software engineers or salespeople or executives. Those jobs are even more important because if you get the wrong person, it affects many, many other things. And they're expensive and the cost of replacing them is very high and the cost of losing them is very high. So on the latter types of hires, companies will spend a lot of time, a lot of money and they'll do a lot of human recruiting. And then on the former they'll try to automate as much as they can. And I think the future of this is smarter and smarter IO industrial organizational psychology assessments built into the AI. This is what Mackie people's doing. This is what Paradox is now doing. And other companies where before you unleash the AI to go out and look for people, the AI or the consulting firm that works with the vendor or the vendor works with you to build a realistic assessment that really pertains to the job and the company and the culture and the role that you're looking for. That sounds like a bunch of extra time, but it really pays off because then you get really good candidates. And then of course the other, you know, glitch in this whole process is that we or those of you who are in HR aren't really the ones that have to hire people. We're doing it on behalf of a hiring manager. So the hiring manager has his or her biases and oh, I don't really like people that went to this school or I don't like people that worked at that company because they don't really understand what we do or, or I definitely want people that worked at this company. And you know, I always like people that do so and so. So that gets thrown in too. I don't know. I think we're kind of at this weird state where maybe AI fueled recruiting has become a hairball for most companies. There's some huge success stories in high volume recruiting, in IO assessment, in selection, in screening, in video interviewing, and then there's a massive amount of noise and junk out there. For those of you that are in small companies, I would try to use a trusted provider or a consultant to help you if you're doing a lot of hiring because you can easily get lost in the weeds of all of these little startup companies trying to do this for you. Even LinkedIn has problems. For those of you in big companies, it's worth spending the time and money on this area because hiring is the most important thing you do. I mean, it's the fuel of your whole business, is getting the right people. And what I've learned over the years is great talent acquisition professionals are some of the smartest, most savvy individuals in the company about what makes the company tick because they know who works out and who doesn't work out and what characterizes a great fit. They that's why contract recruiting can be good and bad. If you don't outsource recruiting to a great outsourcer, you may be outsour one of the most important things that you do. That's it for now. I'll keep you guys up to speed as new things come out in this area.
