While the internet was becoming mainstream in the late ’90s, Milo Mitev was seriously researching something that hadn’t caught on for decades: AI.
Mitev, now a wealth manager, was an early adopter of AI in the financial sector, having discovered the power of neural networks in 1997 while studying at the Vienna University of Economics and Business.
He told CNBC that he sees the potential of neural networks in financial forecasting. “I was hooked on these possibilities,” he said.
Mitev has spent his 25-year career forecasting for banks and technology companies like Siemens. He founded SmartWealth Asset Management. Its decisions are made entirely by a network of AI systems. Its latest fund, IVAC, aims to have $2 billion in assets under management and has a target return of 14% to 15% annually.
Even though humans are not involved in the AI’s decisions, Mitev said that “the humans are the most important part of the equation” because they select the training data, enter the variables, build the parameters, and consistently fine-tune the model.
Once the model is created, “it’s very risky to start intervening,” Mitev says. In fact, he added, trusting the model is his golden rule.
Instead, humans must ensure that the data and calculations are error-free and introduce new data to ensure the model is up to date.
“The worst thing you can do is reverse the results, which happens very often,” Mitev said, adding that people “don’t trust” AI at first. “Even if we don’t know the results now, we look back in two or three months and think, ‘Oh, we were actually wrong,'” he added.
The forces that drive markets, such as optimism, pessimism, and speculation, are very human. Even the European Central Bank has warned that the current AI bull market may be driven less by detailed technical analysis and more by fear of missing out.
Mitev said removing emotion from investing has been proven to yield better results. SmartWealth Asset Management has recorded a return of 407.63% for the 10-year period ending November 1, 2025, compared to the industry benchmark’s return of 145.34% for the same period, according to a graph shared with CNBC by a company representative.
Mitev said it is “impossible” to know what will happen a year from now, but his model can look up to a month ahead. “Assessing this information and making informed decisions based on it has consistently been shown to produce better results than humans.”
Continuous monitoring and introduction of new data are key given that AI systems can “hallucinate” or generate false information. Maitev said the model’s errors were due to “overfitting,” data issues, and misspecification of the model.
Overfitting is when an algorithm pays too much attention to what Mitev calls “noise.” He said this was “meaningless” data because it did not reveal the true causal relationship with stock price performance.
Rigorous design, validation, and real-world environmental testing serve as an antidote to this, Mitev added. This means that even though his fund strategy is executed entirely by a set of algorithms, humans still play a key role in making sure it is effective.
“The reality is that this is a process that evolves over many years…and this is why in-house development of this kind of technology is so important,” he added, especially for those looking to differentiate their AI efforts.
