Compensation is breaking out of the back office. Paul Reiman and Giac Soliman weigh in on AI, judgment, and what’s next.

Ruth Thomas (00:01)
Hello, everybody, and welcome to another episode of Comp and Coffee. And today we're going to be talking about compensation in the age of AI. Over the last year, we know that AI has rapidly been moving from experimentation to active adoption across HR and compensation. And if like us here at Payscale, we know probably teams in your organization are using Claude or Copilot and ChatGPT to support everything in their daily work, but as but specifically.

Aiming to use it for compensation management tasks. But at the same time, we know comp leaders, they're having to navigate important questions around trust, governance, data quality, and defensible pay decision making and what how AI can support that. So today, in this episode, I've got two new guests to the podcast who have not been the comp and coffee guests so far. So I'll introduce them in a minute. But with them, we're going to unpack this, we're going to look at what

Do we mean when we talk about AI in compensation? How are we currently seeing teams use it? Where are we seeing organizations getting the most value from it at this stage? What are some of the risks or considerations we should take into account when we think about using AI? And then whether AI changes are going to impact how we think about using compensation technology. So helping me to do that today, we have Paul Ryman from Novo Insights.

And Jack Solomon, who is founder of Range and a f former head of compensation, where mainly to be found on LinkedIn talking about all things AI in compensation. So I'm gonna let themselves introduce themselves. Paul, would you like to start?

Paul Reiman (01:41)
Sure. Paul Ryman, founder of Novo Insights. I have to correct one thing already. I am a repeat comp and coffee guest from Comp and Coffee prior to Ruth

Ruth Thomas (01:50)
I remember.

Paul Reiman (01:51)
being a comp and coffee host, episode seven in twenty eighteen, which ironically

Ruth Thomas (01:56)
Gosh.

Paul Reiman (01:57)
has AI references in it. before it was kind of a cool kid thing, surprisingly. But

I'm a twenty-seven year comp practitioner as a consultant as a practitioner in house at three different companies. And then I founded the firm four years ago where we help companies pay differently. So glad to be on again, Ruth. Not to give you hard time, but glad to be on again.

Ruth Thomas (02:17)
Apologies. Well, I hope

there are some listeners that heard you the first time that are listening

Paul Reiman (02:22)
That's right.

Ruth Thomas (02:23)
to this episode today. That's some while back. but it yes, comp comp and coffee does

Paul Reiman (02:27)
It's been a year. It's been a few years.

Ruth Thomas (02:28)
go back a few years. So welcome back then, Paul. And

Paul Reiman (02:32)
Thank you.

Ruth Thomas (02:32)
well, as a return guest, can you remind our audience, is your preferred beverage a coffee or a tea?

Paul Reiman (02:39)
I am a coffee in the morning and iced tea in the afternoon. Every day those two things.

Ruth Thomas (02:45)
And where you where are you based, Paul?

Paul Reiman (02:47)
And I am based outside of Chicago, Illinois, in the United States.

Ruth Thomas (02:51)
So you're you're drinking coffee at this point of the day.

Paul Reiman (02:53)
I am definitely still in coffee phase. Almost done with it.

Ruth Thomas (02:57)
Right. And Jacques, welcome to the podcast. Would you like to introduce yourself to the audience?

Giac Soliman (03:03)
Hi everyone, I'm Jack and yes, I'm the freshman here. I've

Ruth Thomas (03:05)
Jack. Mm-hmm.

Giac Soliman (03:08)
been at both sides of the pond. I was first in-house and then actually the other way around. I was first in consulting and then I went to in-house and then I was looking for another side of the pond and couldn't find it. So I just created it. I said, you know what? I don't want to be a consultant anymore. I don't want to work in-house anymore. What should I do with my life? And that's when I decided, you know what, I'm just gonna start, I'm gonna build a community.

You know, I'm gonna be the community guide that bridge, you know, bridges people, connects people. And that's what I do today on Building Range, which is the community, the AI community for toll rewards and compensation leaders. And yes, I am a coffee person throughout day and night. So it takes different shapes and forms. The very morning it's a cappuccino, because I'm originally from Italy, so I always want my coconut well.

It's coconut milk now, it's not more. But I I want my chocolate powder on top. I want to indulge in the moment. it then turns into an iced coffee because I really respect the Greek tradition of having ice cold coffee. If you ask for a coffee in Athens, they will assume you actually want a cold coffee, which is the other way around in Italy. And towards the evening, that becomes an espresso martini. Because why not? Right? So

That that's my I I do I do drink tea as well. After spending eight years in UK. I'm now technically British Italian.

Ruth Thomas (04:40)
Mm-hmm.

Giac Soliman (04:41)
so I do drink a lot of tea. my I've been told I should drink more water. Like I if if if not if not coffee, I go to tea, but at some point it should also actually I I'm not sure if this you're able to see this, but I got this from Payskill. I got a water bottle. That's

Ruth Thomas (04:58)
Well there we go.

Yep, I have my pay payscale water bottle here, but I also have another a secondary glass of water. So I'm

Giac Soliman (05:00)
Pay skill branded. It's helped. Yeah.

Paul Reiman (05:01)
I mean

Yeah.

Ruth Thomas (05:05)
definitely on water by this time of the day because I'm well, you and I are both in London right now, Jacques, aren't we? So so it's the afternoon as we're recording this podcast. Yeah, we're not quite at Espresso Martini time yet. So not long, not long.

Paul Reiman (05:17)
You'll you'll get there.

Giac Soliman (05:17)
Yeah.

Ruth Thomas (05:19)
Okay, so let's talk about AI in compensation. let's

Start at the very beginning. I'll do my best to Julie Andrew's impression. What do we mean when we talk about AI in the compensation context? And the other thing then I'd like to move on to is like how much of a shift do you think we're going to see in the profession? So Paul, I'm going to start with you. When compensation leaders talk about AI today, what do they mean? What technologies and capabilities should they actually be thinking about?

Paul Reiman (05:50)
Yeah, it's a a great place to start. I I think people, you know, not to start with a hot take, but I think we we're s most people are talking about specific tools, not a capability. so you say, you know, let's talk about AI, people's brain immediately goes to Gemini, Chat GPT, Claude, right? They're thinking about a specific model or a specific application. Sure, that's AI, but

There's AI in a lot of other places, you built within your comp tech stack, there are algorithms and models that are AI at their core. there are other tools to create things differently than ever before that aren't just those sort of frontier foundation models. so AI at the end of the day should mean it's this capability to predict something, right? That's essentially what AI is at its simplest form, whether that's predicting code, predicting text, and that's manifesting itself in these different places.

You know, my my call to action for comp people when they when they start with the tool is don't start with the tool. Think about where you need to predict something. And that's going to be where you look for how you can use AI differently.

Ruth Thomas (06:56)
Jacques, would you like to add on there and and then I'll kind of get you to start us off on how significant a shift do you think this is going to be for the profession? Are we going to see the work change? Are we going to see roles change?

Giac Soliman (07:09)
Yeah, I think this re this frame here, Paul, just gave resonated a lot with me. If you think about a change management project, a technology adoption project, which tool you use is probably just a fraction, literally a drop in the ocean compared to the overall strategy. It's like me asking you, "What is our approach to running pay reviews?" And your answer is "Excel." Like you see, there's a clear like, Excel is just a tool. There's a clear mismatch there.

I like to separate two things that I the people I spoke to tend to conflate. There's generative AI that you drive, you know, there's your in-house tools like clouds, Gemini, et cetera. And there's AI baked in vendor platforms and very different governance questions for both, right? Since we are gonna talk about the the use cases, the risk, the governance, etc.

People say A AI and and they mean either, whereas, you know, they're two separate things. And in terms of adoption maturity, I think, I think it's it starts with understanding of different levels of AI adoption. Most of us, I think, are still the chatbot users, essentially polishing out comms, drafting J Ds, you know, explain this regulation to me and and that's great.

Then there's level two, which are growing at building custom tools. If you think about Gemini gems, custom GPTs, Notebook LM. Level three is where things get a bit rare and isolated. There's just a, you know, maybe a dozen out of a hundred people I know in the com space that use code enabled tools like Cloud Code, Python, that build agents. And usually there's one person in the entire company.

Doing it alone just by themselves, right? the funny thing about AI is that we all know it has potential, but w no one knows for sure where it's going, right? how significant it is. I can definitely see something significant happening, and it really impacts the operating model of the team. I think it will in the next couple of months. What AI helps you doing is, well, let's take a step back. Currently,

If you're just using obsolete technology, which is Excel sheets, came out in 1985. We all love them, but still, let's think about it, came, it came out 40 years ago. You would spend most of your time processing, you know, getting the work done. And by the time you have to make decisions on the data that you processed, by the time you get there, you probably have very little time to actually apply any judgment.

What AI is doing is you know is compacting, is is shrinking the amount of time needed to process and to perform tasks. And this is not to say that the tasks will require less time. It could. That it does mean though that you you have two options. You can either reduce the time or you could say, right, all that time I used to you know allocate to processing, I can now redistribute it to other sort of

Flows of the same workflow to other steps, judgment, imagination, intuition, and actual decision-making stakeholder management. I think that's what we'll see in the next couple of years. We'll see processing and then perhaps the need for shared cost centers, you know, coming down, but the need for better imagination as a skill and better judgment actually going up. And it's a really interesting world. No one knows.

how this can actually pan out, but if I had to come up with a prediction that this would be the one.

Ruth Thomas (11:07)
Yeah. I I've always said, you know, c compensation work has always been ripe for this type of disruption. you know, we have comp tech tried to do it, but I think AI is just going to accelerate that. I was rereading one of your papers before, Paul, as I prepared for this podcast. Do you want just t explain to the audience you brought out three papers recently? I'll let you just explain what those are. But one of those papers, which is one I was reading, is very aligned to what Jack was saying there around you know

productivity and workflow improvements for repetitive cyclical work, which is what a lot of us found ourselves doing. And hopefully, you know, we get to park some of that and start to be able to have more time to consider the outcomes of the processes that we've been running. So do you want to talk first of all talk about the papers and then, you know, add on there?

Paul Reiman (11:57)
Sure. Yeah. In April we published a series of three white papers. the first is all about a lot of what Jack just said, right? How is AI going to affect comp? We also believe it's transformative in terms of making you more productive. A lot of the work we do is under resourced, and this is an accelerant to the same outcomes we've been chasing, but we can do it in more efficient ways. So the paper's about how to think about that and, you know, starting finding problems that you have and applying AI to that rather than thinking AI is going to solve everything automatically.

Like taking a problem focus. The second paper is, well, is this going to kill CompTech? Like if everybody just gonna build stuff and AI is gonna make sort of CompTech disappear. You know, the short conclusion is no, but some parts of CompTech are at more risk than others, just given how they've added value in the past. and then the third is if you're out searching, you know, if you've decided that building is not the right answer, how do you buy it? And how do you get past some of the vendor claims of

You know, the the language that marketing uses, but may not be the way that comp needs to think about what AI is doing. so those are available on our website. I I feel like they're already out of date. Like not to not to submarine my own content, but like this world is changing fast. And, you know, the use cases that I was using AI for in March are different already. Like there's already things now that the models have improved, the tooling is improved, and or I just figured out how to do it better.

You know, it's a little bit of all of that that boy, it feels like three months is a long time for these things. 'cause the value

Giac Soliman (13:29)
Absolutely.

Paul Reiman (13:30)
is accelerating, not like everything we said is still true and then a lot more is true is what I feel like is is accurate in the past three months.

Ruth Thomas (13:38)
Yeah. So let's pivot to that value. like you've both talked about focusing on kind of the outcomes and less of less on the tools. where do we see value for compensation teams then in using this this type of technology, either standalone or you know, within a within a vendor platform? Jack, you've been experimenting with AI tools, you've been kind of leading the charger on that. I've seen you know, been seeing

You know, what are some of the practical examples that compensation teams should start exploring today?

Giac Soliman (14:12)
Yeah.

I've definitely been experimenting but more recently I've been observing what other compensation total rewards leaders are doing with AI as well. And there seems to be consensus around one thing, which is probably a low hanging fruit. It's anti hype, but it's generally value adding and not zero but low risk, which is communications and explanations. Think about

Told her word statements, merit cycle letters, explaining how your equity refresh works. These are questions, no matter how polished your policy on Notion on the internet is, these are questions you'll keep on getting because lots of people just don't want to read a policy. So if you can come up with a front end application of something that explains to people.

how their equity refresh program works, what how to read, a guy to read their tour reward statements. Like that's already taking a lot of the dread, a lot of the burden out of COMP COM teams. Suddenly you're not finding yourself having to answer the same questions on repeat. And that's already a great and unlock of productivity for for com teams. That's

Definitely where I would start, but it's also where I see most com teams start inside the community and out. There's another use cases that seems to be getting a lot of tractions. And it started with drafting job descriptions, but we all know what comes next. Job sizing, job leveling. so I'm seeing more and more teams using AI to draft job descriptions, but also to sanity check level mapping.

So it's very important that we say sanity check because you're not suddenly, you know, delegating it all to an AI, especially if you know it's a brand new AI model that needs to be trained. Right. Instead, you're using it as a second brain, as you know, intellectual sparring partner to to to challenge you a little bit on that loving assumption because

You know, although methodologies exist and I'm a hay guy, I grew up my first job sizing, evaluating jobs with the hay method. I am a firm believer of job evaluation, but I remember going to you know a team meeting, they we were all asked to level the same job using the same methodology, and turns out we all level jobs differently. So

I think it's helpful to have like a second brain, AI second brain at sanity checks as levels. These I would say are two of the ways where I can see AI creating value that are lower risk, provide a modest productivity boost, and are a great place for to start.

Paul Reiman (17:22)
Yeah. I might be more aggressive on that though, Jack. I think because humans you know will level jobs differently too, I actually think AI is accelerating

Giac Soliman (17:32)
Okay.

Paul Reiman (17:32)
it faster than I originally appreciated it would. you know, whether it's in Comptech platforms, almost all of the leading market management platforms auto match, right? Sort of both the function and the level. And it's pretty darn good in all of the platforms I've used. like better than I thought it would be. So while I still I agree with you, human in the loop.

I do think it's better than we expected, right? and what causes it to degrade is the same things that causes humans to degrade, which is bad job descriptions, right? Unclear leveling criteria. You know, so if the source is is good, the AI is as good as as my analysts are. And we're we're really good at this. so I do think it's ex that's why I say it's accelerating even faster than we thought, right? I do think the the prediction of a level for a job is a prediction that AI is getting.

Pretty right when fed with the right information, both source, you know, source information about the job and well, what is your criteria against that? If those are both good, it's gonna be pretty darn good. which is a major unlock. I mean, that's a major unlock if you can start with a job description and you get a list of levels quickly. Like that's great. Like that's that's that's a world changer for a solo comp practitioner of one. I mean, you've been there, right? You don't want to spend the time on that. You'd rather spend the time, okay, now what?

And this really does get us there. So it is highly value creating in my in my view.

Ruth Thomas (18:56)
So you've you've talked about like total reward communications or conversations, you've talked about job descriptions and levelings. What about comp benchmarkings?

Paul Reiman (19:07)
Yeah,

what about comp benchmarking? I mean, I think you know, benchmarking, if I if I break that down, like the matching process, it's a it's an accelerant. Like I'll just be full stop on that. Do not trust the data itself, right? So that's one of the theses of the paper about will AI kill comp tech is generative AI and AI as we know it doesn't create data. It interprets data. And the internet is not a great source of benchmarks for comp. so if you go to Chat GPT, you go to Claude and ask it what to pay a job, it'll give you an answer.

based on something not auditable, not trusted, you know, not clear, right? So like if we break benchmarking into either a process or an outcome, like the process, it can help. The data's not trustworthy, right? There's real moat around sort of benchmarking data because you need to know where it came from, have confidence in its quality, you know, be able to trace back and get the same answer if you do the same thing twice. You know, all the things that AI is not going to give you.

You know, in its current form. Could that change? Yeah, maybe. I don't know. You know, somebody builds a MCP into a d benchmarking data source and as a result you can constrain Claude to just look at that. Great. Now you can use your AI tool to benchmark. But that's that's still future state at this stage of the game.

Giac Soliman (20:20)
I wouldn't be surprised if we see more compte vendors going that direction, right? 'Cause I mean, we know that these AI tools exist and people wanna use them. We know a shadow AI is used as well. It's a alternative channel to make your data available in a f in a platform it's getting a lot of traction, but it's not that simple, right? So I I would ex I think it's a smart move, but I I don't know if practically

is something that every vendor will feel comfortable with. And and to Paul's point, like yes, absolutely, if you're using AI to source data, well, it's just like doing a overpowered Google research research. Like every any single data point that's just out in the web, it's free. It's free for a reason. And it is not verifiable. Like I can go on Glassdoor or Elva's FYI and just say, I was paid a million

you

know, US dollars in my previous job. I mean, yes, maybe someone's there, maybe they have someone asking questions, but you know, if I if I if I insist, no, no, no, I was paid a million, you know, who's gonna say no, you you weren't paid a million. but there are ways that you can use AI to benchmark and mark and do market pricing. if you get it to write code for you and then run that code through something that

A lot of us don't know exist. I recently found out there's a terminal, which is essentially just an interface where you run code. I found that pretty damn good at designing salary bands the way it I want. Like I have a methodology in place. You know, I share with the AI the methodology. I share what my data set looks like. I don't share any single employee line of data. I'm just giving it

synthetic data that is very reliable. It looks it resembles quite well the actual data set. The AI drafts the code and then I run the code in the terminal. And the after I ran that code, it creates a spreadsheet with the actual data because it's in the terminal. So it's my local environment and not, you know, the data is not going to any third party, if not Google Sheets, but that's okay because we have a DPA in place and all of that.

and then that I found pretty re reliable. You'll see is not frictionless, like question terminal. gosh, like someone might have a heart attack.

Paul Reiman (22:52)
Ha ha.

Giac Soliman (22:53)
It is it is is not built for that purpose. It is something you use if you have any nothing else, and you wanna do things your own way. you know there is this approach, but yes.

It is not frictionless. I'll I'll have to be honest with you. Yeah.

Ruth Thomas (23:14)
So

Paul, you already kind of talked about data, or both of you talked about data being like one of the risks of potentially using AI. So I want to kind like explore what are some of the other risks. So if I think about some of the compensation processes, I think about repeatability and I think about governance and they're interlinked. Do you wanna, Paul, do you wanna talk a little bit about like what are some of the risks we might see in terms of using AI for repeatability and governance?

Paul Reiman (23:43)
Yeah.

I mean, we can't forget that when most of the time when we're talking about AI in this generation, we are talking about using a large language model, right? And that large language model is predicting words. It is essentially when you are on your iPhone and you're sending a text message, it's trying to tell you the next word of your message to help you along, right? That's in its simplest sort of retail form. That's what an LLM is. the risk is that you make

Like I wouldn't want to send my wife all of her messages with autofill, right? Because odds are I'm gonna tell her I want something for dinner that I we're not even talking about dinner yet. Like it's just gonna make up stories along the way. And that's a silly, silly version, but it's not that apply it to comp and that's a problem, right? So if I'm trying to explain how my stock vesting works to Jack's example earlier, it there's an answer and it and it needs to be the same every time. So the

The risk of overtrusting the model is it's going to predict wrong sometime and it will. Is that prediction quality going up? Yes. But it is not, it's not deterministic to use sort of the math term, right? It's still a a probabilistic prediction of what I think that next word would be or that line of code would be. So there's a big risk of over relying on that probability. The probability is not a hundred percent in a lot of things that we do.

You know, the probability and the quality of those goes up based on the source data. Our source data is not all that great. Let's be honest. People data is messy. Job descriptions are wrong. people data is missing, right? Like the skill that that person has that kind of warrants or justifies a different pay outcome may not even exist in a database. Right. So we have a number of concerns with sort of that data to prediction flow. And as a result, my ability to

Make sure it's the same answer to audit why was it made that way? You know, all those things are missing if you just overtrust the LLM. that's why AI and LLMs are not the answer for a lot of comp decisions, right? A deterministic model where I can say this is why it said that, right? Think regression for the stats nerds out there. Like you could say why it thinks that the y value is something given this X value, because it's math, right?

Those are often better answers for compensation outcomes because we need to be able to defend why we're we're producing something.

Ruth Thomas (26:16)
And so you're kind of like advocating still for the the human in the loop, which I know Jack, that's something that that you talk about a lot in terms of like with AI. Do you want to expand on kind of like the risks that you see and why we need the human in the loop?

Giac Soliman (26:31)
Well,

the risk I see is between at l if I look at at our inspirations, like most comp teams are dreaming, rightfully so, I'm dreaming as well, of you know, living in a in a world where they're gonna build the tools for managers and employees, you know, to self serve. And ultimately I think that's some you know, the goal for some. And I'm talking about use cases where it is

safer and just more sound to to build something in-house rather than buy. We can get into the build and buy discussion, I think be great. But let's say the risk is the the where we are right now to assuming that is, you know, taking for granted that is the aspiration that we build tools for managers, employees. In other words, that we find AI, we find ways for AI to enable ourselves

So that we can better enable the business. That's the that's that's the goal, right? Now the problem is that if you run up against, and this could be just me, but I heard a few times, a tool, may maybe a job matching tool. Like if you want to put it in the hands of managers and employees, it has to be nearly a hundred percent accurate. Just think about how high stakes the work we do is. Think about

the fire and forget problem on pay. Like once once a contract is signed, it's signed, like yes, you can fire people easily in the US, but not in the rest of the world, and still you get into trouble if you fire, you know, just like that. What I'm trying to say is before we put these tools in the hands of managers to self-serve, if the first few times, you know, the agent you build gets it right, but then it starts getting it wrong,

That trust that you spent decades to exaggerate a little bit to build, it's gonna evaporate very soon. So that's just you know throwing some cold water on the hype. We all want to build tools that are self-serve, but realistically, fine if it's just you and your team using them because you know this stuff, but the moment you put it in the hands of the manager and employee.

and they make pay decisions that you can't back paddle, you know, away from easily, that's the real, you know, that that's the real hurdle that we need to find that we need to fight and find a solution to.

Ruth Thomas (29:10)
Great. And then let's talk a little bit about compensation technology then. Obviously, you know, pace scale, we're a vendor. This is something we are living with every day. You talked about pace, Paul. I've never seen like our roadmap evolve with so much or have to evolve at so much pace. For me, AI is a bit like I'm running an 800 meters and it's the pacemaker, basically. And I am trying to keep up with it. That's how I kind of

feel a lot of the time in terms of defining our product strategy here. But, you know, it it's changing how we well, how we even we build software. So like most of our coding now happens, is AI enabled coding. But it's also significantly changing how we think about what we're building. What do you think, Paul? you know, you you have one of your papers, will it kill, will AI kill comptech? Give give the audience kind of like your insights

Paul Reiman (30:05)
Yeah.

Ruth Thomas (30:06)
from that.

Paul Reiman (30:07)
Yeah. There's no doubt that AI is changing the economics of software and the build versus buy discussion compared to, you know, years ago. you know, I'm just old enough to remember when everybody sort of, you know, built their own thing or customized an on-premise solution because that's how technology worked. And the the SaaS revolution told us customization is bad, right? Building is s silly. You should always buy and configure.

And I think that's not as much of a gimme as it used to be. I think there's a legitimate choice to make about build versus buy. And then Jack will speak more to that, I think. you know, I think within the within the buy, the economics of pr you know, the productivity of software is changing, right? So I do see across the veter landscape, more features are delivered faster. you know.

Things that we would see as feature gaps just close themselves. Like, you know, honestly, some of the times when we meet with vendors and we tell them, yeah, that's that's not great. You need to make that feature better. The next time we talk to them, it's done. Because it can be, right? The constraint not long ago was was skilled coders who could make features happen. Now the constraint is knowing what features you want. you know, and having a user experience that's sort of as effective and stitched together as a result. So it's like it's totally changed the way that software.

needs to be managed without question. you know, so I think the advice we give the market when thinking about that then is, well, what's the unique moat that comes from this this buy decision, right? Having a data set. Like AI can't create data, as I mentioned before. So like that's a moat creator. You know, integration and sort of a stitched together experience where your data talks to your ranges, which talks to your cycle, creates value. And

For a long time I was actually not real positive on sort of the need for integration across the space, but now I could see the value of it because that integration is hard to create when you're sort of building things separately. you know, and deep expertise, right? So, like Jack talked about he could create something in his terminal, but that's because Jack knew what to create, right? He understood how these pieces fit together and how to coach the system. You know, not all tools have that. And, you know, workflow expertise.

Not necessarily like the point click of the workflow, that's easy to replicate, but knowing how do good comp decisions get made? How do we frame this? What data needs to be present in the experience? Like that's what you need to evaluate for now. It's not so much does this feature exist? Because it will. It's more is there a proven capability in this provider to to innovate because they understand the problem so well or they have data that that I can't create otherwise?

Ruth Thomas (32:51)
Yeah, so I think, you know, one of the things I would say is just like really ask your vendor like where is AI in the product and how are they evolving it. I think, you know, I talked about the pace of how you've talked about it. Vendors are like roadmap is progressing. I mean, I kind of feel like the quarterly analyst briefing is gonna have to be like faster than that. We're gonna be doing monthly analyst briefing

Paul Reiman (33:13)
Yeah.

Ruth Thomas (33:14)
soon. but what I do observe is that differentiation between the vendors is is

you know, widening because we're kind of working at pace. We're all deciding, we need to go and do this, we need to go and do that. And you are seeing a differentiation. So that's if you're out there assessing vendors as well, is just like g make sure you get to the bottom of that. Don't accept the AI whitewashing, I guess, is is the the the major recommendation.

Paul Reiman (33:39)
for sure. Like I'm big on that in that paper. then we have a separate blog post about it too. Like just just ignore when a vendor says they're AI native because we don't know what that means. Everybody's got AI. You've really got to dig into show me what it's doing that I couldn't do before, right? Like how is it going to show up? And you'll get very different answers with different providers on that.

Ruth Thomas (34:02)
Let's pivot to build or buy then. So, you know, previously the decision when it came to CompTech was am I going to use CompTech? Maybe I'm going to use my HRMS platform, or I'm going to use good old Excel, which we love basically.

Paul Reiman (34:13)
Mm-hmm.

Ruth Thomas (34:14)
And you know, all the surveys out there was like 50% of people are still using Excel to either do wraparound work around their comp solutions for compensation management or to do all their compensation management. And now we're starting to hear this like build versus buy again, where

know, enterprise organizations are considering their holistic AI strategy across the whole organization and thinking about, you know, where which aspects could we build ourselves and which could we not. So do you want to talk to that a bit, Shaq?

Giac Soliman (34:46)
I

think it boils down to a couple of question a couple of questions. First, if you think about the use case you want to focus on, right, does it hold sensitive data or compliance heavy workflows? I mean, the building part, that's become easy right now because you've got tools that code for you. The bottleneck is not coding anymore. The question is how do we wrap around PII, how do we take into account data residency, data privacy?

Those are the real questions in our domain since it's so high stakes that we should answer. The building itself, like, yeah, I can I can ask Closcode to build something, but will it actually work without spilling, you know, data that shouldn't spill to managers, not just the public? Can someone maintain it inside your company? You want to build something, right? Building is just step one, testing it, deploying it, implementing it, but then there's

You know, there's costs of maintaining it over time. Ask yourself, is that what you want to do with your time? Like it if if you're generally a builder and you have the resources and means and it all works out, great. But if not, you know, there's there are companies that exist precisely for that purpose. You know, they've built custom build a product precisely for that need. It is a bit like, you know, we we all have

Paul was he was he was talking about what to prepare for dinner and and and he wouldn't put you know Chat GPT or Claude on Auto mode to reply to his wife. I would use the same analogy. The vendors are the restaurants out there. Like they're they they they were born, they are known for that dish. Like they, this restaurant is great for pizza, this restaurant great for steak. Then there's

There's you. You have all the tools in the kitchen now. That's your AI, the kitchen utensils, the oven, the air fryer. You have access to best in class, best, you know, whatever ingredients. Yeah, y y you can cook, but just because you can, should you? Right? And you'll find out there are things that no restaurant will will accept. Like let's say I go to a restaurant and

You know, they have a set menu and they can customize a little bit, you know. you want your steak medium rare? That's fine. But let's say I want, you know, this is almost like blasphemy for me, but you know, a pizza with cheeseburgers and bacon and and pineapple on top. There are many restaurants who will say, you know, I'm sorry, but we're not gonna be able to do that, or we can do it for you, but you'll have to wait an extra hour.

And it's gonna cost you extra time because our chefs need to prepare it, they need to source. At that point, if you have so specific needs, right, you might wanna build something yourself. If it's real edge or a one-off, that it's not just costly for you to wait on a vendor, it's it's costly for the vendor as well to accommodate the need because most vendors have a minimum viable product issue. They build products that cater to at least a minimum number of customers.

That's the economies of scale. That's economics of business, right? So it's if some of these niche use cases you decide to build rather than buy, I would say it's not just good for you, it's good for the vendors as well. Because the vendors, instead of having to deal with so many custom requests that sort of you know, di derail them from their product roadmap, can really focus on what they can do best: best in class software for the purpose

they've been building for.

Ruth Thomas (38:35)
I love that analogy. Yeah. And I guess it comes back to almost full circle to our earlier conversation around value as well. It's just like you if you're out there buying technology, make sure you know what the valued outcomes are that you want to achieve. And that will put you in a much better position. I mean, I've been selling HR tech for kind of like 15 years. And I would say most times, like or a good 50% of the the use cases.

the buyer, the HR buyer or the comp buyer doesn't really know what they want,

Paul Reiman (39:08)
Mm.

Ruth Thomas (39:08)
even though you know when they're buying, basically. And and that's when problems occur or all sorts of problems occur, not only in them kind of trying to get budget for buy in for it internally, but also like once you implement this system isn't doing what I thought it was going to do or doesn't quite meet our requirements. I think that that that hasn't changed, you know, that being very clear about what you want to achieve and helping you to make that strategic decision about build versus buy.

Paul Reiman (39:32)
Yeah.

The the one thing it's a it's a sort of a hidden risk in the build versus buy that I I was talking to a a client who's built buying a sales performance management system. So commission calculation essentially, which is just a calculator and workflow. So, you know, at the end of the day, some of the things I said about what creates a mode around ComTech kind of don't exist, right? They could but by making the choice to build versus buy, I think there's two risks that I pointed out to them that I think I would highlight for

other systems as well. The first is your requirements will change. And sometimes the configured system anticipates the ability to change better than what you can customize yourself. Right. Because you can go down a hard code path that you're going to regret later, or as a platform has thought about because they have to serve these broader needs. So that's one is like you might not be thinking about the change ability well if you over tailor to your specific requirements today. And the second is

Maybe this is a great time to just stop doing that thing. That's such an edge case, right? Like I know having gone live on workday three different times in my background, there were specific HR processes in practice. I was glad to use workday as an excuse to kill. Cause they needed to go. And it just so happened workday couldn't do it. So instead of saying, we can't use workday because they can't accommodate this process, it's like, no, this is a great chance to simplify what we do. And I worry that if because it's so easy to customize.

It won't make us question hard those edge cases. And do we really need them? Or is this a chance to have a better employee experience, a more scalable experience by just destroying the edge case? So it's it though, those are risks. I think they're, you know, maybe academic because people will see around that corner a little bit when they do the build versus buy. But I'm seeing some early behaviors because with the pendulum is swung back so hard to build feels viable that we forget some of the benefits that.

forcing a buy decision did have, you know, on on those fronts.

Ruth Thomas (41:32)
Yeah, the SAS benefits of like you you benefit from everybody's like best practices and also you know the pace of processing the development of that, which which which a vendor may is doing basically. You that still applies, I think. So okay. So some great conversation. I think we could probably talk for a long time more, but I'm gonna kind of try and wrap us up here. for a closing question.

let's try and think about what our listeners should be doing next. Paul, let's start with you. What would compensation needers prioritize? What should they be prioritizing over the next I was gonna say six to twelve months. I think twelve months is impossible at the moment with the pace of AI. So let's say six months if they want to adopt AI effectively.

Paul Reiman (42:21)
Yeah, I think it's start with the work. so if you're a comp team of one, you know what work you're doing. If you're a leader of a team, you've probably lost some visibility into exactly what your team is doing. and AI is best deployed when it solves a problem, not just deploy AI with a Jedi hand wave. So make sure you really understand what are those time consuming or difficult workflows, processes that are absorbing capacity. it's essentially the plan before you deploy stage, right? Think about what.

you can make better before you just jump in and start building stuff. I just think it's a flaw to jump to the solution. Understand your problem. And the problem is the the work that's hard, wrong, inefficient, you know, or doesn't scale. So that's where I would I tell everyone to prioritize. Start with the portfolio of work, then think about how a techno technological change can power innovation or efficiency given that problem set.

Ruth Thomas (43:16)
And Jack.

Giac Soliman (43:17)
I agree with

that. I would add one more thing. Focus on the problem, but don't fall in love with the problem. Fall in love with the solution instead. Cause as as calm people, sometimes we just get bogged down and overly focused on the problem. But really when we say fall in love with the solution, what we mean is, you know, focus on the outcome. Agree what that desired outcome is with your stakeholders.

To Paul's point before you start building, because that we all feel the urge just because we can, now we're gonna build. I spent so many weekends building stuff that I don't needed. Like I didn't need and I don't use anymore. So I had to learn the hard way. And most people will have to learn the hard way, I guess. But that's good. It's try error. Any learning journey is like that. Agree on the desired outcome, and then you can work out, you know, reverse engineer the steps, the tasks.

the the steps are needed to get there. so that would be my advice.

Ruth Thomas (44:19)
Thank you. Well, if people want to find out more from you, Paul, Novo Insights, they can find you at your website, but also on LinkedIn. And

Paul Reiman (44:28)
Absolutely.

Ruth Thomas (44:29)
Jack, on LinkedIn, where else can they find you?

Giac Soliman (44:32)
On the range website as well, range dot community.

Ruth Thomas (44:36)
Right, okay. So we'll make sure we'll include the details to Paul and Jack's contact details in the episode notes. Thank you both for joining me today for this discussion. It's been great fun. and you know, I think today's conversation has made one thing clear that AI does have the potential to make compensation teams more effective, more efficient, and more informed. but the technology alone is not your compensation strategy. I think, would be my advice.

So thank you both again for joining us and sharing your perspectives. audience, as always, we'd love to hear from you. So you can email us at coffee at payscale.com. That's coffee at payscale.com with your questions, suggestions, or ideas for future podcast episodes. we love hearing from you and your suggestions. Thanks again for listening and thank you both again for joining me.

Paul Reiman (45:26)
Thank you.

Giac Soliman (45:27)
Thanks. It's fun.

‍