There’s a growing urgency in asset management and trading to invest in artificial intelligence. But this task must be undertaken carefully.

Michael Heldmann, CIO for the U.S. Systematic Equity team at Allianz Global Investors, wrote this article on AI in investing for MarketWatch:

There is an urgency for asset managers to invest now in AI because there is a strong correlation between investing in technology and profitability: Between 2014 and 2017, asset managers with profitable growth increased technology spending 7% annually, focusing that investment in proprietary technology, compared to only a 2% annual increase for the average firm, according to a study by Casey Quirk. The report notes that data today remains “probably the least optimized resource across most asset management firms.” That’s why Boston Consulting Group warns that firms must invest now to harness AI and big data or “they will fall even further behind the front-runners.”

As a practitioner of quantamental investing (quantitative research paired with fundamental analysis), more than two decades experience with AI has taught me that the best use for this technology in asset management is to incrementally improve existing investment processes and methodologies.

That starts by accepting that AI is not designed to hit stock-picking home runs, find needles in haystacks or to work independently. Instead, it works best when humans develop an investment thesis and machines test that theory. For example, AI could help identify a cohort of 100 stocks that statistically exhibit certain characteristics and a human portfolio manager could improve performance by excluding potential bad apples among that cohort, based on context and experience. The end result is tilting a portfolio to favor stocks with the highest statistical probability of outperformance while underweighting stocks with a lower probability. The goal is not a “eureka” moment but consistent, diversified, additional gains.

What’s more, asset managers must avoid the temptation to look for short cuts by hiring external firms for this sort of work. Building AI capabilities is better done in-house because AI-driven algorithms need constant human supervision, feedback and adjustment. At the same time, incentives between investment teams and external vendors are never correctly aligned. Vendors have an incentive to develop solutions that look good on paper but may be short-lived or the product of problematic data mining. Similarly, to encourage collaborative and agile progress, it’s better for AI teams to work in close proximity with investment professionals, not housed in far-away centers of excellence in remote locations.