Building AI : From vision to execution | A conversation with Artefact

Arthur Garnier (00:14)  

Welcome, everyone. I'm here today with Elina. Elina, you are a partner at Artefact. Artefact was a portfolio company of Ardian's Expansion team. You are still in the Expansion 'family', from what I get. And Elina, we just ran a very interesting workshop together here at the AI for Alternative Investment Ardian conference that Ardian is organizing. The second edition of the conference. And, you know, we discussed a bit together, the why and how to build an AI chatbot that leverages our company data. But before we delve into these topics, maybe a few questions for you. How would you say, Ardian as a specialized consultant in the data and AI field, Artefact or companies like Artefact, are able to support Ardian and the global private equity industry in general, navigate the AI landscape?  

Elina Ashkinazi-Ildis (01:13)

Well, Artefact has evolved a lot from when it was founded ten years ago, and a different set of skills has been built on as the company's needs have evolved. So, if initially, a lot of our work was concentrated on actually just doing very specific use cases, delving right in, after hitting walls a few times, we realized that the actual choice of use cases was extremely important. So, the starting point really should be, what business are we in? So, in your case, what exactly is Ardian's business? How are the teams spending their time to determine which use cases will be relevant? And, then of course, again, before jumping to AI, we now know that without data, we cannot do anything. So, in that way, I thought it was rather visionary on your part to actually invest first and foremost into your infrastructure, and in building your data platform. So when we started bringing AI, and you want to use it to crawl through your structured data, to go faster, better, the data is already there. It's clean, it's ready to go, versus a lot of the companies doing it the other way around and hitting walls as we have, probably in the past. So, we really work The first layer, the top layer is the use cases. What are we trying to achieve? Second, what would the operating model be? How do we connect the needs of investment teams, which is the core of your business, to people like you, the data scientists? How do we get you to work together? How do you choose your cases? How do you deploy them? How do you train people? There's then, of course, the layer of, as we talked about, infrastructure and data governance. And last but not least, again, is change management. So all of these four layers are extremely important. And this is where Artefact got very strong. And because we're able to work across this entire pyramid, I think that's what makes us a valuable partner for a company within your industry. And, of course, your size.  

Arthur Garnier (03:17)  

For sure, I think Artefact has been that really key stakeholder in building our data platform, as you mentioned, but also GAIA, our generative AI platform that's leveraging the data that we have. And what you mentioned is also really true because this is something that we have been observing here at Ardian. AI is actually a tremendous leverage for change management, because what we have observed compared to my experience in the past is that when you're talking about data governance, data quality, and so on, I would say it's not sexy, and no one is really interested in those things, but they are key. But now with, Generative AI, the fact that you can create an amazing use case and get crazy results with LLMs, suddenly makes people interested and they say: What can I do myself, in my team, to leverage the full capacity of AI? And often, as you mentioned, the answer is: first, you have to align on data governance and so on. And then they really get motivated. That's, really amazing to see that with this particular innovation, we're able to work on the full chain of the use cases, the technological use cases, or really the infrastructure and data up to the AI or generative AI part. But I think this is really key today. It's key for Ardian, but it's also key for our investments. And we also have a duty as investors to help and support our company in the AI field. But more broadly, we have observed specific things in the past: private equity has evolved so much right now, and it's no longer just about investing and bringing financial capital to industries and companies. They also need to bring something more; operational excellence, data knowledge and so on. And so, we also have an AI duty to support them and help them on their value creation journey. And as a consultant, I have also seen a lot of different industries, and not just private equity. What would you say the current AI landscape across industries is? I'm assuming it's pretty diverse, right? Like, we don't have all companies or industries at the same level. And so is this something that you have been observing as well?  

Elina Ashkinazi-Ildis (05:50)

Of course. And I think we have to remember also that the excitement around AI has come in waves and layers. So, different industries were impacted, and some got excited earlier than others. Because of the nature of their data, they got off to a much earlier start. But also companies from the transportation sector, with the coming of machine learning, which made a huge difference in the accuracy of forecasting, optimizing, and then, everything around operations, research, and optimization under constraints. So, for companies whose operations were really data-driven, it could make a huge difference. They jumped the gun a long time ago. You know, by optimizing the transport routes, you can save so much money by sending them to the right place, by making sure the truck is full. So, a lot of the industries, like digital natives, those who had data at the core of operations, have transformed, the way they function through AI. And again, there was this more recent wave that really got everyone excited. And I would imagine within your portfolio, you have companies also that are digital natives that probably have been using a lot of AI, but others that are a lot less mature, often linked to the size of the company. You know, companies of the hard-core manufacturing sector or hard-core physical didn't have to go to AI until recently. Their business was fairly protected. So they didn't have to worry about it. But today that's not the case. And even if not every sector is equally threatened by the arrival of Generative AI, every sector has tremendous opportunities for efficiency gains, for improving operations, which of course has a direct impact on EBITDA, which is something you guys are keen on. So yeah, this is a new wave. And what we're also seeing is even smaller companies, which in the past were mostly focused on, “ Now I'm within the portfolio of a private equity firm. So I will look for growth through acquisitions, maybe launch a new product." But now there's a very strong consensus that data, not just AI, but sort of data maturity in AI, is a major lever of value creation. And we're seeing, in particular, private equity firms asking us for more and more help in supporting the portfolio companies.  

Arthur Garnier (08:30)  

It's a really good point. And then, indeed, we've been working a lot with the Digital Transformation team and Data Science team to try to extend this assessment maturity across all of our portfolios. So not just Expansion funds, but also about Buyout, Infrastructure, Growth, and so on, to be able to have a full picture of how we could prioritize more, the key use cases within each portfolio company and also create a lot of synergies. Because I think what's beautiful about AI in tech, in general, is that once you start building something, it's not just something really niche, that can only benefit a specific company, it's something that you can extend potentially to your industry, but also throughout multiple industries. And that's why having, for example, the AI x AI conference where we invited and had a lot of different CEOs and key executives from portfolio companies, so they could really talk together, specifically on digital and AI topics. I think this is what makes it really interesting. They get to really find and understand that sometimes they have really similar pain points or ambitions, and they can really try to get common initiatives to try to get to the next step. I want to follow up on what you just said on the recent wave of Generative AI. Things have been evolving at a crazy pace in the last three years. We had GPT, and things stayed quite calm for a few months. Then there were a tremendous number of players offering Generative AI services or broader AI services. We had big techs, startups, but also, pure players. I'm thinking, for example, about Snowflake, Databricks, and so on. And I think it's really tough to navigate through the many opportunities that you currently have with AI. On one side, since then and that, now, everyone, not just data scientists or tech people, can use and leverage AI in their day-to-day work. On the other hand, we're also observing huge amounts of investments required to build data centers, train LLMs, and so on. So, we're seeing two potential outcomes. One being every single barrier to entry is being shattered with the commoditization of AI, making it almost possible for everyone to build use cases on their own. And then, on the other hand, we're also seeing a tremendous increase in investment, making it very complex for non-pure players, or the non-big tech, to leverage Based on your experience, and your experience with us, because you have also been involved a lot in building GAIA, what would you say could be an interesting approach today for someone looking at implementing AI in their company, or in their team? What could be an interesting approach? Should they go straight away and buy a generalist product? Should they try to build everything from scratch?  

Elina Ashkinazi-Ildis (11:52)

So again, the first question is, how core is this? What am I trying to resolve? How core is it to my business, to the way the company operates, and to my results? For many companies today, and especially when we're looking at a lot of these chat bots and generative AI solutions, they tend to be more focus on efficiency gains. So, you know, for a lot of companies, it will simply come down to saving some time on writing emails, on doing the summaries, maybe generating ideas. And these elements are very important, but probably, again, this is not the core of their operation. So it's not necessarily worth really putting a lot of money into it. And at the same time, the technology has evolved. So, one way, one approach is to test Constantly. I know you're fond of GAIA, but I also know that you're testing ChatGPT, Copilot, Duster, and Mistral to see how they're doing, how we're measuring up. So that's the way to go. If there's a market solution that answers all your questions and use cases, if it's accessible in terms of price, well, don't build it. Just buy it. We tend to come down to the same way of doing business. What is the problem that I am trying to solve? How core is it to my operations? Do I have the financial capacity to actually build something because of the upfront costs? You should probably also be looking At the total value of ownership. In the case of GAIA, the upfront costs are quite significant but at this point, you are paying for the usage of LLM; you're not paying license fees, and you're not paying for anything you don't need. So, over time, hopefully, it will prove that our AI is really there.  

Arthur Garnier (13:53)  

It's a really good point. And I think we resonate deeply with what we just discussed during the workshop on how exactly to build a chatbot, services, and solutions on top of your internal data. We're trying to share some lessons. that we learn together during this time. Basically, a lot of the work, we're going back here to the beginning of the discussion we just had on data governance, is really at the data and infrastructure level. Not that much anymore around the crazy breakthrough we had in AI; it's more now on data and internal data, When talking about internal data, the only one who can understand it, clean it, and own it is yourself. And so when you are doing 95% of the work, of course, going the last five miles, it's not that complex for a company that has relevant resources, a relevant IT team, a relevant data science team, and so on. And so you need something that we have been observing a lot. As you said, we are consistently monitoring the landscape of providers, trying to be as agnostic as possible to be able to repivot quickly. And it's something that we discussed , the ability to benefit really fast from a new breakthrough, also being able to get rid of obsolete technology quite fast, and not be too locked in with certain technology and a certain way of doing things, something that has been really key for Ardian in the building of GAIA. The last lesson that I've discovered, and maybe it's more related to our own culture since I'm not sure about what's going on out there, is that I've noticed the features that we have for GAIA actually come from bottom-up initiatives. We had this global vision of saying we want to go this way with our AI strategy and so on. But a lot of features actually came up through the users of GAIA, through innovation initiatives that, for example, we discussed during the Startup Studio. It was kind of an entrepreneurship initiative among the company where people were pitching their projects and so on. Some of the projects were actually built, deployed, put into production, and put into GAIA. And so what I've realized, compared to external solutions, is that when you also add these small layers, it doesn't necessarily need to be custom. Because at the end of the day, we are relying a lot on partnership technologies such as Microsoft, Mistral, and OpenAI. But when you add this small layer, you're also creating a cultural shift, and people are really eager and excited about the technology, about the tool that you're building. So they want to be part of it, and they really want to help shape the future of the solution and of AI in general within the company. And so I found it really interesting, because thanks to that, we're also able to get a lot of different pain points from the teams and really understand that, at the end of the day, our vision is really aligned with the operational reality of our business. This is really something I think is key for Ardian and for our portfolio companies and would be even more so in the future.

Elina Ashkinazi-Ildis (17:22)

I think you're absolutely right. And I think it was probably one of the key success factors in that. I remember early on, when we started working on GAIA, it was initially within the data team, but very quickly that changed. You went outside, you've identified the ambassadors, the people who are keen from all over different teams. It wasn't just investment teams. So that was really cool because that's a mistake that a lot of the companies make when they start building data and AI solutions: it stays within the technical team. They don't talk to business or, you know, they speak to them once at the beginning. “Ok. You want a dashboard?” And this is what made a huge difference because there are a lot of iterations. And if you remember, with early solutions, they would test them and they would come back with a huge list of things they were not happy with. As you said, it allows them to, on one side, be very involved: they feel like this is their solution, their voices are heard, they have a chance to shape this investment. On the other hand, it really helps the technical team because they were getting live feedback, and they could make the adjustments. This is definitely a key success factor. And this is as we talked about the operating model. How do we get the users within the company to work with the technical team, to work with the IT team, to work with the security team? Teams who're going to come back and say, “Well, this is all fun and games, but that's not how it's going to work”? So you have to really bring people from all of the different teams of the company to make this a successful project.

Arthur Garnier (18:56)  

Yes, it was all a huge transversal project. And I'm glad you noticed that on GAIA because it's actually really part of our data science strategy: data science needs to be embedded within the business. We need to make sure that we have a real impact, because it's not just about technology; technology was really an enabler. But without the people it's not useful per se. What's really useful, indeed, is how you use and leverage this technology.

And that's why we want to put it as close as possible to the people working across the different teams, whether they're investment teams or our corporate teams, and so on. Like, we want to have as much vicinity as possible between technical teams and business teams, to make sure that we are building the right product and something that's really helping shape their job.

Elina Ashkinazi-Ildis (19:52)

I think it's actually industry best practice. It's pretty normal when a company gets started with an AI topic to not have many resources, much money, or many choices. You have to have a centralized team and have them all in one place. But if you can take a step, if things work out well, if you can place the technical people very close to the business, that's where magic really starts happening.

Arthur Garnier (20:18)  

Now, on another note, in what will probably be our final point. What has also been very interesting is that a lot of people are talking about technology and want to apply it, especially AI, since it's really amazing for the outcomes you can get. They want to apply AI straight away in their day-to-day jobs, their pain points, their problems, and so on. But what we know is that, like we were discussing, there is the data layer first that you need to tackle. But there is also something that is not related to technology that different industries and teams have to tackle, and that's a bit more complex. It's directly linked to the business. I think you have a saying for that, at least one of the top model mottos of Artefact:  

Elina Ashkinazi-Ildis (21:05)

“Let's get process re-engineered”.  

Elina Ashkinazi-Ildis (21:08)

Exactly. Thank you very much. So indeed, what we have been observing, and we have also been discussing with a lot of our peers on this topic, is that the way AI technology and the way a firm is more efficient is not necessarily by bringing in a crazy amount of technology or a very complex workflow, but by re-engineering your process. And of course, you can re-engineer them, having technology in mind and how technology can help. But the way AI is really powerful nowadays is when you have well-optimized, efficient processes within the companies that are well thought out. And this is only then that you can apply AI on top of it and really see the magic happen as you said.  

Elina Ashkinazi-Ildis (21:53)

Absolutely. And this is actually a tough part of a fancy consulting firm: it is a fair bit of work and bring it down to a level where an agent can really handle all the steps. I have a couple of examples in mind. We worked for a major telecom company on one of the use cases we were trying to solve with agentic AI: is the management of a fiber-optic installation. So a technician goes to an appointment to install, and the address is wrong. So, he comes back, doesn't do it, then this information is kind of hidden in 25 different databases. It takes some time to figure out what happened and what the next step should be. As we started implementing agentic AI, we noticed that, in the workflow, when the agent would find the information and say, “the fiber optic was not installed”, and so the immediate answer would be reprogram the appointment and do it again. The trouble was that the address was wrong. So you had to go back and actually explain to the agent again, and an additional workflow to actually identify the reason why it hasn't happened. So the agent can't just jump to the right conclusion. So it has to be extremely precise. This morning, as I was thinking about the talk we were going to have, I realized that rethinking these processes is very similar to how you work with young children. When you're raising children, how you teach them the way to cross the street: “Okay, well, the light might be green, might be yellow. I'm going to see if no one's there. I'll go. I'll go straight. I'll go slightly diagonal. I'll adapt. I'm a human.” They are not mature humans, so when you're teaching a young child, if you give them all these varieties, that's going to end badly. So here you simplify the process: red, you don't move; yellow, you don't move; you see green, you look both ways, then you cross. So, we basically have to look at our processes and all the little things that we do, the corners that we cut, and things that we add on. It has to be taken out. We simplify it, we're ready to make it very precise. And that's when the agents can start taking over part of the job. And that's really tricky because you have to start with a macro process and actually have to go down to every nitty-gritty detail before you can actually go through. So you're right, once again, the technology is going to be a very small part, because there are already a lot of very powerful agent platforms you can use. But it's the first part that's going to be really tricky for any industry and for Ardian as well if you want to take it a step further.  

Arthur Garnier (24:42)  

For sure. And to summarize: process, data, and then AI.

Well, Elina, thank you very much. I think this was an interesting conversation, and hopefully we'll see you next year at our third edition of the AI for AI conference.