“Humans and machines make a very strong team together”

Artificial Intelligence has become mainstream. But will it also change the way we invest? It already does, says Huib Vaessen (Head of Research & Analytics Real Assets at APG). Six years ago, he stood at the cradle of ‘Samuel’. This digital portfolio manager provides recommendations and thus improves the investment decisions made by its physical colleagues. “As a human, you constantly get challenged by this digital colleague to thoroughly justify your decisions.”

Making a decision on whether or not to invest in something is done in an increasingly data-driven way, Vaessen explains. “Samuel plays a growing role in this, because, for example, it offers insights based on data. The role of these digital workers is growing. Indeed, Samuel will be increasingly better able to help the real estate team making (investment) decisions. Those decisions will keep getting better. That’s because Samuel doesn’t forget anything, so its knowledge will keep expanding. For some frequently, more straightforward, recurring decisions, humans will only review and check the proposals.”

Storing data (read: knowledge) is done in a very innovative way, Vaessen continues. “We made sure that Samuel not only stores the data itself, but also the context, such as what the data was used for, where it came from, what the definition is and when it was saved. This is all done in such a way that the digital assistant can perform its own calculations on it. Thanks to this so-called knowledge graph, much more data can be stored than in the traditional way.”

ChatGPT
Here, large language models that power applications such as ChatGPT play an essential role. “This development ties in seamlessly with the knowledge graph ‘Samuel’s Brain,’ making it a nice extension. This ‘brain’ stores both data and the logic regarding suggestions for investment assumptions or ultimately decisions that are derived from this data. Now you still need a specific query to extract the data you need from Samuel, and then a dashboard to interpret the data. Not everyone can do that, so not all the information is extracted from Samuel.” By integrating a large language model with Samuel, anyone can ask their digital colleague specific questions in plain language, Vaessen explains. “For example, about how much money we have invested in the Netherlands. This makes this knowledge more widely accessible. Ultimately, the goal is to have this digital portfolio manager answer increasingly complex questions. Such as; which companies we invest in are affected by increased interest rates? We are not at that point yet, but APG is leading the way when it comes to structurally storing data and using it in our decisions.” ​​​​​​​

Vaessen describes an investment team as a decision factory. “Over time, in tandem with the investors, Samuel also makes a decision. It may suggest option A, while its physical colleague ends up with option B. Then the investor can figure out what the difference is. Maybe Samuel didn’t take into account the consequences of inflation and that’s why the investor stays with option B. But as a human, you constantly get challenged by this digital colleague to thoroughly justify your decisions.”

Humans and machines
The introduction of Samuel has led to a nice combination of humans and machines, Vaessen said. "Machines are very consistent and can integrate much more data when making decisions than humans can. But humans are capable of contextualizing and augmenting such outcomes.” Samuel also depends on what his physical colleagues record in the algorithms he uses. “We do have to keep thinking about whether those algorithms are still relevant, and we can’t blindly rely on Samuel’s output. Thinking about what’s under his hood remains essential, and that engine will need to be tweaked from time to time.”

That the role of the digital assistant is growing is obvious, according to Vaessen. “It will certainly take care of more and more decisions. Gradations do exist between investment teams. For example, there will be teams where most of the suggestions from a digital colleague like Samuel will be followed, while in other teams — where decisions are more complex and less frequent — this will happen less.”

Part of the reason Samuel was first used for real estate investments is because of the size of that team at APG. “As a result, more of the budget is available for launching these types of initiatives. Real estate as an asset category also lends itself well to a digital assistant. The team manages about 150 investments, consisting of some 30,000 buildings in total. A systematic approach such as Samuel offers is very welcome for that purpose.” The digital assistant is now also being introduced to the infrastructure team and will subsequently show up in more and more teams.  ​​​​​​​

Work opportunity
Samuel’s arrival needs not directly affect employment within the investment teams, Vaessen believes. “Every investment team works with a certain budget, which is used to make the best possible decisions. That budget is not necessarily to decrease, however we can realize better results with the same budget. The way the budget is spent will change. Whereas before the investment result was generated by putting a group of experts together, now it is increasingly done by a systematized process where human experts work together with their digital assistant. So, more of the budget will be available for programmers and other ICT personnel, reflecting that skills sets will evolve, rather than necessarily impact overall headcount.”

Not every colleague was as enthusiastic about their new digital assistant six years ago. “In the beginning, people had to get used to it, but people also know that our clients increasingly want to understand the decisions we make and that it is expected that you increasingly use more data to do this. If you don’t systematize that information, you have to look it up every time. And keeping track of that in Excel sheets is not fun work. Samuel can also take over repetitive work. People are grateful for that.”

The latest developments in AI are showing people how much the computer can already do, Vaessen concludes. “This also means that digital colleagues become more and more knowledgeable when using the latest techniques. As humans, we should always consider how best to cooperate with the computer. Together, humans and machines are a very strong team. Only together can we achieve the best results for our pension fund clients.”

For more information on various applications of AI in practice, read the chapter Vaessen wrote in the Handbook of Artificial Intelligence and Big Data Applications in Investments.