Wealth firms are sleepwalking into an outsourcing trap, as they lose control of external vendors and the impact they have on client-facing business

We are in an artificial intelligence (AI) bubble. Many wealth and asset management firms are capitalising on the ‘AI’ buzzword, claiming it is transforming their businesses. The majority are overstating the transformation and are repeating common, avoidable mistakes during integration.

AI is indeed transforming businesses across the board, and asset and wealth management is no different, but it very rarely shakes anything to the core.

To quote the accidental management consulting guru Donald Rumsfeld, for most managers, AI is seen as a “known unknown”, but it is actually an “unknown unknown”.

A recurring and costly misstep across the asset management sector is the blind reliance on external, closed-source AI models. Leveraging household large language models (LLMs) can accelerate deployment, but this approach comes with costly trade-offs.

A recurring and costly misstep across the asset management sector is the blind reliance on external, closed-source AI models

The major concern is the lack of control. When a vendor owns your AI infrastructure, you are outsourcing your core investment judgment. Even a minor model update can drastically alter an AI model’s behaviour and outputs.

For asset managers, any changes to the core model can introduce an unmanageable risk of overfitting, whereby a model learns underlying data patterns but not the logic that drives them. This can cause it to lead firms astray by performing well on past data, but poorly on unseen, real-world data. Reversing this can then mean rebuilding the whole model from scratch, driving up costs and wiping out historical results, with huge implications for investors who rely on stable, proven performance.

The external cost of running the back tests to train an AI model is another limiting factor. At a recent Swiss investment event, a firm disclosed that it spends SFr5,000 ($6,200) to run a single model back test using ChatGPT. Proper model vetting requires hundreds of runs, and at that cost, most firms simply won’t thoroughly test their models. As rigorous testing becomes unaffordable, firms will be incentivised to cut corners. Given that proper out-of-sample validation requires testing hundreds of hypotheses, this cost structure can quickly become unsustainable.

Building a complete AI model in-house requires more time, more resources and more talent, but it’s the only path that allows firms to have a real strategic edge. To ensure the model’s efficacy, it is important to develop models able to capture market dependencies and causalities across multiple market cycles, which can be on average over seven years long. Those who make this initial investment reap benefits in strategic independence. Owning your own AI model is essential for firms that are aiming to differentiate themselves in a market increasingly defined by short-sighted automation.

Even with high-performing in-house AI models, human managers often second-guess the data and undermine results. True AI adoption is psychological and half measures kill performance.

Too many asset management firms let human investment managers make decisions on the AI’s recommendations. One key benefit of using AI in asset management is removing human biases from the investment process. If it is not trusted fully, one human override can break that compounding chain and dilute returns.

Too many asset management firms let human investment managers make decisions on the AI’s recommendation

It is essential to resist the urge to intervene, even when the market wobbles. Too many asset management firms expect AI to be right 100 per cent of the time, then panic when it is ‘only’ right 60 per cent — even though that still vastly outperforms most active managers, 80 per cent of whom underperform their benchmarks. Viewed through this lens, AI offers a significant long-term advantage and outperforms human decision-making not by being perfect, but by being consistently better year over year.

One of the most overlooked obstacles to effective AI adoption is how brutally competitive the market for AI professionals is. The fight for talent is repeatedly referred to as a ‘trade war’ in the industry, leading to a rapid escalation in wages, with reports of Meta offering sign-on bonuses as high as $100m to top AI researchers.

This poaching of talent makes building AI teams very difficult, and the departure of a key mathematician, analyst, engineer or data scientist can massively derail progress. The praxis shows that after a steep three-year learning curve, key team members often leave for rivals or start their own firms. Salaries alone are not enough — no matter how generous the offer, someone else can, and will, outbid.

AI systems do not run themselves. They are not ‘plug-and-play’. They need continuous refinement, and continuity in teams is critical. Without that, even the best models will stagnate and lose progress. Therefore, some asset management firms have adapted by rethinking their team structure entirely. Instead of hiring employees, they bring on partners, finding that giving team members equity and a direct stake fosters commitment.

AI is an equaliser, and access to the technology is not the real differentiator anymore. The real edge lies in how firms choose to utilise it. The asset managers that succeed will not be the ones chasing trends or outsourcing systems. They will be the ones willing to commit to building in-house and prioritise retaining key talent.

There is no shortcut with AI. It demands long-term thinking and hard work, but the payoff is real. The key question facing asset managers is whether they bolt AI onto their outdated models or if they choose this moment to reinvent their systems and build something that can last for decades. The firms that get this right will not just keep up — they will set the pace.