The Productivity Paradox, Finally Cracked?

Here is the uncomfortable backdrop. The asset management industry has been spending aggressively on technology for years, with tech budgets growing at approximately 8.9% annually in North America and Europe since 2019. And it has largely gotten nothing for it. McKinsey’s research across firms representing 70% of global AUM found virtually no statistical relationship between higher technology spending and improved productivity. Margins declined three percentage points in North America and five in Europe over the same period.

The reason, as any honest CTO in the industry will tell you, is that roughly 60 to 80 percent of technology budgets go to maintaining legacy systems, rather than building the future ones. McKinsey notes that firms have been paying this "complexity tax" for decades.

Agentic AI may finally force that math to change. McKinsey estimates that a mid-sized asset manager could unlock 25 to 40 percent of its total cost base in efficiencies--not from automating individual tasks, but from end-to-end workflow redesign. The critical qualifier is that nearly 90 percent of companies have now invested in AI, but fewer than 40 percent report measurable gains, because most are applying AI to discrete tasks rather than rethinking how work actually gets done.

That gap between deploying AI at the margins and reimagining workflows from the ground up is where the asset management industry is innovating.

Three Firms, Three Approaches

BlackRock, JPMorgan, and Vanguard are each taking a different stance on where AI’s value lands, with all three are strengthening their models on their own terms.

BlackRock is playing the longest game. Its Aladdin platform, already the operating system for trillions in institutional assets, is being AI-layered from the inside out. BlackRock’s Technology Services division now generates roughly 8 percent of total firm revenue. AI isn’t a cost center at BlackRock; it’s a product they sell. That structural advantage of a tech revenue stream that funds R&D without cannibalizing advisory economics is something few competitors can replicate.

JPMorgan is betting on advisor speed. Its internal "Coach AI" tool helps private client advisors surface research and market context in real time. During April 2025’s market volatility, when clients were flooding advisors with questions, JPMorgan positioned Coach AI as a speed layer--compressing preparation time so advisors could respond with precision during exactly the moments when clients are most anxious and most likely to make bad decisions. It's not replacing the advisor. It's making the advisor faster when it matters most.

Vanguard is doing what Vanguard always does--using technology to extend its mission at scale. Its AI-powered behavioral nudge tools combine machine learning with behavioral science to push self-directed investors toward better decisions. In the first half of 2025, more than 200,000 investors used Vanguard’s MinTax tax-loss harvesting tool in trades totaling over $38 billion. For a firm whose entire value proposition is giving ordinary investors access to institutional-grade capability at low cost, AI isn’t a threat to the business model, but the purest expression of it.

Three firms. Three definitions of what AI is for. Three different control points defended and strengthened.

The Trading Bot Reality Check

Which brings us to the honest counterargument every sophisticated investor has top of mind.

In May 2026, Bloomberg reported on Alpha Arena--a competition that pitted eight major AI models, including Claude, ChatGPT, Gemini, and Grok, against each other in live US tech stock trading over two weeks. Each model was given $10,000. The portfolio as a whole lost approximately a third of its capital. Across 32 sets of results, a model finished in profit just six times.

One model placed 158 trades under a given prompt; another, given identical instructions, placed 1,418. "LLMs can’t really make money by themselves," said Jay Azhang, founder of Nof1, which ran the contest. "You need a very sophisticated harness and scaffolding and data platform in order to even give them a chance."

The models also showed distinct personalities. In one contest, Claude went mostly long, Gemini was short, and Qwen levered up. Another observer noted watching them felt like overseeing a group of analysts, each with a completely different market instinct.

To be fair, these public experiments are too short and too noisy to support firm conclusions, and they lack the sophisticated quantitative techniques and execution infrastructure found inside the best investment firms. Even so, the underlying capabilities, when housed within expert infrastructure at a high-end hedge fund, for example, cannot be trivialized. The pieces are there. The scaffolding isn’t--yet.

The Control Point Question

The firms that will win the AI era in asset management are not the ones that automate the most. They are the ones that own what McKinsey aptly calls the “control points”--permissioned data, compliance-grade auditability, client trust, and the execution infrastructure that turns AI-generated insight into a completed transaction.

Griffin was right to be sobered. Work that required PhD-level talent and months of effort is being compressed into hours. That is genuinely disruptive to the industry’s cost structure and talent model. But it doesn’t mean asset management has lost its value—it means the value-add equation is changing. Planning is getting cheaper than supervision. And research is getting cheaper than judgment.

The value in asset management isn’t disappearing. It’s migrating—toward the firms that own the relationships, the data, and the judgment that no AI agent can replicate on its own. The race isn’t to automate the most. It’s to scale to become indispensable first.