What Your Data Knows (That You Don’t)

This month, visionary Shon Burton stops by Subject to Talent to discuss how companies are (or should be) leveraging AI to find and retain the skills they need to get work done. As founder of leading talent intelligence company HiringSolved, Shon has plenty of views on HR tech, from why AI in HR is so difficult to recruiting epiphanies found within their clients’ data. Listen in as Shon and host Bruce Morton demystify the hype around AI in today’s workforce solutions.

Subscribe
Shon Burton

Transcript

Bruce Morton: Welcome to Subject to Talent, brought to you by Allegis Global Solutions. Similar to you, we’re always trying to learn more. On this podcast, we speak to workforce and talent experts from around the world covering market trends, technology and our ever-evolving dynamic industry. 

Hi, I'm Bruce Morton, the new host of Allegis Global Solutions’ Subject to Talent podcast. And while I'm new to this hosting gig, I'm definitely not new to the talent industry having been in the game for over 40 years. I'm excited to be here today with our guest, Shon Burton. Shon is the founder of HiringSolved, a leading talent intelligent platform, and we're going to be discussing how companies are and should be using artificial intelligence to help them find and retain the very best talent. Welcome Shon, and thank you for joining us on the podcast today. 

Shon Burton: Hey Bruce, thanks for having me. Good to be here. 

Bruce: So as our regular listeners will know, we like to start off all of our podcasts hearing how our guests got to where they are today. So Shon, can you tell us how you came to be in the workforce solutions industry and the backstory on founding HiringSolved?  

Shon: Sure. Yeah, it's exciting. I actually, I was an engineer and didn't like recruiters very much. And then finally went through a cycle where I did like them because I realized they could negotiate for me, which was great. And then I got talked into, 10 years later or so I got talked into starting a recruiting firm by a great recruiter. And this was in San Francisco in the tech industry there. So we started recruiting for hard to find tech, particularly for customers like Google, Apple, some of the big tech companies and some of the early stage startups that became huge like Dropbox and others. Oh, Twitter. Twitter when they were six people, so we were doing that. 

And I was just shocked, as an engineer I was shocked at how hard that was and how hard we had to work to place people. And immediately started thinking about, well I should say after about a year of beating my head against the wall, started thinking about how can we automate some of this stuff? How can we make some of this process a little bit easier? Because the amount of hours it would take us to place an engineer at Google was just ridiculous. So, that's where the idea for HiringSolved came from. 

Bruce: And what year was that, was that aging you? 

Shon: Yeah, that was 2011. So the recruiting, I'd been doing recruiting for about two years by that time where I sort of was like wait a second, there's got to be a different way to do this. 

Bruce: Right. Well we'll dive back into HiringSolved a bit later on in the conversation, but the topic today is around the high programed artificial intelligence. Every day you see something online about AI or ML. There's a lot of hype around that, but not many people I think really have a deep understanding of what artificial intelligence actually is. So let me ask you a simple question to ask, but I don't know how easy it is to answer but simple for me to ask is, what can AI do and what can't it do? 

Shon: Oh wow, that's a great question. So what we have today in the sort of state of the art of AI, it can do things, some things that humans can do. For example image recognition or finding patterns in video, audio, documents, things like that, that previously only humans could really do. So examples of that are look at a resume. Let's say we've got a stack of 1,000 black and white resumes and I give them to a human and say, tell me which ones are male and which ones are female. No pictures, just black and white resumes printed out. So previously if we go back maybe 20 years, that was pretty hard for a computer to do and it wouldn't be very good at it. The best humans are looking at, they're great at pattern matching, they're great at looking at that snapshot and very quickly figuring out based on, essentially a pattern, whether that's a male or female and they'll put that in the pile. 

Where we are today is that artificial intelligence lets us do a bunch of stuff like that because we've been able to train it on lots of samples and there's lots of different techniques. So we think about, only give me the pictures of dogs, and I'm going to give you a million pictures. Or only show me the YouTube videos that show cats, or show me which emails are spam, or the diversity sort of resume idea that I just talked about, machines are actually quite good at doing this. What they can't do is what we call generalized artificial intelligence. So that's sort of your C3PO where it can just do anything. It could tie a shoe, it could play a game of chess. The same system could tell jokes, could fix a broken dishwasher, we don't have that. And just to be clear, nobody in the world has that. This is an unsolved problem as they say in AI, which is this general intelligence. 

Bruce: So is the word artificial even the right word for it? I mean it's in the lexicon now, but is it truly artificial? 

Shon: Well that's a great interesting question too. Because I think what AI is showing us in, particularly when we think about neural networks, what they're showing us is that maybe what this looks like is this is an underlying organizational principle of the universe. Like literally it doesn't matter, the techniques that we're using in the neural networks that were using, those structures happened to be based in software, but the melon on top of my head does a lot of the same stuff. So there's this idea of, for example of reinforcement learning. 

A depressed human brain has certain chemical changes that make it less good at reinforcement learning, it's less good at perceiving the pleasure that comes with doing the thing right. And so it's literally less good, a depressed human brain, at that type of learning. We have reinforcement learning in artificial neural networks that work the same way, but they tend not to get depressed because we can control that. But yeah, it is a good word for it. And there's the whole other debate of whether this or that thing is AI or not. But when we think of artificial neural networks, they earned that moniker because they really are a structure that is very similar to what's going on and what we think is going on in the human brain and animal brains. 

Bruce: Right. And obviously this has been around for a long time. I know it's always been called AI, now that I'm thinking back. And Watson was playing chess and beating the world masters, that is the artificial intelligence. I'm sure that the machine is understanding if joining the dots of 30 moves ahead the same way as a brain would there's been trained that way. So it's been around for so long, but why is AI so hard in the world of human resources? If it is, I mean do you feel that human resources are behind other industries? 

Shon: Yes, it's crushingly hard in human resources. Yes. If I had only known that Bruce before I went in. So human resource, what it is is neural networks are not good at explaining why they know something. So it's still today, if you look at the science around how someone actually catches a baseball, there's a lot of complexity on there and there's not a great description of how that really works at a really granular level. The problem, and computer neural networks are the same. There's kind of a saying in the industry that goes, the better the thing works, the less we understand why it works. So a lot of these artificial neural networks are very much black box things in the sense that we generally understand the structure, but unlike a sort of traditional software, deterministic software, it's very hard for us to trace through step-by-step and understand why the AI arrived at a decision. And that turns out, because HR is this heavily regulated industry, as is something like medicine where there are lawsuits and stuff for doing it wrong. It turns out that that's crushingly hard. 

I think aside from that when we use our example of resumes. So today for example, HiringSolved, this is something we've been working on since probably 2014. And it's just get a system that can reliably tell us based on no other information than sort of a black and white resume paper idea of a resume, whether a candidate or an applicant is male or female. We now have systems that do this better than humans. So of course the machine can do that in less than one second. A human would take quite a bit longer to do 1,000 resumes, the machine is going to be more accurate over any decent sample size than the human. Problem is that's not good enough in HR tech

So in other industries, like for example financial trading, we have AI-based decision systems that are making trades and doing all this crazy stuff at light speed, but the only thing we expect in finance is that they perform well. So at the end of the day we don't care why they thought what they thought, why they just dumped Bitcoin today and it went down 30% or 40%, we don't care what the answer is as long as they're performing well over a reasonable amount of time. In HR tech that's very much not the case. Because of the compliance in the industry it isn't good enough just to perform well, you have to perform well on a compliant way. Which turns out to be really, really different. 

Bruce: Yeah, well you made your bed and you got a lie in it I guess. You got into the industry. 

Shon: Yes. Yeah, it's full of spikes in glass, but yes. 

Bruce: Yeah. That's a nice segue into my next question in terms of defining terms and you're a leading organization in talent intelligence, so how do you define or describe what talent intelligence is? 

Shon: Well I look at it as the analog to business intelligence. So I look at it as, in the '80s and '90s we had built these database systems that contained all this data. We had accounting, and supply systems, supply chain, ERP, all this different stuff, but we didn't have much intelligence. We had systems that could tell us what the price of something was, but we didn't have systems that told us large-scale trends and could give us predictions based on those trends. Business intelligence in the IT world changed all that by networking all those systems together and pulling all that data together to bring a layer of intelligence above all that, that the business could look at and realize, gosh, we use this one part in 50 different planes. If we could save $0.10 on that part it would really create a lot of savings, for example. 

So talent intelligence, I think of as the same thing. It's really, there's a tremendous treasure trove of data in all these companies. It's the HRIS system that's telling us what type of people have been successful at the company, what does longevity look like at the company? How does that break down geographically? And by department, by hiring manager, all these different trends that are historical. And then we've got the sort of ATS and apply side, and who's applying, and who's being accepted, and how are they processing through interviews and things like that. And then we've got things like performance management systems and CRM systems to figure out all this other data. So talent intelligence is really sitting on top of all of that just in the same way as business intelligence was. And it's producing intelligence literally, it's distilling intelligence from all that information and lets us make recommendations and show patterns that we just didn't know were there. 

Bruce: That's exciting. And as you were saying all that I can imagine some listeners thinking, okay, that sounds like a lot of integrations, a lot of different systems. Is this one of those, here's the danger of the CIO putting it in the too hard bucket when you get into those types of conversations? The size of the prize sounds amazing, but how do you get across some of those challenges of having to plug in those different data sources? 

Shon: I mean it is a huge challenge. It's a huge part of the challenge just as it was with business intelligence back in the '80s and '90s. Yeah, it's brutal actually. We've seen a couple startups pretty well-funded, $10, $20 million in funding just pack up and go home largely because of the integration or compliance challenges. So it is challenging Bruce, but I always have said we are probably one of the world's foremost experts at getting data out of systems, we're pretty good at it. So that's one of the things that we focus a lot on is, and we actually find that's the very beginning stage of that intelligence. 

What we're seeing right now is that as soon as we get access to the first dataset, we're able to generate a data quality report, give it back to the customer. And before we've even done anything, they're not even in the HiringSolved system, what it does for them is it shows them how ready they are for automation. And we have these amazing findings like, well some company has 5 million records and they think, gosh, this is really valuable data. Once we analyze that data we give them this report and we realize, hey, 27% of those are missing emails. 35% of those millions of records are missing resumes, there's no attachment to them. 

And then as we step through this data and show them, that's super valuable because they're still collecting all this data every day, in some cases in a really poor-quality way. That's the other interesting thing about AI is it runs on data, it doesn't really work well in the absence of data. So if you've got millions of people applying but you're not good at collecting a resume for them, there's going to be some real limits to the type of automation you can do for example. 

Bruce: Right, right. That's a great point. And I love that line I've heard you use in the past saying, what does your data know that you don't? I think that's such a provocative and a great question. The data is there as you say, it's just getting access to it and making sense of it. I remember some years ago an IT software client of ours, we just did some manual tracking of who their superstars were in their sales team and who were the ones that they were wanting to get rid of after a year or were leaving. 

And amongst all the data we were collecting, there was this one organization, one of their competitors. If they hired people from that competitor that had less than two years’ experience, they were a superstar. If they hired people from the same organization that had been with that competitor more than seven years, they bombed and they left within a year. And when I presented that to the global head of sales it's like, "Wow, why is that? Do you care?" I said, "No, I don't." Exactly, just stop doing that and do more of that. And that's just one little nugget of, if you spend the time to do that the answers are there, right? The answer are within, it's just how do you get to that and what do you do with it? 

Shon: 100%. 

Bruce: Talking of which, could you just for our listeners to bring this to life a bit, just share some examples of how some of your clients that converted, the ones that have seen the light, how they're using talent intelligence to their advantage? 

Shon: Yeah, sure. I mean I'll give a couple anecdotes that are real world. One customer, very large staffing firm had a diversity initiative and was working for a large multinational company doing a ton of hiring. And the diversity initiative essentially said we need to be hiring more women and we need to see those percentages increase across the pipeline. So we want to see more women in interviews, we want to see more women in the pre-interview pipeline and all that kind of stuff. And ultimately what we really want to measure is a large gain in women hires across the board. They didn't have the intelligence available to kind of parse back to the customer and say, here's where this is realistic and not. And one of the major requirements because this customer had, won't say there any of these names, but they had let's just say a major company, two major companies that do a lot of warehousing. So one of the major hot recs was forklift operators, and they were beating their heads against the wall because they just couldn't achieve this required diversity mix, particularly around female forklift operators. 

So one of the things our system does is it can provide, you can run a search and you can actually, in our system you can actually talk to it like Siri. So you can say, find me forklift operators in Wichita, Kansas with five years’ experience or something. And it comes back and you say, how many are female? And it'll give you the percentage. And then you can say, how does this compare with the workforce? And from there it will go out and do a web search and try and figure out how your data, how you're apply data or your database numbers compared to the general workforce that we see out there on the web. 

What the system back with, and I did this live with them, what the system comes back with is about 4% of all forklift operators are female. And across the entire United States there's about 4% that are female. It's probably intuitive to us, but that was just a landmark moment to watch that team see that data presented. And it highlights a type of capability, this is where I think artificial intelligence is really interesting because manually, you couldn't do this computation. It would be grossly manpower excessive to sit there and try and analyze that, analyze the web. But we can do that in seconds and they can go back to their hiring manager to say, hey, here are these five positions, one of them is forklift operator, there's four more of them that we don't think we can hit the diversity on, or you're going to have to give us way more resources because we're fighting against the current here. So that's one example, other examples are even simpler than that. We helped Lowe's hire 70,000 people in 90 days. 

Bruce: Wow, 70,000 in 90 days. 

Shon: 70,000 in 90 days. This is pre-pandemic obviously, I think 2019. And one of the things that, they have very good data, Lowe's has amazing data and they're very artful in the way that they think about it. But one of the things our system was able to do was find and boost the relevance of anyone that had ever worked at other similar stores and get those people to the top of the list because they had similar experience. And that seems really simple and silly but again, sometimes artificial intelligence isn't really the sexy, cool stuff that's talking to you and making graphs. Sometimes it's something as simple as parsing through a bunch of again, black and white resumes and figuring out correctly, just parsing the data. Figuring out correctly who worked at some other hardware store that might be more relevant that we should bring to the top. 

Bruce: This got me thinking, do you think this is filling a gap that potentially should have been in the ATSs from the get-go? Or I guess, well when that intelligence is there it seems as if you're adding value and getting more value out of an ATS system, would that be correct? 

Shon: Yes, you're definitely adding value to the ATS. And I kind of look at it as going back to that business intelligence versus talent intelligence. There are some real technical reasons why the ATS isn't the tool to do this with. And it's similar to the business intelligence stuff, in business intelligence we have the data warehouse. And really what that was was a different technology that was much better at crunching data in a non-relational way. So the ATS is there. And the way I look at the ATS is ATS is a gold mine, but the goldmine never said it was easy to get the gold out of, that's not the gold mines job. The gold mines job is just to hold the gold and keep it safe, and it does a great job at that. 

The other interesting thing is that the ATS does a lot of stuff. It's the reason we've never become an ATS. These days it does a lot of stuff related to compliance and making sure that that data stays whole, and we can track it, and all that sort of stuff. But we have the flexibility as not a system of record, I think you want to keep those things separate. I think you want to have the flexibility to go in and say, we're going to access data across a bunch of containers and we're going to bring them together in a new way with a new type of technology. And I think that becomes sticky for the ATS because of what it is. 

Bruce: Yeah. And that's a great analogy that goldmine. You're right, it's there to protect it in a way and hide it. That's why ATSs were designed for compliance. And then we complain that they don't do the cool stuff that you can do. They were never built to do that, that's a great, great point. My understanding is you are not unique but slightly unusual for a software company that you serve both the staffing industry and organizations TA functions as well. Is it the same products, same conversations you have on both sides of that fence or is it coming at it from different angles? 

Shon: Absolutely the same products on both sides. They do come at it from completely different angles. I'll be honest and say we've really pivoted more, we were always into big corporations because we tend to solve problems. It's easier to see the benefit of our solutions if you're having a lot of scale, like the Lowe's example. But post-pandemic we're really focused on staffing firms from a sales perspective. We find that they move very, very fast. We find that they're very, very busy because of the shortage of talent that's out there. And at the end of the day we just make their bottom line fatter quicker. 

Bruce: Well they get paid for finding the needle in the haystack. 

Shon: Yeah, exactly. They get paid for doing that. And corporate, although that's exactly the same product, corporate uses actually completely different features. Staffing doesn't really get applies, staffing goes out and hunts, they're the hunters. So they don't get people just applying, so there's literally feature sets, even though they run the same software, staffing doesn't so much use some of our apply scoring stuff. They just don't use it because they don't get applies. Whereas corporate, in a lot of the big corporates they're just drowning in applies, especially when things go crazy with the pandemic and stuff like that. 

I remember when I was out of a job in the tech bust of 2002 or whatever, I was sitting there without a job. Yeah, I would apply to anything. I'm like yeah, I can cut wood. I'm a security engineer but I can do a little cooking or whatever you need, why not apply, right? What the heck, what's it cost to me to apply to all these different jobs? So yeah, we do see a lot of differences. We love the staffing companies right in the last year because they're moving very fast and they immediately see that increased bottom line with our software. 

Bruce: Yeah. Well that's great, that's great you've been able to pivot in a way to that market. Awesome. Well Shon, we're coming up to time here. So obviously I want to thank you for joining us today and making sense of what is sometimes a bit of a black hole, or these phrases people will throw around. So I think our audience are far more educated than they were 30 minutes ago in terms of the ins and outs of AI. So those folks that want to learn more about talent intelligence and the particular HiringSolved, how do we find you? 

Shon: Just go to HiringSolved.com and check us out. 

Bruce: Great. Thanks, Shon. And really appreciate your time today. 

Shon: Thanks so much, Bruce. 

Bruce: To learn more about AGS, please check us out at AllegisGlobalSolutions.com. You can also send questions for me or our guests. Just tweet us @Allegis Global with the #SubjectToTalent or email us at SubjectToTalent@AllegisGlobalSolutions.com. Until next time, cheers! 

Get Podcast Episodes Delivered Directly to Your Inbox