In December 2023, we had our first workshop on AI. We set aside a whole day to dive deep, bringing in an external expert. It was one year after ChatGPT was launched. Now, three years later, we've become a bit wiser, but still feel we are quite early in the development here. In this presentation, we will go through three things. First, we will share some of the success criteria for implementing AI in the fund. Then, move on to some reflections on how we have used and feel about content development in communication, and then Sara will give a deep dive into one of our US cases.

Here are three success criteria that we feel have helped us implement AI across the entire fund. And I think one of the most important things, which you got a little taste of at the start, is that leadership must be involved. They must set the direction, and thereby get the whole organization to change and change the way we work. We had an event at Arendalsuka earlier this year and now want to show a short clip of how leadership thinks. I think one must do if one is to lead this extremely important development is that you must attack it from all sides. You must have a crazy person at the top who is on it all the time. Every time I get the microphone, every time we deal with something in the fund, I stuff everyone's ears full about how AI is what matters. It is AI that matters. It is AI that matters.

So, all the people you hire, the young people, they must be totally AI-minded and they must be really keen on this. Because what AI has done is to turn the whole pyramid in the organization upside down, so that it is the older people who should learn from the young and not the opposite. So you must have ambassadors. We now have 40 ambassadors in Oljefondet who are trained by a company in America called Tropics. Several times a week, they receive training in this, so that you have specialists in all parts of the organization who PC all their colleagues. And then you must have events all the time.

You must have Techweek, Tech Year, you must have trophy talks, you must keep going all the time. You must simply, totally, saturate the organization with initiative so that you increase the pace.

And so on, the days go by here. Second, knowledge sharing across departments is important. Nikolai mentioned the ambassadors. We now have 70 AI ambassadors with representation from all departments. We have two people in our communication department. They meet regularly to be updated on the latest developments in various technologies. They share use cases and learning, which they then take back to their teams and implement good US cases. So that has been an important success factor.

Third, one depends on being updated and having the best technology. We feel it is really important. First, perhaps to attract the best people, but also to be able to get the best utilization. And what we have is what we call our three horses. Cloud, made by Antropic, is our AI assistant. It is where we input and link it up with our company data and have various projects and daily savings. Second is Cursor. It is mainly for those who work with development and programming, but it is so easy to use now that we see more and more people. Now, a third of the organization writes code. And it has actually become four horses now. Because Cloud Code has also come in here. And the third, which most people here have probably heard of, is Copilot, which we also have available for all employees. And we see good development there too, but feel that there is a lot to improve in terms of maximizing the potential there.

And we are working with all these initiatives here to become more effective and increase quality. And we have set ourselves a goal of 20%, and that is 20% every single year. And that number was thrown around a lot in meetings. And we wondered a bit where it came from. And we found out. When you have the cost of intelligence coming down so dramatically like it is now, what is it going to do to productivity in the world? I mean, supposed to go up a lot, right? >> That's what theory tells us and >> that's what I think. >> So, so eh I've told everybody in in our company that hey, we should we should improve our productivity by 10% of the next 12 months all. And that's and you know how I got the number. Did you ask? No, I just I just took it straight out of the air. Do you think what do you think about that number? Is it low high? Underambitious. Underamb. What should What should increase by? How do you how do you measure the stuff we do? me just kind of stuff that I produce. How much of your company writes code? Eh, 15 people in in technology probably 15 20% of us more actually. But okay let let's say that's 20% writing code. I think in overall goal of like you know 20% productivity increase in a 12 month period is appropriately ambitious given the tool and given the tools that we will launch over the next 12 months. >> Ok. Sounds like I should up the game her a bit. >> I think so. >> Yeah. I'll just tell everybody you told me to. So that's fine. >> Ja.

So, the 20% figure was established. And this was August 2023. So, some experiences from us in communication. So, we decided to go all in on AI. We were going to see how we can best utilize the technology. But experience we gained quite early was that we have to take a step back and actually work with digitalization. We have to find where all our data is? Is our data structured and are we able to use it correctly, and then we can add AI on top of that to get the best utilization of this. So we also notice that this is not going to remove us from our workplace. AI is middle to middle, not end to end. So we are absolutely critical to have, in a way, our knowledge and critical sense to help AI produce the best output and have human in the loop all the time, as we say.

And third is to maintain momentum. Things are happening all the time. What you perhaps couldn't achieve a year ago, you might easily achieve now. So, we have good ambassadors who maintain momentum, and events, and all that Nikolai mentioned, mean that we suddenly find out that this is actually working, and better than we thought it would not work a few months ago. So, maintain that momentum. There is a tremendous amount of product development happening here all the time. And it is quite critical and we have found it important to push all the time.

And four things that we feel we have done quite well is we have developed a mail assistant, which is an application in Outlook that helps us respond to many inquiries we receive from external parties. We now have 150 people out talking and receiving a lot of requests, and then we get an autogenerated response as a suggestion for us based on our criteria. So we have now implemented that and are using it.

Then, when we have agreed to an event, we have now created our own event platform, which has not been launched yet. It has been a bit of a big project. But then we go from, in a way, an Excel sheet to a dedicated application, and then we can spice up AR on top of that.

And then we have talked about Cloud, where we have worked on creating good cloud projects, which is like GPT for those who use the OpenAI model. And we have regarding social media posts, how you write intranet articles, things that we do often and that the rest of the organization can also utilize. So we have worked a lot on that. And then perhaps the crown jewel is Eko, which Sara will now talk a bit more about.

Yes, thank you. So Eko, it all actually started in May 2024 when we had a discussion with the management team about our priorities going forward. But they felt they lacked data, they lacked an overview of everything we did. So our task was then to come back in three months with a complete overview. And that became EKO. So Eko is a dashboard that gathers data from all our communication activities under one roof.

And now we will share some of the experiences we gained by creating it. Because the problem was not that we lacked data. We had tons of data. The problem was that the data was isolated on different social media platforms, in Excel sheets, on various desktops and SharePoint pages. So, to get an overview of all activities requires you to log in to different places, you have to ask people who have a file that you don't have access to, so it becomes difficult to get an overview. And you become very dependent on those people with login credentials.

So the solution was to gather everything in one place. And there our choice fell on Snowflake. And there is the first lesson learned. And that is to take control of your own data. And it's not like Snowflake itself is magical, but the principle is important. Get the data in a place where you have control over it.

Then you can combine it with other data and you can use it in ways that the vendors don't think of. So much of the work in the beginning was not about AI or fancy dashboards. It was about APIs and data flow. But it is building the foundation, and it is incredibly important. But once that was done, AI came into the picture because then we had to find a way to present it. And I am not a developer. Nor is anyone in our department, but with the help of AI tools, such as Cloud Code, good colleagues, and stealing code without shame, we were able to build this in just three months. And there is the second lesson learned.

Because even though it's starting to get boring to hear it, AI technology is democratizing, and you no longer need to be an expert to develop. You need to be curious and you must dare to try and tolerate failure. So the result was then this dashboard Eko. And the dashboard itself is not that interesting, but it is what happens when you gain control over the data.

You can combine it, you can see new connections, and you can share with more people. Because it is not just us in the communication department who have access to Eko. The entire organization has access. And they also have access to the underlying data so that they can use it in their own reports. And we get a lot of information. And here you see an example just so you get a little look at how it looks. And this is then Jammer or Viva Engage, which we use as our corporate social media.

And previously, to access this, you had to be an admin in a community on Jammer, and then you could go in and see stats. I think it was a maximum of three or six months back in time and sort of trends for the individual community. You could not see user statistics for the use of the entire platform. Then we had to ask someone in the technology department if they could go into the backend, fetch a report to Excel, send it to us by email, and then we could sit and combine several Excel reports to see the trends. So it was exhausting.

Now we have all the data instantly and we can follow the trends. So if we see that usage starts to drop, we can take action immediately. And that saves us a lot of time. And then we have corresponding overviews for everything else we do.

And it is useful, but it is a dashboard. It is static, so it is limited. It answers the questions we have already thought of asking. So therefore, we put a chatbot on top, which we have called Eobot. And it has access to all the data and can therefore answer all possible questions. So we say that Eko largely gives us an overview, but perhaps Eobot gives us a bit more insight.

And here is the third lesson learned. It is that documentation is key. It is not enough to just have the data. AI must understand the data. So you must give it context. So we have told Eobot that when we look at the figures for our podcast, it must combine two different sources. And we have explained things like that, so that with the context, Eobot can give us reports in seconds and answer questions we haven't even thought of asking.

Here, for example, I ask if it can compare engagement on our different social media platforms. Previously, this would have required us to log into the three different platforms and retrieve data there. Now, this is done automatically. We get a summary of the last six months. Breakdown per platform and then some recommendations. It really likes to give recommendations, and then you decide for yourself whether it is useful or not, but it saves us a lot of time at least. But it also changes how we work. We go from spending time fetching data and reporting to understanding.

So going forward, we will focus on collecting even more data and creating more tools so that Eko can become a central hub that can help us with everything from readability of external reports to podcast research. And the first project we are about to complete is our own sentiment analysis. We have set up seven agents, simply put, seven different LLMs with their specific mission, analyzing media articles about the fund and categorizing them according to given criteria or in given categories.

And this is not new in itself. Many of you probably have it. Some of you make it. Some of you buy it. But the advantage of doing it yourself is that we know ourselves best. We know what information we are interested in. And we can combine it with other data. Here you see a very clear prototype of this, and it's a lot of information, but it's actually not important at all for my point, it's just that it has full flexibility and we can do it exactly as we want.

We see what works and what doesn't work, and then we can change it super fast. So I just want to reiterate the learning points. It is to take control of your own data, move into an environment where you have control yourself. AI makes it possible for non-technologists to build. We have managed it. Everyone else can manage it. And documentation is everything. Structured, well-documented data is what makes AI truly useful.

So as Sigurd said, AI is not end to end, it is middle to middle. And what we have done is cut time and costs on fetching data. And then we use the time to try to understand it.

And finally, we think it will be critical for us communication advisors to be critical of what we get out of the elements and not take it for granted. Our input here elevates the output to a new level. Going forward as well, it is in relation to recruitment and what kind of skills are needed to be a good communication advisor changes and to be a bit more like Sara, I think is the future. And then comes AI integration.

The way we are going to tackle this more going forward now in 2026 is to look at what are the big processes we are working with and how can new technology, when you have mapped out every single step in a large process, streamline, automate, and improve.

And then there are new opportunities that this technology provides. And for us, a few years ago, to be able to create our own sentiment analysis, tailor-made for our needs, I would have seen as completely impossible. Now we achieve it. And I think there are many such opportunities that will come. Thank you for us. Over to Halvor.