Subscribe
Share
Share
Embed
The WorkOps Podcast is your weekly conversation with HR leaders and People Ops practitioners doing the real work.
In every episode we dig into one story. A process that went sideways, a system that just didn't work, and what someone actually did about it. Packed with practical lessons you'll want to bring back to your team. Whether you're supporting 500 employees or 5,000, this is how the best People leaders are building for what comes next.
The WorkOps Podcast - Ivan Nosov
===
[00:00:00]
Jeet Mukerji: Hey everybody, welcome to the Work Ops podcast. Today we are joined by Ivan Nosov, who's the head of people, AI, and HR technology at Campari Group. For folks who don't know what Campari Group is, which I would find very surprising, it is a major global spirits company with a lot of the drinks that we all love, whether that's Aperol, Campari, Wild Turkey, Grand Marnier or Courvoisier. They're publicly listed on the Milan Stock Exchange. So Ivan, very excited to speak with you about what you're doing with AI. Before we dive into that, would love to hear how did you choose HR and people operations?
Ivan Nosov: Yeah. Thank you, Jeet. Really glad to join your podcast. Now actually I'm not an HR by education or by something I was doing or planning to do. I was initially an IT guy and with engineering education, starting doing my job well, but working for some reason with HRs a lot. And at some point they told me, "Hey, we have a [00:01:00] position.
Want to give you a bit more money." And at that time I was like all about, "Hey, I need money to spend time well." So of course, I joined it, and naturally that allowed me always to control a lot of the things that are not that usual for HRs. Rewards, budgeting, technology, analytics, like anything around those areas, and obviously last three years because of a whole, a little bit slow shift towards more of a technology driven companies, I was specifically pushing hard in the area out of my b- personal interest, passion, and volition
Jeet Mukerji: Love to hear that, that technical background is becoming so increasingly important in HR practices. Of course, the people part is central to it, but to enable all the things you wanna do, it's great that you've come in with the builder mindset already. And tell us a little bit about your role specifically.
You're looking at AI within the people space as well as choosing the tech stacks and what you decide to do in [00:02:00] terms of build versus buy.
Ivan Nosov: My major role is initially like we can say there is a foundation part which is foundational HR technologies like SuccessFactors at core, recruitment system, et cetera connection to SAP Analytics Cloud, whatever. And the second more emerging piece is around AI which is exactly what I'm pushing for starting with using HR as a proof of concept but also embedding and enabling the understanding of what can be done to other functions.
So we're doing a lot and right now it becomes more and more important AI enablement topic with hackathons capability building, changing the p- way people work, making people drop Excel and PowerPoint in favor of AI solutions. I'm really pushing that hard. And really I wish the companies to change how they work thus this is a weird combination of what I am supposed to do against what I'm actually doing
Jeet Mukerji: So that's really great to hear 'cause at the end of the day, yes, how do we build an [00:03:00] effective organization is a question that falls within HR's remit. It's really interesting to hear that you said that you're using the HR function as a proof of concept before you roll it out to other departments. What we've seen from other companies is that either there's like an AI transformation person that sits outside HR or it's driven by IT and it's more of a technology-led transformation as opposed to a people enablement-led transformation. What are your thoughts there? Is it something that should sit within HR or is it just something that's happening because you're passionate about it?
Ivan Nosov: Surprisingly, it's, I think right now for companies, a never-ending question. Because when you put it in IT, of course, the focus is on the technology. But technology itself without people understanding how it works doesn't work, because you need to bring the capabilities to needed level, and then you need to manufacture how the job should look like, how the architecture of processes needs to be.
Now, on the other hand, HR without IT cannot do [00:04:00] that because there is no architecture, right? So the question is, how do you create a proper synergy across functions and maybe even a third function which is separate from both to enable? Because very important thing that AI pushes the ownership towards the business, towards the final client, closer to final client.
So meaning if you can cut those the tails loose,
Jeet Mukerji: Mm-hmm.
Ivan Nosov: you can create much more efficiency. And HR has one very important benefit that often is being overlooked. HR speak with all the people, and all the people in the organization are the client of HR by nature, meaning HR can read signals very well.
It can see where the time sinks exist, where the processes might be less efficient, and how the jobs are architectured today to make a proper adjustment or proposal to make it better
Jeet Mukerji: It sounds like one of the most exciting times for HR actually to almost redefine their roles, their remit, as well as help to [00:05:00] redefine how work gets done across the entire organization, which is awesome to hear. Would love to dive into your story soon. Before we do that, I'm sure there are people who are listening in who are wondering, "This is all great to hear but don't know how to even get started.
I'm in the HR team. I'm maybe non-technical. Yes, Iman's got this amazing background of coming from engineering. My relationship with IT may not be so strong." What's the kind of the one thing that they can start doing from maybe tomorrow to get some of the outcomes that you're driving towards?
Ivan Nosov: To be honest, this is one of the interesting challenges I'm putting in front of me. If I just give the AI coding tier tools to usual HR who never touched a line of code and never were thinking what is possible, guide them a little, mentor, give support, and suddenly they do amazing things. So it's all about trying.
If the company doesn't provide a proper toolkit [00:06:00] to do that, try it outside. Buy a subscription to your preferred tool, like Cursor, Codex, Claude, doesn't matter. Install the proper of a coding level solution, not the usual desktop layer. Try to work with files. Try to create something. Any idea, any simple one.
One of my colleagues, I remember what he was trying to do is to create a fantasy team kind of collection that is made by AI running through a process and analysis based on the data that is collected. Simple yet reasonable, very safe to test. Go and do anything, and suddenly you will notice how fast the border of this invisible wall becomes invisible because it's not there.
It's in our imagination
Jeet Mukerji: That's awesome to hear. Okay. Let's with that dive into the projects that you're working on. And before the call and when we previously spoke, you mentioned that you tried a few things on Copilot Studio, it didn't quite work out and you took a different direction. [00:07:00] So we'd love to hear what was the original problem that you were looking to solve, and then how did it go with Copilot Studio?
Ivan Nosov: Yeah. So first of all, it's really tricky how Microsoft calls their solutions, and s- they, for some reason, call everything Copilot, which are completely different things and work at a completely different level of competency. I'm not really a fan of a usual Copilot because I believe they're abstract the fundamental model more than they build on.
Now, they're catching up. To be clear, they're catching up. Copilot Studio is a nice entry point for AI agent builders, but the problem mainly is that you have a limited harness of how exactly it works. You have to rely that it clicks based on the way they set it up, or you have to over-engineer that you lose all the simplicity benefits.
So meaning, generally, if you're not focusing on the real workflow changes, but rather the knowledge retrieval, it is adequate. Adequate can [00:08:00] work has its own limits in terms of specifically work with artifacts, but can solve many questions, the simpler one. It can solve a very good entry point to build a proof of concept, and in this area, proof of concept that something can work is also an important step.
Hence, when I was pushing this agenda, I was starting with Copilot Studio because it was already there, meaning I was able to build proof of concepts that already proved the ideas actually worked
Jeet Mukerji: That's a really important point. I think what you're saying here is then that actually you may want to start with a tool that is more easily accessible but you know it has limits, but it's just enough to validate whether it's gonna work or not. Once you've done that, it doesn't mean you have to carry on with that tool.
You can then move on to the more appropriate tool for the V1 or the V2. Tell us a little bit about more-- what do you mean by harness? I'm sure a lot of people are probably wondering that.
Ivan Nosov: So generally when I say harness is the ability to impact how AI will think against the contents you have. [00:09:00] So meaning like I'm making it a bit more practical for HR use cases because of course you can turn harness to any different direction specific to specific area of a application. HR would generally have to apply it against the knowledge that we collect or the data that we collect in any systems, in anywhere.
So meaning when I think about the contents in HR, there is a part of that has to apply to everyone at the same time. Your basic context, you don't want ever to bring it out. Understanding how the domains of knowledge are connected, what are the maybe some critical dates or how the HR organization looks like or what is the business unit or supply chain, right?
But majority of knowledge of HR are still domain-driven or process-driven. And of course be- because how it works then, now that goes a bit technical, but for instance, even RAG or just a context injection, this is a very big question for anyone working with AI because [00:10:00] like when you do RAG, you have a higher possibility of failure on the edge cases.
But you have a full download in the context window of AI of the context regarding the topic, then you force a more reliable reply. But you have then to work with optimized context and the routing against it, so it's always within the real reasonable limits. And the ability to impact this and then on top impact what's happening with artifacts, what's happening with data queries, is all possible when you like do the proper harness of the logic behind.
But when you outsource it, suddenly you have no idea if the problem is hallucination or how can I impact it
Jeet Mukerji: Exactly. You lose the control. So - If I can summarize what I heard from you it's basically like the tool or function calling infrastructure. It's also the let's call it the orchestration loop. So like you send a prompt, you get the response, you execute a tool call and then you repeat that. How do you manage [00:11:00] the context and the state across multiple steps? And then the really important piece you've mentioned around guardrails and permissions and how do you stop conditions and move things forward. So we're saying that actually Copilot Studio is limited when it comes to giving you that control. Fine for a proof of concept, but when you put it into production, then it's not so good. So what were you looking to build and where did you hit that limit with Copilot?
Ivan Nosov: Yeah. We hit the limit when we started to add too much of information, and especially started to want to help the real processes on the ground. So meaning, very simple example. If I have a very detailed description of my compensation package and how all the benefits work in Italy, and I have another file in Germany, if we do the RAG approach, quite high chances you will not hit the mark
Combine those two, because RAG goes into the chunks of the data, meaning it might lose the clarity.
But when I do the routing specifically, if [00:12:00] user asks about Italy, I load only the Italy context. N-none of other context exists in the memory of AI, meaning I can impact it this way to be very efficient and very targeted to hallucinate less, and especially it's important when you grow in the size. Now, of course, the RAG can be better, can be improved.
N-not a problem, but the level of an error will still be relatively high compared to the direct harness against the activity
Jeet Mukerji: Yeah. And that is what was not working so well with Copilot Studio
Ivan Nosov: Correct. Correct. It starts a little bit to stutter when the objects become too large, and it tries to search way too actively across the whole level of database. And you can't really point it out, "Hey, check there." So maybe there are ways, but then it becomes increases in such a level of complexity that it's easier to buy a proper router app because then at least you have a very fast way to test it out, while Copilot Studio creates the [00:13:00] challenges.
Again, it has its own use cases, but when you go to complexity and actual outputs, like artifacts creation, it's very hard to create good artifacts for Copilot Studio. Now, it is possible, but the complexity grows more when you create a separate solution that just goes and does it
Jeet Mukerji: And when you talk about artifacts can you give us some examples of those artifacts?
Ivan Nosov: Yeah. Look, the most simple example. Whenever HR comes to manager and stakeholder, they want to have some slides. Now everyone loves slides. I know of it's a bad thing, and by the way I really hate it. But to make people use the tool, with this kind of a digital tool I'm first of all referencing about as the entry point instead of a Copilot for the functional workers, is that it creates you either a very beautiful HTML based on the context that it knows plus the context you provide, with charts, data analytics, anything, like good pacing, and everything that can be done, or it produce good enough PowerPoint that you can reuse.
[00:14:00] And this is also important, like game changer. I hate PowerPoints, but I know that my clients still love it, and I want to provide them the tool that generates it very well, which we never had before. Unfortunately again, Copilot doesn't really do PowerPoints well. No one
Jeet Mukerji: I see.
Ivan Nosov: it. Maybe there are some ways to trick, but they are too hard to replicate
Jeet Mukerji: Gotcha. What did you guys try instead?
Ivan Nosov: So what we did is created this kind of a digital twin app that is fundamentally a router application. So meaning you have your chat screen, drop your question, your conversation history, all the recent files you've generated. You drop a question. Now, by the router logic, it knows who you are, it knows what to inject first.
If you ask about rewards, here go rewards module, talent reward talent module, anything, and uses all this data plus whatever you provide to satisfy your requests, which are most oftenly are artifacts. And also, it's not just like [00:15:00] the decks, right? It may be Excel tables, just well-organized Excel. Now the view is "No, but Claude does it quite well," right?
So and that-that's fair. And by the way, it's built on Claude. So it's built on Claude models underneath. But what we're using, like using Sonnet as the main model for efficiency, Haiku for any helper queries, and Opus only for the final artifact generation because Opus has a better design taste.
And design is the number one thing that in this specific topic for HR business partners decides do they want to use your tool or not
Jeet Mukerji: It's so true, right? Like it's gotta have the output at the same level that you would expect a human would get to. Now, is it the right use of a human's time to round up slides, to move boxes from one side of the slide to another? Probably not. But it's crazy how important it is for it to look good, to be digestible and be useful so that the information doesn't go to waste. Are you, are you pre-selecting the models for your end users the HRBPs? Or they having to [00:16:00] decide, like what does their workflow look like when they're creating this artifact?
Ivan Nosov: In this situation , in this solution, we are presetting everything.
Jeet Mukerji: 嗯。
Ivan Nosov: So meaning templates, colors, spacing, rules of a better design, how to make it work, how the assets are being injected in the HTML through external sources and everything to make it as easy as possible, as consistent as possible and as predictable as possible, which is very important here.
Because sometimes when you switch from the LLM logic to actual workflows, you also want to have a quite high level of predictability, is contradictory to how LLMs work
Jeet Mukerji: Yeah. You want some deterministic flows in there, or at least some kind of structure and guardrails to follow. And before we hop into the chat, you mentioned that, okay, yes, we've got Claude but the data is being stored where? 'Cause it sounds like you're pulling s- sources from...
You've got your templates, you have your information somewhere else of what needs to be in the presentations. And then there's also the routing [00:17:00] problem too
Ivan Nosov: Yeah. There's a couple of elements on that. So first of all, and this comes back to the context engineering discipline, and by the way today is one I believe one of the biggest problems of companies in general, too much of the knowledge is tacit and not known, not written anywhere. So at the first stage, even during the, our experimentation with Copilot Studio, what we were doing, we were trying to distill all the knowledge we have.
So basically what I was doing, I was asking all the colleagues from COEs to HRBPs, "Give me all your decks you have." Like everything, like whatever. I will run it through Claude or anything I have on the VIP coding side to just distill it carefully, slide by slide, graph by graph. Policy's a bit easier because it's already a text.
But distill it all, and distill it to such a size that there is no duplication or ambiguity created, and it becomes compact. So to be honest, our total context [00:18:00] is around 250,000 tokens. Obviously not everything is loaded a- any time. Usually session loads no more than 50
Jeet Mukerji: All
Ivan Nosov: to have enough of a space for any discussion to be had.
Plus, users can bring in their own files, images, PDFs, PowerPoints, Excels, almost anything, depending on the size of course, because any chat solution like that will crumble at certain point limit, but that's why we had other ways to do with, to work with that. But what's important is to create as much flexibility to make it as plug and play as possible with minimal technical knowledge and minimal barrier to entry
Jeet Mukerji: Nice. And this is where it gets even more interesting is great, you've got this, you've got this capability and you built out the proof of concept, which already I'm sure generated some interest that was tested. You ran discovery. Now you've got the version that is in production. How did you drive adoption within the team?
Like, how did you know that yes, this is gonna be sticky and it's working?
Ivan Nosov: [00:19:00] Yeah. First of all, I was specifically working with personas, with people I needed, I need to win over.
Or those who are already, let's say, on a more, "Hey, we're interested already. How can we bring it inside?" Work with them, ensure that the outputs are exactly what they expect, open event till the knowledge fits, till things click.
And after they click like one of our regional HR directors were, was basically, whenever we had the recent round of engagement survey, we're just dropping the raw files that we have from Viva Glint in our solution and getting beautiful presentation country by country without even asking the country.
And then his point was, "Hey, why do I get it in, like in 15 minutes and the quality is so good when the team has to work a day or two, like just to get the same," right? And then people start, people get to it. Like our rewards guy was really engaged with the [00:20:00] dashboards generation, right?
Because suddenly he had a tool that make a dashboards on what was happening within our compensation cycle available without asking anyone from the team, based on his taste and his vision. So the things are really getting fast as soon as you get those people, because then they share
Jeet Mukerji: So that's really interesting 'cause you're actually starting from the regional directors, heads of centers of excellence, that level. A lot of folks start with things like hackathons and let's build from the bottom up and do it that way. But it sounds like you went almost the other way of get the champions and then filter down.
Ivan Nosov: But actually, like I was doing it on the all directions, so we also run the hackathon with all those people. Now, it, the hackathon focus was different. It was more on the product capability through AI. But it was also very important to a little bit let people touch what it is in live session within a day, right?
That's also a very strong tool. We're really banking on this. But they work from a bit of a different directions. One, [00:21:00] awareness on the product capability and what AI can do in terms of micro products and products in general. Another, how I apply it to my day-to-day work, which is very different
Jeet Mukerji: Yes. Absolutely. face any kind of resistance? 'Cause it sounds pretty great. It sounds like it's being used a lot, but I wonder if there was any resistance
Ivan Nosov: No, of course. Of course there is. So I'd say there are three main lenses of resistance. One people got used to working like how they were working, right? So it took, for instance, a while for some of L&D colleagues, for me to convince them that, "Hey, you no longer need to draw slides. Stop moving boxes."
Because that's exactly what they were doing quite a bit of part of the day, right? Move the boxes. Stop doing that. Agree that 85% of quality is enough, but it is done in a fraction of time that's good enough. Or another type of resistance, the people that don't believe AI brings something new For them, you have to [00:22:00] get their friction out of them.
What is the problem they're struggling with? Something they hate. Address something they hate. Don't target what they love. If they love to create an Excel tables, don't touch it. But if they hate something, I don't know, reconciliation process at the end of the month, time management activity, to write emails to people, address those activities that they hate, they really want someone to help with them, and suddenly you help them, and they can no longer go back
Interesting. On the stuff that gets in the way as opposed to just trying to get the machine to do the things that they actually like to do, which is very important. And then your first point was, basically make sure that there's a mental shift of be okay with 80, 85% good enough. I imagine some people felt a little bit uncomfortable with that 'cause they were like this isn't quite how I wanted." Did you find people slipping back into their old ways after you'd given them this tool? Or did they hit a certain point and now they're never going back? They're never gonna shift another box from left to right. [00:23:00] Or is
No I do believe it still happens.
Jeet Mukerji: Okay.
Ivan Nosov: It's also about how convenient it is to do so you have to enable the ways to work with that. And one of the problem with these chat solutions that don't really allow to do fine-tuning of things, especially when we talk about the PowerPoint because the environment is not fitted.
The same if you ask in Copilot to add some... Edit something in a deck. It will regenerate the deck instead of editing it, right? The same thing. Editing is hard for those type of solutions. So you have three ways around it. One, your PowerPoint script underneath is excellent, so people just take the PowerPoint and then move it a little bit as much as they want.
This is what we achieved, and this was one of the very good alternatives. Two allow them middle ground solutions if it's possible, like CodeX, again, the Claude co-work type of solutions that work with files directly that can fine-tune the files. If it's possible, it's a good [00:24:00] alternative. And three make the regeneration as easy as possible.
Optimize how it's being done. Now, of course, it's costly, but you have to provide options
Jeet Mukerji: Because you're gonna get different people at different levels of maturity and different amounts of free time that they have during their day job. So that optionality is gonna be really important. How are you finding that level of maturity? Is it improving? Previously when we spoke, you mentioned that the initial kind of first few folks that came in, it was not so easy to bring in.
It was a bit slow, but then the next lot came on really quickly. Are they then also starting to become more advanced in their usage, or are you seeing they're running the same use cases over and over again through the same methods?
Ivan Nosov: There are a couple of things that are actually changing the needle, and one of them is people really loving how they have to do their job right now, because then they share.
Jeet Mukerji: Ha.
Ivan Nosov: And when they share, then when there is a colleague who already had more experience, suddenly you also can get a tip. You can exchange.
And to be honest, I'm also always learning new things about [00:25:00] AI from other people who haven't used this for a while. Why? Because some questions they ask, I never asked. I never thought that it could be asked, right? It all happens because AI surprisingly creates very personal experience of how you work with that for each individual person, because it depends on how exactly you think.
How do you organize your thoughts and files around? And that creates very interesting dynamic. Now, it's also, by the way, very interesting point exactly to that, that as your desire for the output grows, so let's add the layer of analytics, actual data analytics, research, et cetera. Suddenly, you start to get result you would not be even able to get by doing it yourself.
And then these concessions start to happen. Okay, maybe I don't need to fine-tune two words just because I already got result that I would not ever be able to get
Jeet Mukerji: You hit that aha moment, and you suddenly see that kind of unlock of "Wait, I can do all of this? This? is amazing." The question though [00:26:00] becomes, you mentioned earlier on that, "Hey, I had to wait for two days for my colleagues to be able to give me this presentation. I can now do this in 15, 20 minutes." Are you seeing the interactions change between the team who was being asked to surface that data, create those slides? What has been their reaction? Are they relieved or are they potentially a little bit concerned that actually, "Hey, we're not being asked for these things anymore"? What is that operating model been like?
Ivan Nosov: That's a very good question. And here what I also observe, and to be honest, like I never seen a person who mastered AI who started to work less
Yeah. So true.
Jeet Mukerji: There, there's always
Ivan Nosov: suddenly you can think so much more create so much more and actually reflect on what you are doing that you are not even necessarily every time spending severely more time.
But instead of making lookups, collecting the files, and figuring out, "Hey, how do I build a chart?" You get it and you iterate. You figure, "What I wanted to tell? What was the story that I was supposed to tell? [00:27:00] Can I add some more to that story?" Now like I was analyzing turnover growing in my country, maybe I can add the dynamic of our sales or the succession of sales KPIs on top to see if there is any correlation.
So you suddenly get a lot of time back, and this time usually people use to ask questions. Or well, some people were always overworking, and suddenly they can stop overworking
Jeet Mukerji: so right. I think all of us there is not a single person I know who's working less. They're either working the same amount or they're using their time more effectively, or they're in fact working a little bit more because they're experimenting, and experimentation, at least during these times, takes a little bit more time. Does this change the HR team's role and function? Are they doing the same kind of work? Are the KPIs for the HR team the same, or are you seeing that transition take place too now that there's a lot of AI being used internally? We were talking [00:28:00] about how the KPIs and the expectations from what HR need to do How have you seen those change now that there's more AI being used and now that what took them a long time is now taking them only half an hour?
Ivan Nosov: Yeah. On the KPI side, it is too early to say in clear numbers. But I see people change how they work and where they spend time. instance, when my colleague, senior HRBP, who before spent four hours to prepare for the talent review, now spends half an hour and is happy about it because she was able to do that herself.
Chances are this time goes back to business. And this is observed very consistently acr- across those who start to experiment, because those who start to experiment also come from the top, from those who really don't have time, and they are the most, let's say, engaged. Now, it still takes a bit of an effort to make them start doing that.
But when they start, [00:29:00] share they compare, and it
Jeet Mukerji: Yeah. Yeah. mentioned it's people who are busy, it's people at the top. How do you see this if at all, the shape of the team, let's say in two years? Maybe three or five years is too far. Because it sounds like a lot of the work that the AI can now help with is administrative stuff that you may have asked a junior person in the team to help with. So what does that shape look like if a percentage of the junior person's role is no longer required?
Ivan Nosov: That is a very clear threat to overall logic of career development in the companies. Before it was clear that the technical job no one likes was done by interns and early level specialists. Exactly while they do that by hands, they had a chance to learn how those things work. Now, this pipeline suddenly disappears.
The question is, how do we make it better regardless? Now, I believe here [00:30:00] actually question will a little bit go towards the logic of a business. Because suddenly when the situation is that I don't need as many people in certain function, actually it does not mean that I just need less people as a business.
Because obviously, if I just increase my margins, chances are my competition will actually use those new resources more efficiently. Then wouldn't they be the one who are losing if I just save the money? So the question is, how do you create the mechanism for invention, reinvestment? And how do I move the value creation and the efforts I spend closer to how I create a value as an organization?
Jeet Mukerji: Exactly.
Ivan Nosov: a complex question. And by the way, here also I feel that the actual victims could be at large vendors. Because then suddenly if I can move and develop capabilities inside, chances are this is how I prefer to spend resources because I then have smarter people inside
Jeet Mukerji: Exactly. And it comes down to the question [00:31:00] of build versus buy, right? And I think to your point earlier on, we have to be very, pain point led to really understand what is the right solution as opposed to starting from the solution. "Hey, I have Claude code or code work now. What can I do with it?"
Yes, you need experimentation, but you go for do the things that people hate to do, like you said earlier on. Focus on those pieces, and then we can find whatever the solution might be. And I also really liked how you said, yes, Copilot was great for the proof of concept. It starts to not be good enough.
And there may be, again, that curve where actually, yes, we take it as far as we can, and then you have a really clear view on what you need to solve for. And then the question becomes maybe we don't wanna solve it internally because we don't wanna maintain it. And at that point, it becomes a build and buy conversation.
I really like that kind of thinking that you've presented. What's next, Ivan for you? So you've got these artifacts that have been generated, people are happy, good adoption. What are you building
Ivan Nosov: So two things, more complex work related activities like automating actual [00:32:00] services job,
Jeet Mukerji: Okay
Ivan Nosov: Let's say if I have a team, how do let's say of HR ops, how do I make them work together through AI? Because individual productivity is just the first step. We the team productivity. We want people to work asynchronously on the same preset projects in the software development terms that allow AI not just to serve you, but serve your team.
And right now we're running experiments exactly with that to bring the people in one group, in one folders, in one activity, and they work asynchronously on that through AI. And we do that with people from HR who never touched those things and never thought about that this is possible, right?
This is the proof of your concept and where you move. And the second, of course, is obviously how do you move it outside of HRs? Because now as HRs get more knowledgeable, they're happy with results, they produce more, they share it and see what happens in other functions to make the next step in how to bring this capability out.
Because this is a game changing [00:33:00] capability for everyone. We need to control it as a company in a very nice way, but we have to let it grow as a kind of a virus
Jeet Mukerji: Yeah. I'm looking forward to continuing the discussion and seeing how the virus grows within Campari and it goes beyond HR. Ivan, thank you so much for joining us and sharing your experiences. We started the conversation by saying, "Hey just get started. If you don't have the tools necessarily, maybe do a weekend project." Are there any kind of final thoughts beyond that, that you wanna leave listeners with who are going through process optimization in the age of AI, what else they might need to consider?
Ivan Nosov: So probably I think the most important point is that we have to accept that AI for the knowledge work as whole is like electric engines for the factory floors, right? This is a good analogy. Even though like electric, again, like engines were a center AI is going from the all the directions.
But the transformation around it is [00:34:00] very large. And although I speak a lot about the capability per se within the existing processes, as soon as you have enough of the lay-- first layer of capabilities, you can start thinking about how you change the process it. The process that fits the new ways of thinking, modeling, and sharing the accountabilities.
Like very good example, do you really need as many different people in different functions even within HR? Do we need a separate rewards talent learning person? Maybe not so much. But for that, we need first to clear the initial capability gaps
Jeet Mukerji: that you're leaving listeners with a big question to, to consider. But again, thank you so much. If people did want to look at what you're building, follow up on the conversations can they find you on LinkedIn and reach out there?
Ivan Nosov: Absolutely. Thank you for calling me, Jeet, was glad to speak with you
Jeet Mukerji: Likewise, Ivan, thank you so much for joining us on the WorkOps podcast, and to everyone listening, we will catch you on the next one