Lowering the Cost of Alpha: A CIO’s Framework for Integrating Humans and AI – CFA Institute Enterprising Investor

The active asset management industry has reached a breaking point. After decades of benefiting from high fees and growing assets, active managers now face brutal margin pressure. Passive investing has eroded income, while the costs of producing alpha remain stubbornly high due to large teams, complex data needs and heavy infrastructure.

While some companies have been able to reduce absolute costs through traditional budget cuts, these savings rarely keep pace with the relentless margin squeeze. With additional burdens due to regulations, cybersecurity and technology maintenance, companies are under structural pressure: declining reimbursements and weak inflows on the one hand, rising or inflexible costs on the other. The battlefield is no longer just performance, but the cost of alpha.

Technology is supposed to solve this, but in many cases it has done the opposite. Years of investment in AI and automation have failed to reduce costs, as most companies remain locked into an outdated architecture that consumes resources and imposes an increasing complexity burden.

Much of today’s technology spending simply maintains existing systems (often 60% to 80% of total technology budgets), leaving little room for innovation. Even when modern tools are introduced, human resistance often limits their impact as portfolio managers and analysts fear loss of control or job relevance.

For CIOs, the real transformation is cultural: success comes when AI is used to empower experts, not replace them, freeing teams to focus on the highest value decisions.

Blueprint for a cost-effective Alpha Factory

There is a high opportunity cost associated with highly compensated portfolio managers spending time manually collecting data rather than making high-quality assessments. The industry is full of talk, but short of actual, working blueprints.

So how can asset managers escape the clutches of fees, generate sustainable alpha, break free from the old trap and engage their people? The solution is to reimagine the investment process itself and build a new kind of alpha factory that is highly efficient and scalable, but still places human expertise at its core.

Based on over two decades of experience managing institutional portfolios (over €1.6 billion under management) and designing Human+AI investment processes, I have designed and tested a specific end-to-end blueprint that reduces the cost of alpha by addressing these root causes.

For example, during a live run in early October 2025, the model revealed an unusual valuation dislocation in the Japanese company IHI Corporation that a traditional factor screen could not detect. The warning prompted an immediate review of the company’s fundamentals. Within hours, the portfolio manager validated the underlying factors, judged the mispricings to be real and took a position. This transaction was part of a live model portfolio designed to test the entire Human+AI blueprint in real time and measure its impact on alpha costs.

This is what the new alpha factory looks like:

1. The new IP: licensing models, build prompts
   Today, the edge no longer comes from building proprietary AI models, but from how companies use them. Instead of pouring capital into internal development, CIOs should license multiple best-in-class external models and focus on the real differentiator: implementation. This means that you need to know which models to use, where to use them in the investment process and how to combine their results effectively. A company’s real intellectual property now lies in its fast library: the custom workflows that anchor its investment philosophy in common models. This Human+AI approach shifts spend from heavy CapEx to flexible OpEx, often at modest costs of around $500 to $5,000 per model per month, and requires constant monitoring of the AI ​​landscape so that new and better models can be tested and integrated as they emerge.
2. The new process: a four-stage human-AI funnel
   The traditional linear research process must become a multi-phase system in which humans and machines work together from the top down. In a global equity example (equally applicable to fixed income or multi-asset), AI first supports regime-aware allocation decisions, such as steering cash levels based on market signals and adding a critical layer of risk management before individual stock work begins.From there, portfolio management runs through a Human+AI funnel in four phases:
   • Phase 1: pre-screening (e.g. 17,000 → 5,000 shares)
     This first step is purely quantitative and does not require AI. It involves screening the global developed markets universe – approximately 17,000 stocks – against key criteria such as minimum liquidity and market capitalization. The goal is to narrow the field of study to a more manageable universe of approximately 5,000 companies that meet fundamental investment standards.
   • Phase 2: Generate ideas (e.g. 5,000 → 500 shares)
     This is where the power of AI really comes into play. Machine learning and generative AI models are applied to the universe of 5,000 stocks to bring forth new investment ideas tailored to the current market environment. Unlike static screening, this process is adaptive: AI can dynamically shift focus between value and growth styles, identify emerging industry trends, and flag outliers that traditional methods might miss, such as the example of IHI Corporation.
   • Phase 3: Deep analysis (e.g. 500 → 100 stocks)
     Now you can deploy generative AI features as a team of junior analysts. Using the company’s own prompt library, AI reads and analyzes company filings, management tone, technical indicators, sentiment data, competitive positioning and much more from the 500 companies that have made progress from the previous phase. The AI ​​takes care of the mechanical workload, while the human analyst or portfolio manager takes care of the critical interpretation. Together they compile a high-conviction shortlist of approximately 100 candidates. In the IHI Corporation example, the manager used AI deep-dive analysis to validate the company’s balance sheet strength and closing price, going from idea to conviction in a fraction of the usual time.
   • Phase 4: Portfolio construction (e.g. 100 → 70 shares)
     Finally, the portfolio manager takes full control and uses AI as a co-pilot in the construction phase. With the shortlist of 100 stocks in hand, the manager uses AI-driven tools to optimize position size and manage risk exposure at the portfolio level. As described in my previous post, this final step – bringing together human judgment and machine precision – can significantly improve risk-adjusted performance and ensure that alpha generation is both scalable and cost-effective.This funnel compresses portfolio management cycles, strengthens process discipline, and makes alpha generation scalable – whether the team is analyzing 100 or 10,000 stocks – while directly addressing the cost side of active management.

3. The new architecture: a portfolio with four pillars
   The ‘human in the loop’ principle must be more than a slogan; it requires a clear and transparent portfolio architecture. Rather than relying on a single black box, a robust Human+AI portfolio is built from discrete, purpose-driven components.

   A practical design includes four sleeves:

   • AI-driven top ideas: The largest allocation, made up of high-conviction opportunities, highlighted by the AI ​​funnel and validated by the portfolio manager.
   • Human expertise: A special cover for hidden champions and specialist areas where the manager’s unique insight adds value and exploits opportunities that AI may miss
   • Core Stability: Strategic positions in key index heavyweights to anchor liquidity and manage tracking error.
   • AI-driven risk: Diversifying AI-selected positions to reduce overall volatility and improve the portfolio’s Sharpe ratio.

This four-pillar structure is transparent and auditable and shows exactly how human judgment and machine intelligence work together. It keeps people firmly in control – not as the ultimate veto, but as the architect of the entire portfolio.

Maintaining the edge

Investors haven’t lost their appetite to beat the market, just their willingness to pay high fees for weak performance. If active managers can significantly reduce the cost of generating alpha, they can once again provide compelling value versus passive products.

For investment leaders, especially CIOs, the mandate is clear: the future belongs to those who redesign their workflow, not just buy new tools. The first step is to test a processnot a product – a product that enables teams to scale alpha generation efficiently and profitably.

It is crucial that the cost savings do not come at the expense of performance. When human experts are freed from manual data work, they can focus on the real drivers of alpha. The outcome is simple: the same or better alpha at a fraction of the cost.

Initial results from a live model portfolio applying this blueprint suggest it is possible to combine competitive performance with a more efficient cost structure, without requiring more staff or increasing technology budgets.

To maintain that lead, a dynamic system is needed. With new AI models emerging every week, continuous evaluation, testing and integration of the best tools must become standard operating disciplines for any CIO focused on long-term competitiveness.

The companies that succeed will be those that master the integration of human judgment and AI at scale. They will be the ones who can crack the cost of alpha and secure a sustainable advantage in the next era of active management.