Case Studies

Situation

A sovereign wealth fund in the Gulf was expanding its allocation to quantitative equity strategies. Five managers on the shortlist all claimed artificial intelligence as a core component of their investment process. The presentations were polished, the backtests were strong, and the language was similar enough across pitches to make differentiation difficult.

Challenge

The fund's investment committee had deep experience in traditional asset classes but limited internal capability to assess AI-specific claims. Standard due diligence (track record analysis, AUM checks, team backgrounds) could not answer the question that mattered: were these models built on genuine economic relationships, or on patterns that would disappear out of sample? The committee needed a structured way to evaluate claims they could not independently verify.

Approach

Working directly with the investment committee, I applied the SPEC framework to each manager's strategy. Each evaluation focused on four dimensions: whether the model's specification had a causal economic rationale, how the manager diagnosed periods of underperformance, whether they could explain their model's behaviour in plain language, and how sensitive the strategy was to reasonable changes in assumptions. Interviews with each manager's research team were structured around these dimensions, with a standardised scoring rubric that allowed direct comparison.

Outcome

Two of the five managers could not provide coherent answers on model specification: their strategies were effectively curve-fitted to historical data without economic grounding. A third had strong technical foundations but could not articulate them to a non-technical audience, raising governance concerns. The committee narrowed its allocation to two managers with demonstrably sound structural foundations. The evaluation framework was subsequently adopted as a standing component of the fund's manager selection process for all quantitative strategies.

Situation

A Nordic pension fund had, over three years, increased its allocation to AI-driven quantitative strategies from under 5% to nearly 15% of its total portfolio. The board, composed primarily of professionals with backgrounds in traditional finance and public policy, recognised that its oversight capability had not kept pace with the shift in portfolio composition.

Challenge

Board members were asking the right high-level questions ("how do we know these models work?") but lacked the vocabulary and framework to interpret the answers they received from managers and internal staff. Reporting from the quantitative strategies team was technically accurate but effectively opaque to the board. There was a growing gap between fiduciary responsibility and the ability to exercise informed judgment on a material portion of the portfolio.

Approach

I designed a governance framework tailored to the board's composition and the fund's regulatory context, with particular attention to EU AI Act requirements. The framework had three components. First, a structured question set for quarterly manager reviews: plain-language questions that nonetheless probe structural integrity, not just performance. Second, a traffic-light reporting system that translates complex model diagnostics into board-level signals: green (model behaving within expected parameters), amber (deviation requiring discussion), red (structural concern requiring action). Third, a decision tree for when the board should escalate to external technical review versus when internal oversight is sufficient.

Outcome

The board moved from reactive oversight (responding to drawdowns after the fact) to proactive governance (identifying structural concerns before they manifest in returns). Manager reviews became more substantive, with managers themselves noting that the fund's questions had become markedly more sophisticated. The governance framework was presented to the fund's regulator as part of its compliance documentation and was cited as a model for board-level AI oversight in the Nordic pension sector.

Situation

A Western European central bank was exploring the integration of machine learning techniques into its reserve management process. Several internal research papers had demonstrated potential improvements in forecasting and risk estimation. The question before the reserve management committee was not whether AI had potential, but where precisely it could be introduced without compromising the conservatism and transparency that reserve management demands.

Challenge

Central bank reserve portfolios operate under constraints that most institutional investors do not face: extreme conservatism, near-absolute transparency requirements, and political sensitivity around any perception of speculative behaviour. The standard AI integration playbook from the asset management industry was fundamentally incompatible with these requirements. The committee needed a framework for identifying specific, bounded applications where AI adds demonstrable value without introducing opacity or model risk that could not be fully explained to oversight bodies.

Approach

I mapped the bank's entire portfolio construction process and identified four candidate integration points: liquidity forecasting, currency exposure optimisation, credit risk monitoring, and rebalancing timing. For each, I assessed the trade-off between model complexity and explainability, using a modified version of the SPEC framework adapted for institutional contexts where the primary concern is not alpha generation but operational soundness. Two of the four integration points were recommended for pilot implementation, with the other two flagged as introducing model risk disproportionate to the expected improvement.

Outcome

The bank implemented machine learning for liquidity forecasting and rebalancing timing, both areas where the models could be fully explained to the oversight committee and where performance improvement was measurable against a clear baseline. The other two applications were deferred pending further research. The assessment framework I developed became part of the bank's internal technology evaluation process, applied not only to AI but to all proposed quantitative enhancements to the reserve management process.

Situation

An Asia-Pacific sovereign wealth fund with significant allocations to global quantitative strategies was concerned about the implications of escalating technology competition between major powers. Export controls on advanced semiconductors, sovereign AI initiatives, and diverging data governance regimes were creating a new category of risk that did not exist in the fund's allocation models.

Challenge

The fund's risk framework captured market, credit, and liquidity risk, but had no mechanism for assessing technology supply chain vulnerability at the manager level. A quantitative manager reliant on cutting-edge GPU infrastructure for model training could face a material disruption from export controls, but this risk was invisible in standard reporting. The fund needed to understand which of its managers were exposed, how severely, and what this meant for portfolio construction.

Approach

I conducted a systematic assessment of the fund's quantitative manager universe through the lens of AI infrastructure dependency. For each manager, I mapped compute supply chain (hardware sourcing, cloud provider concentration, geographic location of training infrastructure), data governance exposure (cross-border data flows, regulatory jurisdiction of training data), and talent pipeline vulnerability (concentration of key researchers in jurisdictions subject to visa or employment restrictions). This was combined with a macro-level assessment of how sovereign AI initiatives and export control regimes were likely to evolve over a 3-5 year horizon, drawing on my work with the OECD and engagement with policymakers.

Outcome

The analysis revealed that three of the fund's twelve quantitative managers had material concentration risk in compute infrastructure sourced from a single jurisdiction. One manager's entire training pipeline was dependent on cloud services that could be affected by proposed regulatory changes. The fund used these findings to initiate conversations with affected managers about infrastructure diversification, and incorporated technology supply chain assessment into its ongoing manager monitoring framework. The geopolitical risk overlay was subsequently applied to new manager evaluations as a standard component.

Situation

A large North American public pension fund with a multi-billion dollar allocation to quantitative strategies initiated a comprehensive review of AI integration across its entire portfolio. The catalyst was a series of board questions following media coverage of AI in finance: how much of the fund's quantitative allocation genuinely used AI, and how would the board know if AI-related risks were materialising?

Challenge

The fund faced a problem of scale and consistency. It allocated to over twenty quantitative managers, each using different terminology, different reporting formats, and different levels of transparency about their AI capabilities. Some managers had genuine machine learning infrastructure. Others had added "AI" to their marketing materials without meaningful changes to their investment process. The fund had no standardised way to compare AI claims across managers, no governance framework for board-level oversight of AI-specific risks, and no clear view of how AI integration (or its absence) should affect portfolio construction decisions.

Approach

The engagement was structured in three phases over six months. Phase one was a systematic evaluation of every quantitative manager using the SPEC framework, producing a standardised assessment that allowed direct comparison across the entire portfolio. Phase two developed a governance framework for the board, including a quarterly AI risk dashboard and a decision protocol for when AI-related concerns should trigger manager review, reallocation, or termination. Phase three translated the findings into portfolio construction recommendations: which managers warranted increased allocation based on structural soundness, which required enhanced monitoring, and which should be placed on watch or redeemed.

Outcome

The review reclassified the fund's quantitative managers into three tiers based on AI integration integrity, a categorisation that differed significantly from the fund's existing performance-based ranking. Four managers previously considered strong performers were flagged for structural concerns that performance data alone had not revealed. The board adopted the governance framework and quarterly dashboard as permanent oversight tools. Portfolio reallocation based on structural assessment rather than trailing performance was implemented over the following two quarters. The fund's CIO subsequently described the engagement as having "fundamentally changed how we think about our quantitative book."