Moving Beyond the AI Pitch: Asset Managers’ use of AI 

For asset owners and allocators, looking beyond the marketing and well-rehearsed pitches and understanding not simply what an asset manager is doing with AI, but how they are doing it, is critical. After all, “we use AI” or “powered by AI” can mean anything from basic automation to production-grade models that influence live portfolios. Effective due diligence should determine whether a manager’s use of AI is purposeful, credible, well governed, empirically validated, delivering value at speed and sustainable at scale.

Mercer’s 2026 AI in Asset Management Survey^[1] shows adoption is real but uneven: 55% of asset managers report AI is integrated in at least one of their strategy’s investment processes, 27% are at pilot/proof-of-concept, and only 18% report no integration yet. This dispersion matters: investors should pin down what is truly in production, what remains experimental, and how the manager defines “success.” With 91% of managers responding that they plan to increase their use of AI in the next 12 months it’s a topic that cannot be ignored.

As AI becomes an increasingly common topic in engagements with asset managers, below are some areas to consider when asset owners and allocators are assessing the use of AI in managers’ investment processes.

These questions aim to distinguish between business-driven applications and adoption that is primarily exploratory or for marketing purposes. Strong answers articulate a clear hypothesis for how AI may improve outcomes (e.g., signal discovery, speed, coverage, cost reduction, or risk management) and where it is applied (idea generation, forecasting, portfolio construction, execution, etc.). 

The Mercer survey results suggest today’s most common “already integrated” use cases are idea generation/research, processing unstructured/external datasets, and signal generation/market trend analysis. Far fewer asset managers report AI embedded in portfolio construction or trade execution. As a result, mapping where AI sits in the workflow is not just documentation – it clarifies whether AI is mainly expanding research coverage and analysis, or directly shaping exposures, sizing, and trading behavior.

Our survey indicates most firms describe their AI integration as being operational (74% using for automation/efficiency) and/or “co-pilot” (69% using for insight/analysis), with only 6% saying AI is used for decision-making. 

When managers cite “AI-driven outperformance”, it is worth understanding what ‘outperformance’ means and calibrating expectations. In the survey, the most common measurable benefits were enhanced operational efficiency (69% cited this benefit) and faster/higher-quality insights or decision-making (55% saw these improvements). Improved returns and reduced risk/volatility were cited far less often (8% each). 

Operational sustainability matters as much as innovation. The choice of AI model will determine output explainability and robustness, as well as operational complexity, risk and monitoring needs. A credible response from the manager can clearly describe the model architectures or techniques used, what are the key features/inputs and why these models were selected over alternatives. Managers should also be able to explain practical trade-offs between performance, interpretability, and operational risk. Weak answers rely on buzzwords. 

A key difference between prototypes and production-grade systems is governance, measurement discipline, and ongoing monitoring. Without these, back tests can mislead and performance can degrade unnoticed. Investors should assess whether the manager uses rigorous methods, controls for common pitfalls (overfitting, look-ahead bias), and can demonstrate ongoing reliability. They should determine whether the operating model is durable with clear accountability, adequate staffing, third-party risk management, and suitable resilience controls. 

Vendor dependence can create concentration risk, and weak governance can introduce model, cyber, and continuity vulnerabilities. Understanding which components are vendor-provided (data, models, infrastructure) and what due diligence is performed on vendors is important. Consider data sources and licencing, what are the usage rights and auditability? Weak data governance can create operational, reputational, and regulatory risks, and can lead to fragile models or hidden exposures (licensing, privacy, and bias).

Our survey showed 63% of managers make use of vendor’s off the shelf AI tools, with 51% using vendor tools with some proprietary customisation. When it comes to data, 58% are using some vendor provided data. Over two thirds (69%) of respondents cited data quality/access as the main barrier to further AI adoption.  

Our survey also suggests many programs operate with lean internal resourcing with most firms reporting only a small number of dedicated AI specialists within the investment team (57% citing 1-5 full-time employees dedicated to AI development, implementation or oversight). 

Assessing AI is not about how advanced a manager’s technology sounds. It is about whether AI addresses a defined investment problem, is supported by robust governance, is validated with empirical discipline, is monitored in production, and is operated under clear governance with resilient controls.  

When assessing asset managers, it is critical to distinguish durable, repeatable AI capability from marketing narratives and unmanaged risk.