The Rise of AI and Voice Capture

Tullio Siragusa (00:10):

Good day everyone. This is Tullio Siragusa. I'm your host of Tech Leaders Unplugged. Today I am getting unplugged with Mathew Mozer, who is the CTO of HireTalk. Hi, Matthew, welcome to the show. Good to have you.

Mathew Mozer (00:26):

Thanks for having me.

Tullio Siragusa (00:28):

We're talking about the rise of AI and voice capture, specifically in the context of using technology to redefine candidate selection. So, we're going to dig into how AI and voice can be utilized or are being utilized to improve the ability to find the right candidate for the right job. But before we dig into that, let's find out how you got here. Tell us a little bit about yourself.

Mathew Mozer (00:56):

Well, I was working on a project involving a candidate selection and helping companies find people to do work for them. And one of the things I realized quickly was that the one-click apply just seemed kind of old in the sense that all you're doing is sending resumes to different companies and handing it across, you know, through email and whatnot. And I also realized that there's a lot of bias in emails knowing people's email addresses, knowing people's names, and things like that. And I said, there's got to be a better way here. At the time, it was 2022 when I started and started thinking about how AI can really help us out. And so I kind of imagined doing a higher talk.

Tullio Siragusa (01:50):

Okay. Great. Let's, let's dig in. Let's go right into it. Especially today some job postings are getting hundreds, if not thousands of applications, and companies have invested in ATS systems, and applicant tracking systems to try to, weed out those that are not the right fit. However, you know, let's be realistic. The AT system's looking for keywords, and you might've used different words because maybe you speak differently than what you know, the ATS has been programmed to do. And so sometimes really good candidates don't even get presented. They, because the machine doesn't really know that you know, might maybe use a different tonality, et cetera. So how does AI help with that? How do, you know, are you proposing that the whole system is just a waste? What's, what's the approach that would work better? Tell us a little bit about that.

Mathew Mozer (02:45):

Yeah, when I approached the idea, I kind of thought I need to step back and think about why were resumes created. And I can imagine a long time ago that there was a point in time when somebody said, I have way too many candidates. I have hundreds and hundreds of people to interview and I can't possibly interview everybody. So how about everybody write down what they're good at, what jobs they did previously, what's their education? And I can quickly go through within a couple of seconds and weed out as many people as they possibly can. And with that, I think ATS kind of does the same thing. A lot of modern systems kind of do the same thing. They're just doing, the thing that a human would do and, and look over the interview or look over the resume very quickly. And I said, there's got to be a better way. So how can we, let's say, interview everybody if, if you get a thousand candidates, how could you possibly interview everybody and not actually care what the resume is? Cause as we all know not all resumes are a hundred percent accurate and truthful. So using something like AI and, and getting people directly to the interview gives people a much more even chance to get into the proper candidate pool.

Tullio Siragusa (03:53):

All right, so let's talk about a couple of different things. I, I got some notes here about the value of applying AI in terms of candidate selection and qualification. And one of these stood out for me and I, I thought, wow, this could be really cool, or it could be an absolute disaster. Let's see if we can dig into it a little bit. Like an AI-powered voice, an analysis could help identify and assess a candidate. And it says here that voice characteristics such as tone pitch and speaking patterns could tell something about the person's ability to have confidence, their communication skills, and even their emotional intelligence, and provide insights about the suitability of the role. Now, when I look at this on paper, it looks great, great idea. However, in reality, there's also the fact that there are many different accents, and waves of speaking. And depending on how the machine's been programmed, it could have a lot of inherited bias, right? If someone doesn't sound like this, well, they're not the right fit. So what are your thoughts there? I mean, there's, there are some ethical risks with that too, aren't they?

Mathew Mozer (05:07):

Yeah, I think you, you nailed it right on the head. I mean, it sounds great right at the very beginning, but at the end of the day, there are just so many different ways that can kind of go the wrong way. And you have a lot more explaining to do about why you decided to like not to choose that candidate. So we, we kind of not going that route. We think that it's, it's best to still give people the ability to just use their voice so they can naturally answer. But also knowing that the tonality and things like that aren't going to be affected. People tend to be probably less worried about it and they're less nervous, let's say, so they can actually provide a better answer. Giving people the ability also to, let's say, rerecord and, and be more comfortable with it, I think gives a better picture of who the candidate is. And then also not using the voice itself gives people it kind of removes that bias. So we have a system that just kind of decides that they had a great answer because their context was extremely accurate as opposed to, oh, they didn't pronounce words correctly. But using ai, it, it kind of can fill in the blanks as needed to make sure that the proper picture is being painted, not that they said the correct words.

Tullio Siragusa (06:26):

Interesting. So, I mean, obviously, the goal here is to simplify the process of how a candidate gets matched to a job opening, but also eliminate the bias. There's a lot of bias around how a resume is written, how it's presented, and certainly if it's using voice and analyzing tonality and things of that sort, it could also have significant bias too, depending on how it's partnered. Not, not that it's intentional, it just is what it is. You know, the machine doesn't have any ego or emotion around, they set it this way, they speak this way. They just kind of take the programming and try to replicate it as best as possible. So, so how could this technology, so what you're saying is you use it, how does it work? Well, walk us through it. Like, I, I have a job opening someone applies. How do they, what, what do they go through? What's the process like?

Mathew Mozer (07:21):

So, there are a couple of different ways that we're handling it. We're letting companies have their own flow, have their own explanation of what they think the answer should be, rather than letting the system decide what is the expected answer. The company gets to decide what the expected answer is, allowing a little bit more control over what is a good answer or not. And so this gives more of a company fit type scenario with all the questions because we know exactly what the company's looking for. And I'm sorry, what was the rest of the question?

Tullio Siragusa (07:57):

Well, I'm just curious, like, what's the process? You know, someone has a job out and how does the system work? How does the candidate engage with it? Walk us through, the whole workflow. Sure. How does it, how does it come together?

Mathew Mozer (08:12):

So once the candidate gets to, let's say, the interview flow, whether it's through an invite link or, or through an email and whatnot all they, they put in their name and they put it in their email. But we don't ask for a whole lot more information until towards the end of the interview because we, want them to feel comfortable and not think about whether or not we care about certain aspects of their, of their life. More just let's go ahead and answer the questions. So as they're answering the questions in the background, the systems, analyzing them, and then we get the, the system gives a score to the, to the company. And then based on that score, the person with the highest score is a better candidate. So we then give them what's called a mango score. And this gives it a little bit more fun for companies, but it also doesn't necessarily say, oh, this is a percentage of the correct answer. We just know that this person had a much better answer than the others. But then we could also use AI to determine that, why this wasn't a good answer. And then we can also give proper feedback to the candidate to say whether they, depending on their score, can give them some, feedback on their interview once the interview's over. So it gives companies the ability at the end of the day to give proper feedback rather than kind of the normal I don't even know how the interview went, so.

Tullio Siragusa (09:34):

All right. So, kind of gets a link to get through basically an interview. The AI platforms, analyze the answers, matching it up to who's given the better answer. But how's that set up? Like what produces, what triggers the link to the candidate? How do you go about initially qualifying whom to send the link to? Does AI send it to everybody who replies? How do you program it to know the typical things that, you know, every company's different, right? Sure. Different job is diff So how do, how does that programming happen? Is this something automatically based on the job description, you just kind of feed it through the system and it learns. How, how does it come to you?

Mathew Mozer (10:12):

A great big question. So, as a company, I get to decide if I want to have a public interview process where I can post the link with like in a LinkedIn post or Indeed and whatnot, and people can interview right away. You can also attach it to an HTS system for everybody that interviews. You can say, Hey, you immediately are able to do an interview. And then at that point, they can go through the flow. So they have the option to keep it a public, private invite only and then pause it or close the interview process.

Tullio Siragusa (10:46):

Great. So, but, but let's say I say, okay, I want to deploy this technology in my recruiting effort, how do I I might have 10 different roles that I'm hiring for a big company, let's call it a manufacturing company. They need someone who operates the plant. They might need someone who's on the IT side, sure. Might need a driver. There are so many different roles they might have in a manufacturing environment, hundreds of different roles of which maybe there are 15 different specific unique jobs, right? Right. So, how do you, how does the system learn who's answering the right questions based on what is required? Is that a long process, of training it? How do, how do, how does that work?

Mathew Mozer (11:35):

So the way, the way it works is the company will create a campaign for that specific job and then that basically creates a candidate pool that when someone uses that link, they're able to go through the interview. The company creates questions and they can decide if, which questions go into what campaign, because they know, depending on whatever their role is, what kind of questions they want to ask. So they'll have the basic standard, tell me about yourself questions.

Tullio Siragusa (12:03):

So it's a human, the, there's a human interface. Someone actually programs the questions that need to be put in there. It doesn't, does the AI system then enhance those questions based on its learning? Or is it strictly, these are the 10 key questions you want to get through it? So it's the same for everybody. I'm curious how that works.

Mathew Mozer (12:23):

Yeah, so the, it's kind of mocking the same kind of interview flow where you know what questions you're going to ask in this particular interview. So you'll prepare those questions ahead of time along with an expected answer. Now we do have the ability to kind of generate an expected answer, but we would hope that companies would build off of that so that way it can be more fit for them. So, and then eventually we're coming out with the ability to that companies will put in their information about who they are, so that way we can help create the expected answer based on a company profile. We can do things like company fit, culture fit, and things like that based on who the company is. So, you can use AI to kind of learn who you are, and then the expected answer for each company is going to be different. The expected answer for each role is going to be different. And so that's why, but we want that human interaction because the, there shouldn't be kind of a lot of variation in the questions that are asked for different candidates, and that keeps it kind of fair for everybody.

Tullio Siragusa (13:26):

All right. So let's talk a little bit more about how to enhance this, right? I'm curious we talked about bias and removing bias in human errors and I'm always, I'm still concerned about the possibility that the questions that the person is asking are, might have some bias built into it, just because it's their own version of what the questions are and their own version of what the answer should be, right?

Mathew Mozer (13:51):

Sure.

Tullio Siragusa (13:52):

Are there multiple inputs that allow for continuous learning and adaptation so that the system can, can, you know, determine that? Yeah, there might be a different way to answer that question. And both, and those three different ways of answering are still very good approaches to the same problem. I'm just curious how you get that continuous learning and improvement in place so that it really creates an environment where it helps even the interviewer level up the quality of the questions that they're asking. Or is that a future state initiative on, on your, on your end?

Mathew Mozer (14:30):

Yeah, so currently we don't have our own model that we're using. So when we're doing a semantic analysis to compare the expected answer and the, the spoken answer from the candidate, it's just basically based on context. And so, when you like the removal of the bias is that because it's on a level playing field, that we're not having it learn off of whatever the company decides or having whatever candidate's answer, it's, it's going to be straight even down the board you know, down the flow.

Tullio Siragusa (15:05):

Yeah. So you know how it is, Mathew, the many, you learn that you could do something next and you're like, what else can we do? What's next? How much more can I automate it? So I want to go back to the fact that, you know, here's a comment from someone on LinkedIn, wow, this is going to change the game for employers like us recruiting and interviewing is such a time-consuming process. So anyway, to streamline it will be huge help. So, a great comment. And the targeted value for this is it to enhance the candidate experience or is it really to just help get through hundreds of applications in a very streamlined way, but also make sure you're not losing out on the top-quality candidates because maybe an ATS system is not smart enough to figure out that they use different words and don't match up to your keywords. Where, where are you guys focused on right now? What's the main problem you're solving? Is it, the, you know, the comment related or is it the employee experience, or is it both?

Mathew Mozer (16:08):

So initially, like our current MVP we want companies to be able to use this to bring in more candidates, to give a fairer chance to candidates. But we realize that this can be a thing that candidates aren't really a fan of. And we know that we just have to go through sometimes the motions as candidates to like to get through the interview process. But we, we actually have in our works that we realize if a candidate has already answered this question in a certain way and we're already re you know, keeping track of that and we're already keeping track of the candidate, we're going to build a candidate portal that's going to allow them to go back and kind of review their answers and adjust them, but not for the interviews they've already done, but for future interviews. And then based on ai using ai, we can use all of their answers they've already done plus eventually we're going to be able to link them to companies that don't even know this candidate exists and vice versa. So by knowing everything we know about the company and everything we know about the candidate, we're able to kind of match them without them even knowing giving them a much broader candidate pool and then telling them, Hey, here's Joe. He already is a great candidate for you because he is already done really well in your campaign. The other thing that does for candidates is sometimes you might really like that you're a software engineer, but based on your answers, you've kind of answered it more of software like a manager or a leader, but you're not applying for those roles. So we can actually take the role out of it, use your answers, and, and use it as kind of a skill assessment and kind of place them with companies and positions that they didn't even know that they would be good at or they would like.

Tullio Siragusa (17:56):

I love that. So this is what I've learned so far. Let's see if I got to recap it, right? And again, I'm thinking about it the way it's done today, right? You got to put a job out, and then someone applies, and then there's a link that's triggered based on the person qualifying because some ATS system decided that you qualified. What I'm hearing is everyone can get a link. You don't have to limit who gets a link. You, if 400 people apply, 400 people could do an interview because it's not being done manually by the individual. It's being done Sure. By the AI system. Now some people get through it, so there's, there's an ATS built into the system that basically puts someone through an interview, and then two things happen. They either qualify or they don't. Right? And then there's a stacking that determines who's better qualified than others. But also it creates a scenario where you can identify some additional talent. I e you know, have to do some kind of an assessment. You know, usually, these are multiple steps, right? Yeah. So it's built in the ability to say, Hey, this person actually has leadership skills, or this person might be good at architect or solutions engineer or something else, right? So Right. What I'm hearing is it eliminates the need for an ATS basically, right? Yeah. Because every candidate who applies, gets an interview, right? Yeah. So, democratizing the whole process of applying for a job gives everybody an equal opportunity. It eliminates the need to do a follow-up assessment, whether it's a briar, disc, or whatever people use, right? Yeah. Which a lot of people don't like doing anyway. Sure. Cause it gives you the ability to identify some ancillary skills or qualities that the individual has. And so in essence, by the time the recruiting team or the hiring team sees the results, they're going to see the best-ranked individuals and they're going to see some insights and intelligence about who they are potentially for other roles in the organization. I may not have even been publicized, am I on the right track here? What, what, what have I missed? Anything?

Mathew Mozer (19:56):

Yeah. And that's, that's really great for the company also, but for the candidate, it also gives them a little bit more insight too about what roles they could be looking for. Maybe you didn't realize that you would be much better as, some sort of architect or whatnot. So being able to kind of be assessed by the system to give them a little bit more like broad range scope of applications to apply for, jobs to apply for in different roles where some people get really in their silo of, I'm a software engineer, I'm a project manager without realizing that they would be probably really good for other roles.

Tullio Siragusa (20:30):

It sounds to me Mathew, that this could also be used for your existing team to do some kind of what do you want to do next? Or what's the better role for your kind of assessment? Is that fair to make the assumption that this could also be used for internal promotional purposes or internal realignment purposes, especially for organizations that are doing changes or reorganizations? This, also removes the bias, right? Yeah. In going through that is that a plan eventually to just make this the default way of just assessing talent, whether they're a new hire, or an existing one moving into another role? What's your thought process there?

Mathew Mozer (21:10):

Yeah, sure. Internal surveys also to find out how people in different positions feel about certain topics. And then assessing their skills where they've, what they've done in the company and where they could go based on their, you know, previous tasks, things like that can all be done in the system as, you know, as part of the, the interview process. So, you can basically interview your own employees, for different things. Not necessarily for a job, but for maybe even certain tasks that need to get done that are large. But we need to find somebody with leadership capabilities.

Mathew Mozer (21:46):

I think you've got a winner there, Mathew. We've gotten some engagement. We had another comment here from Brendan Roman on LinkedIn. It says, based on what I've heard, it's using AI to create a more humanizing and people-focused screening process. The beautiful irony, and very promising signing product service. So that's amazing. And I mean, that's ultimately what it is, right? It's just democratizing the way that people get access to opportunities. Exactly. And creating equal opportunities, really equal opportunities for everyone without having some ATS system, you know, knock 'em out of the process. Right. So you know, we're wrapping up any last-minute words of wisdom of putting this together. You probably got a lot of different feedback on how to do this, you got some today. What, what how do you, are you focusing your attention here?

Mathew Mozer (22:39):

Our focus is really going to be making sure that, that it's fair and that companies like give everybody a fair shake, but also giving people opportunities. They did then they didn't normally have people with, maybe they didn't have a degree, but they're still able to prove their skillset. Maybe they didn't meet certain criteria of the job description, but they still answered all of the questions correctly. So giving without knowing the background and giving everybody a fair share right from the beginning, and then giving companies the ability to assess that without, you know, spending a ton of money on, on and time, just interviewing them, I think is how we're handling it.

Tullio Siragusa (23:18):

Mathew, it's been great to have you. I think this is a fantastic idea and platform. The idea of giving everybody an equal shot at an opportunity is, is so compelling. Whether a thousand people applied or 10 people applied really doesn't matter. Just send everybody a link, let 'em go through it. Maybe their resume wasn't that great. Maybe they're not good at writing resumes. Maybe they didn't spend $1,500 with resume consultants to write their resume, whatever the reason might be. There's some great talent out there that gets overlooked just because they don't have the right piece of paper presenting them. You know, especially, for folks who are not in a marketing role, right? They may not, may not know how to market themselves. I love the idea that everyone can get a link, they can get interviewed, they have a fair shot, and then the system stacks 'em based on actual input. Not some piece of paper that may not have been written perfectly or not. So great to have you with us. Stay with me as we go off there in just a second. Sure. Thank you so much. Great. Thank you. We got another guest coming tomorrow. Patrick Cason, who's the CTO of Raise Financial Thursday. We're going to have Charu Goel, who's the senior technology leader at the Bank of America. So looking forward to speaking with them. Keep your eyes open for what the topic is. They'll be posted wherever you're following us on social media. You can also check out techleadersunplugged.com. There you will find all of our upcoming guests, all of our previous interviews, as well as blogs around every interview that we do with some additional insights and thoughts. Thanks for being with me and we'll see you again tomorrow at 9:30 AM Pacific.