From 4 hours of job matching to 10 minutes

Welcome to Range. I'm Jack Solomon and this is the podcast for Comp and Reward People building with the eye without losing their grip on the pay decision. Every episode, one practitioner, one honest conversation about what AI can actually do in our world and where you still need the human in the room. Be curious, stay a little skeptical and here we go. Good morning, good afternoon or good evening and welcome to a new episode of the Range Podcast. I'm here today with Greg Leney. Greg is a compensation and HR technology expert with over 20 years of experience, improving HR processes through systems, analytics and automation. He has led workday in HRS strategy. He built practical compensation tools. He taught older rewards and he regularly shares insights on HR tech and innovation. Greg, welcome to the podcast. Thank you. I'm glad to be here. What did you have today for breakfast, Greg? You sound very upbeat. Well, you know, my dogs actually got me up. It's a fear of what I'll have to deal with if I don't get up. This is really what drives me. More so than the breakfast. I did have toast and toast and orange juice to get me motivated. I'm curious. How did you end up being like professional in both areas? Yeah, so I mean, I started out like a lot of people. I did recruiting was a generalist. I found that I was more interested in like, hey, point factor analysis, any type of the statistics and analytical components of human resources, which really naturally pulled me towards compensation. I did compensation for till about 2011, 2012 and throughout that time, I really got heavily involved with like learning VBA, visual basic for applications using Excel. I had a mentor that really got me into Microsoft access and I started building access databases and I found that all my compensation stuff, I was really gravitating towards what's the best technology I can use for it. So when a manager, a former manager approached me back in 2011, 2012 for an HR analytics position, I naturally took the step. From there, we implemented a system called Workday and I've been in the Workday kick for a long time and only recently pivoted back to some compensation roles. Now, throughout this time, because of my natural pension for technology, in 2024, pretty much as soon as I heard about this concept of AI, I got involved. I got a plus membership with chat GPT and I just started going from there. When do you use NAN and when do you use chat GPT? So what I would recommend to people is start with a basic AI model. Like you talked about Clawed. I use chat GPT, whether it's you're in the office and they only have co-pilot or like MS 365 co-pilot. Learn what you have available and start from there. I started my process using AI for the basic stuff I think everybody does. Writing better emails, how do I write an email response to my manager without getting fired? Just simplified this document, this 20 page document that I don't want to read the full thing. All the basic stuff and then what I've noticed is all these tools over time, it's typically like every two to three months, they just get better and better. So I'll put in a prompt, it'll be total nonsense. And then I'll put in a similar prompt two or three months later and the answer is just gets better and better and better. So when I notice this, I, and I think, you know, once the world saw this, I think the idea came to me that, wow, I can really start doing things. So what I realized when looking at this was that with the technology, it's not perfect on day one, it's growing as just as you're growing. And so as you're growing, start it, you know, start basic. If you have a concept and you have a spot you want to get, just start with the basic, put in, help me write this email. Once you understand that, then start looking at help me analyze the spreadsheet and you work iteratively up. And that's, that's what I've done. Start with the very basic components and, and then you build. And then as you see that there's new functionality, what's going to happen is you're eventually going to hit ceiling. So there was a point where chat GP to couldn't longer do the stuff I needed to do just from a regular prompt. I tried all the prompt engineering I could do, but I could not get to where I needed to go. Then agents came into play and I started building out agents, built a lot of agents. And I was able to increase my level of capability within chat GPT, but I was still getting ceilings to where I couldn't no longer get to a particular point. Sometimes the workflows got too little, too complex. Sometimes the information was just too great. And so I started looking for there has to be a better way. And that's where I went to actually you to me, took a $20 class on what is this innate and or make or zap here. Anyways, help me understand how I can use it to create much bigger, more robust business processes. Once I did that, it helped me get to where I needed to go. And that's a crucial point, Greg, because in compensation and solar awards, like we have had access to data through pay surveys more recently, starting five years ago, we started having access to purpose built tools, platforms for that integrate with HRES systems, HRES systems, integrate with ATS application track and system all of that. So we again have access to new forms of data, but in terms of processes, we haven't always had tools that help us build workflows talking about what needs to happen, what needs to trigger to what triggers a new market review. Typically is a business leader and that is complaining that they're not able to hire, but there's no like more systematic way to say, hey, for example, our anchor to the market has moved by above our threshold, that's a 10% above market median, and that triggers a new review and initiates a workflow. Step-by-step with some tasks that are automated, some that are where a human is monitoring in real time and somewhere just working the background, then the human reviews the output. I'm not a huge NAD end user, but what I understand is that NAD end helps you mapping these workflows and connecting other tools to automate the bits and bobs that no one wants to do while keeping you in control. Is that a fair assessment of what NAD end can do in the HRES in compensation space? Yeah, so absolutely. One of the things I do want to call out though is while these large language model systems can move at exponential rate, there is a constraint. The constraint is the end user and the people that do it. Your creativity and your creativity really drives how fast this tool can move. That's why I'm definitely trying to encourage more and more people to use tools like NAD end and whatnot. I think at the end of the day, it's not going to be everybody using it, but I think a lot of people much like Excel. I've talked a lot of HR business partners surprisingly that don't use Microsoft Excel, but if you're going to be a compensation person, you need to know how to use Excel. I think we're going to see similar things for systems such as chat GPT or quad or co-pilot, NAD end and make and whatnot. A lot of these tools are going to be a natural requirement for development of compensation careers or HR careers for people that are much more technically savvy. Really the key point I'm trying to get across here is not everybody is going to be an expert in these tools and that's okay. Excel has the same functionality. Every compensation person believes that everybody knows Excel inside and out. The reality is certain roles do. Your compensation people do, your financing, accounting people do. Other people have some general ideas, but not everybody knows how to run a pivot table or use VBA or create macros. Macros exactly. It's a select group of people that have an invested interest in that particular area. The same thing is going to be with chat GPT. I think there's going to be a large population that uses AI for writing nice memos and creating meme pictures and stuff like that. That's okay. But I think for those that are really trying to dig deeper into to how can I automate my processes? How can I streamline? How can I make things more efficient? How can I make sense of the chaos in my compensation world? That's where I think the AI tools and the workflow automation tools are going to really be of great value to be. - And speaking about great value, Greg, there's one part of our job that I hardly ever hear someone telling me, oh, I love doing it. - I'm referring to job matching and also drafting job descriptions. But it's not just those things then the connecting those things together 'cause what happens next is you gotta benchmark your roles, you gotta find appropriate matches into survey, you gotta create a conversion table between your external pay sources, pay data, and your internal job architecture. - I understand that you've built a workflow presumably with a combination of Chatchy PT and 8N. - Yes. - Back in pairs, once a job description is submitted, compares the external survey benchmarks and internal roles and then recommends a likely market match. - So this is a great example. This is something that I originally started building an agent for in Chatchy PT probably a year and a half ago. It wasn't great at first but eventually the technology caught up to where it started working well. And what I did was I had the spreadsheet, I had the spreadsheet from the compensation consultant company that had all the survey matches. I used VBA to create a PDF file. So I created a file and I loaded that file into the agent and then I was able to successfully create something. But I wasn't able to go far enough but that only took me to a certain point. I still needed to, I had an NGPT which was at home, I wanted to use it at work as a tool. And I had the concept tested out in my home computer. So what I did was I replicated in Copilot. And I think one of the key things to touch on is that these AI tools are similar enough that once you understand one, you can usually bounce back and forth. So I created an agent within Copilot and was able to do this for work pretty successfully. I took work that took four hours a day basically and I was able to cut it down to five, 10 minutes a day for each particular instance. And so that saved me a lot of time and allowed me to work on other functions. So when you and I started talking, one of the things I really wanted to share was the functionality of a workflow automation. And so what I did was I spent about four hours creating this workflow automation. I would love to share with you. And I can share my screen. It's got a process and what it does is it starts simple. It starts with a job description that a manager has passed and not even a well-written job description just a quick paragraph. I loaded in and the workflow basically it compares the description provided against the market. It then looks at the internal structure and then what it does is it comes back with a grade recommendation and even prepares an email for me to send out to the manager. I'd love to show that to you. Let's have a look at what's on screen. This definitely looks like a workflow to me. I can see some connected nods. Absolutely. So this is the workflow. A lot of companies will have various things that I think are common in compensation. They'll have like a salary structure where they may have a grade or band or level or whatnot that have a range that I've created for this example. They'll have salary surveys that they participate in. And here I've just created a sample for this demonstration where you see I've got a simple survey description. I've got midpoint. And then I have an internal job catalog for this particular area. Now in this I have a couple of job descriptions that have been provided to me and lead budget analysts. I'm going to go ahead and copy this description that a manager is just sent to me. So that's an internal job description or is a market survey job description. This is just a paragraph and an email that a manager sent to me. But if they did send me an internal like a two page job description, I can copy and paste that. And then what I do is I have a tool right here and you'll see the workflow. The workflow, so an important concept I think, especially for someone that's not really sure about N8N or Zapier or Make, anybody that's considering it, why do I use this versus chat GPT or call out or whatnot. The reality is those tools work fine for a lot of components. However, if you want to follow the work step by step and you really want to map out the particular assignment that you're trying to do, these tools are fantastic. So what I've done is every tool typically starts with a trigger that launches a product. I am assigning variables. I'm then going out to my spreadsheets. This is a fictitious company data that I made up for testing purposes that I just shared with you. And then what it does is a code is created. And then what it does is it uses chat GPT to do some analysis. I'm analyzing the external market, the internal market and doing a great system. These are all things that if you think about your role in compensation, you're like, well, that's naturally I do that. I get a job description. I match to a job. And then I put a dollar value to it. And then I compare it to the internal jobs and say, you know what? This job is probably a great six or a great five or whatnot. And then what I do is I come back and say, yes, this is where it slots within our organization. And then I say, you know what? It is, in fact, a great six. And so this is what the range is. And now I need to send a communication out to the managers so the thing goes. So what a workflow does, basically, is once you kick it off, they start off with like a survey, which is really, you'll see here my survey is very basic. Let's say I create a job and call it lead budget analyst. I paste a job description in. So nothing special, nothing exciting with it. And now I hit submit. Now as I hit submit, you'll notice that the workflow is running smoothly. So it's like, it's doing something. And so as it does the analysis, it's doing a comparison of all the particular data that it's grabbed. And what we can do is, I want to show you the brilliance of what the end result is. But I think more importantly, it gives you an audit trail on what's happening. Both in the moment and six months from now. And so that's where I think the beauty of this is. So in this particular example, I loaded this particular job. And what I do is I come into the step here, I've got my download file. And as I pull it out, is that the underlying code? - Yes. Well, yes, and yes, it is. However, I didn't type up. I don't know, I'm not a software engineer. I asked Jack GPT to create the code for that. And so, and it pused Claude. Claude is fantastic at coding. So really what you do is, at the very beginning of the design of these, what you do is you say, I have a concept, this is what I want to do. These are the things that I need to have happen. I need to review a job description. I need to go to outside market. I need to compare it against inside market. I need to create a level. I need to assign it to an order, the existing level. And then I need to send an email to a manager. Those five steps I've created. And then what I'll do is they'll say, this is what you need to do. You need to create a node. And a node basically, if I take a step back, a node is basically each one of these boxes. It's a step within a flow chart, basically. That's a fancy, a node is a fancy way of saying a step within a flow chart. So what I've done is, in this particular example, I've created an analysis where it's got the email to the business partner. It's done the external analysis where it says, this is most similar to a senior to budget analyst, and which is position finance professional for with a market value of 122. So, and then an alternate position would be, let's give me a primary match. And then it's giving me a secondary match. And then you'll notice here internal perspective as well. It's like, why could not match to an internal match? Well, I'm going to, I'll show you in a second. But then what it does is it comes back and it says, we recommend that you use this particular range and it provides additional notes. And then it gives me a summary of what I want to do. And this is just stuff I put together in four hours. An organization, a company without it. Obviously, probably put a little bit more time into it. If we take this analysis, what we'll notice is in this particular job, I was looking at a lead budget analyst. And if I look at my internal, I have senior budget analysts, but I don't have a leader or anything like that. The senior was at 109, which is why it decided. It decided, okay, let's look at it, salary grades. Okay, 109 was the one below it was this one. And this job was a little bit higher than that. And so what it did was it assigned at a salary grade of 12788. And the reason I did that is because looking at the market survey, where it matched it was this particular job right here, which was 122. So it took the 122 and it made a determination that 122 most closely matches to 120. And so this is all stuff that a compensation person will typically do, but the GPT did it. Now, a few things you'll notice is, I mentioned the internal confidence was not high because there is no internal position that's created for this. The other beauty of this whole thing is, what happens is I've got a step where it writes all the data back into the spreadsheet. So this is the run that we just did right here where it provided feedback. It provides a summary of the analysis, it provides additional, it provides what the match recommendation is, it provides everything. And I can even add a field here if I wanted where I can go in and say, you know, HR business partner sign, or compensation partner sign off so that if I needed to do an audit report down the road, I could do such a thing. And so that's the beauty of the particular process. If you go in and look at it, really what you're doing is you're really just mapping things. And it's not hard. I think this is where you to me or something like that comes in helpful because where I was afraid, and the reason why I heard of these tools for a while and I was afraid to look at them. I signed up for a $20 class. I took the class until I felt confident enough to move around in it, which that was probably about a, I'll say probably like three or four hours with a training after I got to that point. If I'm familiar with using Excel or PowerPoint or any of these tools, it's not really hard. I think some of the hardest stuff is really just understanding the backend stuff. How do I connect by Google Sheets or my Google Drive to my AI? And if you can do, if you can do that basic stuff, then AI is going to help you with everything. I use it and say, I don't know how to code in JSON. Give me the script. And all this script that I shared right here, this, none of this is anything I wrote. This is something that the computer wrote. Gotcha. I had a few questions. So well, first you mentioned auditability. And you showed us a sheet where only you can go back in time and say, hey, six months ago, we made a record of what the tool did. We can go back and check it. But it also explains what it does. So you've touched on two major pillars of whenever you've met AI features in a new platform and a new tool, you want to be able to explain the results. You also want to be able to audit them. So those two pillars have been met. Now, my question for you is you also mentioned connectivity to other tools you use. And com people, we've been obsessed with this for so long. We have performance data living in one system. We have our HRS in another. We have applicant jump, applicant data in another. How does any connect tool this different tools? What you do in a NADN has the capability to basically, it's using APIs. And I don't want to get too technical for a podcast. But what it does is there are ways to have NADN paying a particular system. So in this particular system, I'm picking my Google drive and asking it to look at a specific document. In a corporate world, what you might do is if you're using Microsoft co-pilot, for example, you might have it paying just right into your organization's Google Drive for your particular files. It's not hard to set up. Like, for example, you have accounts that are set up once you put in the API keys that the NADN does the rest. So I do this once with a setup. And I'll tell you, to be honest, getting the API code and running that, that was probably the hardest part for me to figure out how to do it. But again, using an AI, it's not that hard. I sit there and I'll say, I have no idea what you're saying. Explain to me like I'm a fifth grader. How do I find this particular API code? And then what I'll do is I'll say, go out to the website and this is typically where you find it. And, you know, strangely enough, even myself, and to be honest, with the full disclosure, I am not a software engineer. I am a human resources person. I have a bachelor's degree in psychology. I did a master's in industrial and labor relations. None of this points to computer science engineer. Yeah, because I do compensation and whatnot, I have to have a natural curiosity. And I think if the end user has a curiosity, day one, this, for example, took me about four hours to create this process, which does a lot of work for me. And it could save how much time. Yeah, how much time it saves you? Like, it will save me three and a half hours on the first run. And then three and a half hours every run after that. So, I think that's critical to keep in mind is, it's important to invest in these particular tools. Had I designed this, had I tried to build something like this, day one, I couldn't. It would have been too hard. And so what I did was I started with, help me look at job descriptions and look at survey data. That was step one, and then it just became an iterative process to where, okay, I think I can make this a little better. I want to compare against the internal market. Help me look at this survey data and then compare against internal data. Okay. Now-- >> Gradual progress. >> Survey data, internal data. And then I want to look at my company's salary grades to see what it is. And then survey data, internal data, salary grades. I want to write a letter to the manager to tell them what my results are. Oh, and by the way, I want to record so that if Jack and Greg leave the organization, there is some sort of record that says why we evaluate the job the way we did. So I think it's only going to help as a tool. There's a lot of upside to this. It just takes a little bit of effort to get in there and peak under the hood. >> Yeah. And I love how it connects all this different tools. And just to recap, like we mentioned API, it just stands for application program interface. It's think about it for people that are listening. It's a connection point. It lets one piece of software talk to another without human in the middle. So let's say if you have your mercers, your WW, your A on RAD for data, you know. When someone says they've connected those systems, so your HRS for API, it simply means that the two systems exchange data automatically without the need for manual exports without the need to copy and paste. My last question for you, Greg, is, I saw the tool like designing salary bands for a new level. But how does, maybe it's connected to the API, I don't know. How does it know what your compensation philosophy is? How does it know that you anchor your midpoint of the range to the 70th persons out of the market? How does it know that you want your range spread to be 40% or 30%? How does it know that you only allow up to 10% overlap between ranges at different levels? How does it know all of that? - That's a great question. You've got these, the two brackets, it's a yellow or orange depending on the screen. What that is, is that's a code in JavaScript. The way you build the workflow automation using AI to help the AI is as you're typing, you first, you list the concept, this is what I'm thinking I want to do. And then what you do is you build this node by node or in common terms, box by box. And so you'll say, what do I need to do in this box? And then it will say, in this box, you need to insert code, the code needs to be this. And then what you do is you say, okay, I need more than what the code is you're suggesting. I needed to consider our compensation philosophy, which might be x, y and z. And you cannot have any exceptions regarding A, B, and C. It will then go back and it will recreate the code, so that it does exactly what you want it to do. And the beauty of this process is that, instead of doing a prompt engineering project, where you have a 20 page prompt that you're loading into the system and then you're asking information. What you're doing is you're breaking this down into bite size, chewable pieces of the process so that if I want to go back in and I want to look at, what did you provide me here? If I just grab any random node right here, you'll see it shows me what is the input that came that came in and then what is the output that I have here? I know with this data, with this code, this is what I get as an output. And so I'm able to test it step by step. And as I go step by step, I say, okay, is it working? Is it working? Is it working? Is it working? So again, it's that iterative concept of get the first box to work and get the second box to work and get the third box to work. And then eventually you've got a full process. I'm doing this at four hours. I was doing this at probably about less than five minutes a clip for some of these. Some of the coding, I wanted to go in and I wanted to test. I wanted to, I have to go in and set up the Google Sheet just to do it right. And I created some samples for demonstration purposes. That took some time too. So it's really not hard when you use your AI as a, basically as a colleague or a coworker or a co-designer. It does all the heavy lifting. All you have to do is conceptually, what do I want it to do? Tell the AI what you want it to do and it will build it for you. Now you've got to go back and test it to make sure did it do what I want it to do or did it misunderstand me. I think that's the critical thing. - The doing is not a problem anymore. Like you have a tool that can do it, do the job earn you. But it's having that intentional imagination, like having the idea, right? This is how I'm gonna leverage technology. Here's how the technology is gonna enable me to get this job done. That's what really, that's what you bring now. Like just before we wrap up this episode, Greg, first and foremost, like thank you so much for sharing all of this. All the nodes, they can, in the beginning, but they're all interconnected to your point. You start small, you work from one node and then you expand to another and another and another. So even if it gets big, you've built every single node. So you know exactly what they do. - Oh, by the way, these are things that I think compensation people have done for years regardless. They just haven't done it in this particular format. So I remember years ago, I used to do a compensation, like a merit process or a bonus process. We were introduced to a concept called project management where they use the project management book of knowledge where you would say, okay, I need to map this out. I need to have a very sequential process who's responsible, create racey charts. So I know who's responsible, who's accountable, who's consulted. I did all that stuff anyways. Now what I'm doing is I'm just creating it and I'm putting it on an app, NADN. And I'm using AI to help me build it. So I don't even have to be an expert in these particular areas. Now I think where the compensation piece comes into play is you have to know if it makes good sense. AI hasn't caught up to us as far as our professional expertise in particular areas. It will tell you, yes, this is the process and yes, I fold your process, but it doesn't know, if I should make this job instead of a grade six, I should make it a grade 20 because it's higher than five. You as a professional need to make that decision in that professional call. So I think it's critical to understand that the end user is ultimately responsible for the results. And I think there's still a need for someone that's an expert enough to do testing and to verify. But with these types of tools, the grunt work, if you will, the building out of these processes, things that, honestly, some of these processes, I built out stuff that would take me one to two years that I've built out in two months. So this will save people a lot of time. So don't come up with the penultimate brain, brainiac idea on day one. Come up with something small and work up to it. The second thing I do too, for a lot of these processes, what I've done is I've started very, very small and said, that's too complicated. I don't think I can do that. Let me start with this and then what I'll do is instead of just coding, coding, coding, coding, you know, working on it, what I'll do is, I'll do a little bit, I'll get frustrated or whatnot. And then what I'll do is I'll take a day or two off from looking at it and just going for a walk, talking with family, sitting at a barbecue, or sitting at a beach even, and then all of a sudden, it will hit me, wait a second, I didn't think about tying this into it. And that's usually how I come up with my best ideas. So I wouldn't encourage anyone to start small and take a class, take a basic class. You don't have to take the whole class, but listen to it for an hour or two until you get an idea of the concept. And then as soon as you get frustrated with it, walk away for, you know, for a few hours, a few days, whatever, and then just go back to it. Ideas will pop up. I am not a rocket scientist by any means. Anybody can do this. It just, it just takes a little patience and it takes, you know, it takes imagination, basically. That's a wrap. If this was useful, follow the show, and send it to the one other com person, you know who needs it. Thanks for listening, and I'll see you in the next one. (upbeat music)