Why Focusing on Small, Risk Free AI Wins Accelerates Large Scale HR Transformation

Hi, I'm Jack, and this is the range podcast. Every episode is a conversation with someone in comp or total rewards who's actually building with AI. We talk about what it's doing to our work, what's working for them, and what they do differently if they start it again. Hello, hello, everyone, and welcome to another episode of the range podcast. I'm here with Joshua Levin, who is Senior Director of Total Rewards as Brazil, plus his co-founder at Slash Benefits. He's been in comp and benefits for over 20 years. He is a bit located in Minnesota, is that correct Josh? That's right. And he's here with us today to talk about AI, specifically AI adoption and rewards. What a surprise. Josh, why don't you take a second to tell us a little bit more about yourself and your journey? Yeah, so thank you for that introduction. I guess a couple of things that I would just say about my journey. I am very much of an AI enthusiast. I would say one of the earliest adopters, I think the first one-tenth of a percent of chat GPT, and I AI tools in a lot of my creative work, and that oftentimes extends beyond the total rewards work that I do. But it was not long after I started playing around with these applications that I learned that there was going to be very significant impacts in the work that we do in HR. And so I embraced it early, and here we are. And early adopter, indeed, I came across Josh on LinkedIn, actually, yes, because he's posting about his creations and his AI builds from simple task automations to more fully flesh tools. And that's how we met. Now, before we get into the whole AI thing, I wanted to ask you something more personal, Josh, that we know what your job is, day to day, right? And I'm sure you're a manful secrets. Now, I want to know, what's something you're surprisingly good at, given what you do for a living? You know, I will say if we're talking about little secrets or things that sometimes people don't always know about me, I love to make music. I have since I was a teenager. And one of the great things that I found with AI is that it's a great enabler to be able to make music. I've got three kids, a very busy family life and everything like that. But with the tools that are out there today, I can get that full band sound and, you know, express my creative interests in that way. And so I do quite a lot. In fact, if you find me on YouTube under lemon sour music, you will actually find a number of songs out there that I've created. Will I see you with the same microphone you have on the screen for those that are watching? There's a quite professional microphone that Josh has. But also, will I see you with long hair like the 70s rock and roll style here? If you see my AI edited videos, you might see me with some long hair. I think I've got a video on Instagram or something like that where I look like a rock star with long hair and leather and everything like that. You can tell that's a wig. Josh, that's okay. We're all on the same boat. Great. Great. So Josh, you've been experimenting with the AI for some time and you've been thinking about how to adopt this new capability, how to leverage it in our domain, which is full of surprises, right? Confidential data, personal and identifiable data, regulations that change in every jurisdiction. My first question to you, which is probably many of our listeners' mind is how did you get started? How did you lighten up the fire? Well, I will say the very first thing that I've done in order to kind of get involved with these different AI applications is I play around. I mean, of course, I just told you about how I use it in my music, but I've done certain things that are definitely not work related to be able to see how can I use these tools in a work related way. And honestly, playing around first in a safe space where, you know, I can't really mess something up when I'm making a recipe application, right? It helps when I decide that I want to then turn around and build an HR application that has, you know, a lot of those restrictions that you're talking about. So just your advice, you would be to start small with something that is risk-free, where, yeah, I mean, worst case scenario, your partner will complain about bad recipe that you come up with. And how did, like, there's one point where you went from, okay, I'm I'm using AI for personal matters. But at some point, I start using it for work. Can you recall what the first use case was? Because a lot of people ask, what are the realistic use cases, right? Right. You know, I probably started using AI very similar to how a lot of people start using AI in their work, you know, I want to write an email and I don't want to come across as frustrated as I feel right now. I want to be professional in the way that my approach is. So, you know, write out the email, pop it into co-pile that are chat GPT or your large language model of choice and say, make me sound professional and, you know, but still get my point across and it rewrites things so that you do carry the voice that you want to carry professionally with that email that you want to write. And that's probably how I started, just like other people might start. I quickly moved though into trying to dive a little bit deeper and see what kinds of what kinds of applications can I actually bring into my workspace. And the good thing about Residio is we actually have a generative AI review process that includes cybersecurity and it includes legal review for good reason. We want we have an intellectual property that we want to protect. We have, you know, personal information about our employees that we want to protect and we, you know, need to be compliant with the different AI rules that are, you know, applicable, like you said in different jurisdictions. So I started basically keeping the cybersecurity and the legal team busy by submitting a whole bunch of things for review. Over time, I got a little bit better at figuring out what kinds of things they were looking for. And so now I look for that when I'm looking at an AI tool that we want to review. Specifically, the types of things that I'm looking for are what are their safeguards around data privacy? Do they train on the data that I'm using and inputting into the models? Do they have a way to turn that off? Those are some of the things that I'm looking for. I'm also looking for, do they have a privacy policy? Do they have a privacy page that that talks about how they treat data? Because that's not just important for my company, but that's important for other companies that are using these tools. Excellent. Seems like you've made quite a lot of friends in the legal and IT department over your company. A lot of the reward leaders I speak to often ask, how do I make a solid business case for adopting AI rewards? Some of the struggles, like we tend to be small teams. We manage millions and millions in the resources and dollars of employee paid data, but sometimes there are departments that are bigger than us and get a little more attention. I wonder if you had to do anything to get their attention. No way. Yes, you're a small department, but yet we're happy to work with you and maybe to prioritize your use cases. How do you make sure that your counterparts are in legal, maybe even HR leadership? Before they object, they understand the power that you can bring using AI and comp and benefits. Yeah, that's a great question. I think having regular conversations and demonstrating the value that you can bring to the table is really important. Sometimes that takes a little bit of trust, but the good thing is a lot of these applications are not very expensive. So asking somebody to take a small risk on something that you're trying to accomplish is usually a little bit easier when that small risk is a couple of hundred dollars, not something that's $30,000 a year, or multi-year subscription, sort of a tool. That's basically kind of the entry point that we've taken on my team is we've said, "Work and we end up making real gains with some small investments, and then we can demonstrate what's possible with these tools." We've done that in communications, making videos, podcasts, and other educational content for our employees. We're doing it with applications that we're building specifically for for our team, for the broader HR team, and possibly for the broader employee base, but it's those small gains and demonstrated sort of capabilities with these tools that show people what's really possible bringing them along the journey. - Excellent, and I think you've touched on a very important point, 'cause when we think about AI adoption in our domain, we often think about how can we be, how can AI make us a better version ourselves, a more productive version ourselves? Is a lot of the use cases I've heard from other compensation toward leaders are pretty much focused on enhancing your team's capabilities, upskilling them, and just doing the work either more quickly or better. But to your point, this is an opportunity for us, not just to upskill, AI is giving us an opportunity to going back to why our function exists in the first place, improve the allocation of resources and improve the employee experience, the manager experience, the employee engagement. So the natural question for someone that, like you has started, and this journey, and it has already deployed tools that take care of, maybe some bits and bobs that get automated, but also, you've got AI embedded in some of the, more like heavy workflows. How do you measure success? And also, if you've developed design tools for your audience that sits outside of tool rewards, how do you measure adoption by managers and employees? What are the tells? - Yeah, well, some of this is actually, those are great questions, and I can tell you how we are approaching this. I don't think we've got it all figured out, but what I will say is to answer a couple of those questions, in some cases, they're new capabilities. So we're really comparing against zero, right? The team never did this before, and now the team does this, right? Making videos to talk about things like the Secure 2.0 rules for the 401K plan that we're trying to educate our employees about we probably would have put a newsletter together and sent it out over email. Now we've got a more engaging piece of content that people can watch and they can understand. We never did that before, or if we did, it took resources that went beyond our team, right? We had to engage with our communications team to be able to try to partner with them and get on the schedule. So yes, we're doing some things that are faster, but we're also doing some things that are just completely new. When it comes to the applications that we build, a lot of the ability to see adoption is inherent in how we're building the tools. Most of the applications that we're building have like a backend administrator dashboard, and as part of that, we also do user management. So we're seeing specifically who has accessed the application and we can end up telling when they last use it, or we can end up getting information on how frequently they're using it and that sort of thing. Now that is a little bit of a tricky scenario, because in some cases, the applications that we build for certain purposes might need with that sort of rigorous audit log, might need a works council review like for example, in Europe just because of the possibility that that might be seen as something that might be performance based tracking somebody's usage of something over time, right? And so there are hurdles with that, but there are also capabilities that you can build directly into the applications that will help you gauge that utilization. Out of curiosity, what are the typical questions you're getting from the trade unions, the work unions of this world, like specifically in Europe? I ask because there's a lot of, I'm based in Europe, right? There's a lot of, I wouldn't call it necessary hesitancy or anxiety, but they're companies, they're deliberately playing the waiting game. They're saying, right, we're gonna wait until everything gets sort of dealt first from a regulatory to a certain point, but also we don't want the trade unions to think that we're gonna use this as an excuse to cut cost because we've seen it in, like a lot of the narratives coming from big corporates, especially in the US is, you know, we're investing in AI and look, now we don't need 10 of you, we just need five of you, right? Now whether that's just clever narrative of the reality, we don't know, but I'm curious to know more about, like if I was a reward leader that I'm adopting AI and I've got a huge presence in European countries, or heavily unionized think about Spain, France, Germany, like what are the typical questions I might get from a representative of the workers? - So I can tell you that the types of things that we know are concerns that we just try to avoid right now are any sort of AI decision-making tools that might, you know, decide whether or not somebody should be offered a job or something like that, we're not even playing in that space, right? If we are using something like one of the things that we built is a job matching tool that will basically take a job description and find out how it fits best into our job architecture. There's no personal information, there's no personal, you know, no decision point around a particular individual's job that is associated with that exercise. And so that's something that, you know, usually is free and clear because it's not impacting somebody's work. And that's primarily where we focused our applications. We've built applications that will look at a holiday calendar and understand the statutory holidays in a given country to pre-populate it and then give it over to somebody for further review to review it and we have a workflow and a dashboard and things like that. Pretty safe from, you know, I'd say suspicion or scrutiny from like a works council, for example. - Yeah, it's an interesting debate for sure. It's many different countries like adopting it differently. I think I read last week that Dave Band lay off due to AI adoption in China, which is quite a different move compared to what is happening in different parts of the world. But to your point earlier, like you're building tools for other humans. You're not building tools that replace humans. These tools you build are meant to be used by the same humans that someone might think you're trying to replace but there's actually other way around, right? You mentioned the job matching tool. Tell me a little bit more about that. - Yeah, so at Residio, we built a tool that will or an application that would take a job description and clean it up based on the, you know, best practices that we use internally for our job descriptions in terms of a job summary, the preferred qualifications, the, or they're required and preferred qualifications, skills, competencies, et cetera. And we have loaded on the back end information about our job architecture, our job families, the descriptions for those families, our job levels and the descriptions for those job levels are different career streams. And then it will take something that oftentimes managers have put together the best job description that they can, right, but they're, you know, doing this on a one-off basis when they have a need for talent. So it's not like they're doing this every day, right? So they take the job description and hand it over to an HR business partner or somebody in the talent team and they can then turn around, feed it into this tool and it will clean up the job description so that it meets our standards. And then it will also evaluate and stack rank kind of where it fits best within our job architecture, both from a job family and from a leveling standpoint. So it's really great in terms of helping a manager and like a talent acquisition specialist or an HR business partner have an intelligent conversation about what kind of role do they really need to hire for what they're saying needs to be done. And then if they want to, they can actually level up the job or level it down depending on how things play out and then they can have a conversation about, okay, is your need really this or this? Is it something that, you know, has these problem-solving responsibilities and accountability or is it something that maybe we can end up, you know, saving a little bit of money and giving them a little bit lower accountability that are aligned with our job description but properly leveled. I think it's really great in setting, you know, appropriate expectations for the levels of roles that different managers are hiring. - It sounds like you've made your manager's life easier, right? 'Cause this comes all on top of their regular day-to-day work and they are doing their job. They're also managing people. They're also, you know, involved in matching jobs half-count planning and all of that. You know how sometimes you You design a job and you agree on the job family mapping, the job level. But then you go to market. You meet some candidates, suddenly the business falls in love with one specific candidate. And they ask if there's any room of flexibility to adjust the job requirements to fit with this particular candidate. They're so desperate to onboard. Have you thought about a way of making a tool flexible enough to reopen certain sort of decisions, or go back to a few decision points based on how things can evolve? Or is it deliberately, is the tool deliberately built in a way to prevent that from happening? So I would say the tool is flexible, but the structure of the conversation is what probably would dictate that more than anything. In the case of the application itself, we have the ability to level that job up or down. So if somebody wants to go back and revisit a saved job description and see what would this look like if we were to make it a higher level role, they can do that. The great thing is it's not just changing like the years of experience. It's aligning the whole job description to the leveling guide that we have. So when we do change the level of the role, we have a clear understanding of how does that change the accountability and some of the other aspects of the job, which is really great and makes for a more robust conversation. Now, I will say if the conversation has been had upfront about the type of talent that's needed, based on how the conversation went initially, they probably aren't sourcing that candidate that is outside of the scope of the role because they've appropriately leveled the role to begin with and described it as such in the posting. But I guess there is a chance, you know, somebody knows somebody in their network or something like that. They get involved in the pool anyway and that conversation is possible. But like I said, I think that's more of a way the conversation goes not so much a way the application works sort of thing. And it also stresses the point that you can implement new technologies state of the art. You can build a lot of things with the eye, you know, whether they work or not. It will expedite things. It will extrapolate in size that you might have missed, but it's likely not going to fix any behavioral issues. That's why, you know, the human that has built the tool is still pretty much required in running all these processes and why having a human in the loop is oftentimes the best way to do this. The job matching tool, Josh, that's something you'd be willing to share like on the screen. I was going to say, do you want to, do you want to take a look at it? I'll go ahead and do a little show and tell show and tell. All right, so I've gone ahead and pulled it up. Let me know when you can see what I'm looking at right now is a job on Indeed. And so what we're going to do is we're going to just take a job. This is a software engineer that I found it's, you know, in the Twin Cities area. I don't know where Oceola Wisconsin is, but must be close enough that it came up. And I'm going to grab some of the things from the job description here. We'll go ahead and take up through the working conditions. And I'm going to take it into the Residio match tool. So this is the application that we built, and I'm going to go ahead and find a match. So we're going to see where this random software engineer role that I found on the internet fits in within the Residio job architecture. We do have software engineers, so this should be something that we can find. I'm going to just kick pop it in here. I could either upload it as a file if I had it as a word document or, you know, like I did with just copying pasting, I can paste it in here, and I'm going to hit that analyze. It's going to go ahead and look now at our job families, our career streams, and our job levels. And it's going to give me a stacked, ranked list of what it thinks fits best. So it's come up with a senior software engineer in our organization. This is a professional seven rule. It says it's a 90% match to that, some things that it says are key alignments and why it's a good fit, some things that might, you know, be a little bit different. It's also said that this could be maybe a little bit higher level rule, same job family, but a professional eight senior advanced. And then it's also said, hey, maybe it's a, maybe it's a slightly lower level rule. Looking at the job, I will say, I'm just kind of basing this based on the salary and everything like that. I'm not doing a deep dive because we're doing this live, but I probably take a little bit closer look to see what are the true responsibilities and everything like that. I'm going to say just based on the pricing and how I know these jobs. It's probably more like it's got some experience, but it's probably not our senior level just given the pricing on it. So I'm going to go ahead and refine with this professional six. It's an experienced, not an entry level, but not quite at that career level either. And as you can see here, it takes me into a screen where now it's populated stuff along the lines of our best practices. So I've got a job summary. I've got an outline of primary duties and responsibilities required and preferred experience, competencies and skills. And if I want to, I can edit any of these. I can just go into the fields and type something different or I can go ahead and add something. Let me just do something that's totally odd and just say this is, this is the team party planner. Obviously not typically a responsibility of a software engineer, right. But let's say, you know, I added that and it somehow got through, right? I'm going to go ahead and validate the alignment. Now it's going to take another look at this and it's going to say based on the changes that you made and everything like that, here are some suggestions that I'm going to make for you. It's saying that, you know, there is some good alignment, but as I look over here on the field analysis, it's going to give me some suggestions. And down here at the bottom, it says, hey, you know, you should really remove the informal line. This is the team party planner, you know, maybe consider adding a sentence about how it contributes to process improvements. It's that. So I'm going to go ahead and take that suggestion into consideration. I'm going to delete this. I'm going to validate the alignment again. And I'm going to have a cleaner validation now. It's pretty similar. I didn't change a whole lot with the job right in terms of the percentage match. But down here, the, you know, suggestions, you can see that it's, you know, different than what I had put in previously. And it's giving me different suggestions. Now I can continue to find to him this if I want to, but at a certain point, I might say, well, 92.5% is good enough for me. I can go ahead and finalize this if I want. Now I'm not going to use this because I'm not going to finalize this right now. But let's say I want to, you know, get some feedback from the manager. I can actually export it as a Word document or I can copy and paste it into an email. And I have the ability to, you know, take it somewhere else after I've done that. Gosh, I've got so many questions. Just this is, this looks amazing. In an interest of time, I'm going to aim for those that I think the audience, mostly audience that might have, so like the elephant in the room first without spilling any secrets. We had a 92.5% accurate number. One might think why not 92, why not 89, like what is driving all that? Yeah. There is some matching logic in terms of how it comes up with that score. And we've, you know, refined that a little bit. I couldn't tell you from the top of my head how we ended up calibrating that. But that was a conversation at length with the AI tool as we were building this application on how it would score the match. So there's, you know, appropriateness in terms of the career stream, appropriateness in terms of the level, appropriateness in terms of the match to the family. And I couldn't tell you about the proprietary blend, probably just from recall. But we do have, we had a conversation at length with the AI to be able to refine that the way we wanted it. Excellent. Second question. This needs to map your job architecture and we all know there are some very naughty managers out there and that, right, we cannot go to the salary number that Canada was asking for. Let's beef up their job title, right? Is that, like, how do you, how does the tool handle requests like that? Can someone, managers just go in and change up titles? So you can actually modify the business title. There is a job title that's tied to our architecture so they can, they can modify the business title. If they want, there's a certain amount of leniency that we allow in our organization. It's not going to change the underlying job title. But let's say we do want to increase the level. So there's a couple of things that I'm going to just call attention to, you know, easiest thing for me to call out is something that's quantifiable that we can identify. So you're supposed to have experience, but, you know, certain things like the preferred experience or sorry, the competencies of skills and things like that. Intermediate technical knowledge, developing professional expertise, things like this. These are characteristics that align to both the role and our leveling guide. Now, if I level this role up, you'll see that it's not just going to be the years of experience that's modified. It's also going to be some of those competencies and other things. I mean, I'm not going through all the detail here, but now all of a sudden I'm leading the architectural design. I'm driving the full life cycle development. I've got more years of experience, yes, but when I go into competencies, I'm a subject matter expertise. I have the ability to solve complex ambiguous technical problems. So the great thing about this is if a manager says, you know, in order to get the person I really want, I want to kind of juice up that title. Yeah, that might be possible, but by the time they actually change the leveling, in order to have it still align well with our job architecture, it's going to have also other characteristics that are going to set different expectations for what the role is and what it does. Josh, thank you so much for walking us through this tool. We're going to end our recording with a very simple question that looks forward. Sure. Job matching, you've already tackled one of the biggest beasts in compensation. The kind of grunt work that many of us love to do and also hate to do it the same time. Now, if we look in like two, that's too long. One, because things keep on changing. Let's say six months to a year time. What is that one task, one piece of work in the Torah word space that you're looking forward to either enhancing, automating, augmenting, whatever verb you want to use with AI? Well, I can tell you, you know, we've got, in the Torah rewards team right now, we've got four projects that we're working on. And I sure hope that none of them take six months to a year for us to get across the finish line because we're actively working on these things. A couple of things, though, that I'll tell you that we are doing. We're building a microsite, the microsite that we are going to manage and use to engage with our employees and constantly maintain an update. Now, there is as a part of that an AI component where we're going to have a chatbot that will use the microsite as the knowledge base for the responses that it gives. So as we have additional documentation that we want to put in front of our employees and things like that, we can load it into the microsite and they can then engage with that, you know, chatbot on the site and get answers. That's something we're really excited about. But what the other ones really come down to and I'm just going to say, you know, before I tell you what they are, why we chose them, right? They're the things that have caused different members of our team pain in the work that they're doing. So one of the things that you can do as a person working in Torah rewards is you can probably figure out what your pain points are, right? What do you want to complain about with your spouse when you come home at, you know, from work to say, man, this is really terrible today, right? Anything that you can do that about is an area where you have space to design something that can make your life better with AI. And so one of the things that is that for us or for somebody on my team is a global holiday calendar. We've got 35 different countries, a couple of business units, some differences in how different groups basically take time away from work and everything like that. So we built an application that will go through. We can load in the countries. It will pre-populate based on the year that we've chosen the statutory holidays within the country. And then we can assign that to somebody in country that can then do a review and make adjustments as possible. The great thing about this is it never is a spreadsheet that goes out. That has to have version control or something like that. It goes out. The person who manages this, that on my team, can see on a dashboard when they reviewed it, when they've clicked submit. And once everybody who's gotten an assignment has clicked submit, they can end up generating the global holiday calendar for our organization. Really great application, a limited use of AI, but, you know, we used AI to help build it. And there is a little bit of AI that's, you know, enhancing the functionality of the tool. Another thing that we've built is a compensation offer tool, something that uses all of our ranges, all of our geo-differentials across our entire footprint, and can end up helping somebody who is in a talent team or an HR business partner understand what kind of offer do we need to make to an individual that's going through the candidate process to make sure that we're offering a competitive salary. And that gives that full access. Again, that one's not using a whole lot of AI in the output, but we used AI to build what we needed. So that's where I would say is think about your pain points and you will find solutions in AI. Very deep and insightful for those that are struggling to envisage what's possible with AI. Josh had the answer for you right there. Josh, thank you so much for coming to the podcast and sharing your experience and insights with us. We look forward to catching up soon. All right. Sounds good. Cheers. Take care. Thanks. Bye. Thanks for listening. If something in this conversation was useful, send it to a colleague who'd get something from it too. Next up is Evan Noseb, head of People AI and HR Technology at Campari Group. His episode drops June 21st. We get into what AI adoption actually looks like inside a large global HR function when nobody on your team writes code. If you've been wondering whether that's even possible without an engineering background, Ivan has done it.