AI in asset management scales when firms build “AI factories” — shared infrastructure, reusable components, and governance that turns AI into a repeatable capability.
Most asset managers are still treating AI like a set of isolated experiments: one pilot for research summarisation, another for meeting notes, a third for client reporting. The problem is that one-off use cases don’t compound. They create local wins, but they don’t create an enterprise capability.
The most important shift happening now is structural: the move from “use-case delivery” to AI factories — internal infrastructure that makes building, deploying, and governing AI fast and repeatable. For leaders focused on AI in asset management, this matters more than the latest model upgrade.
MIT Sloan Management Review describes how “all-in” adopters are creating AI factories: combinations of platforms, data, methods, and reusable algorithms that make it easy to ship AI systems quickly.
This isn’t about building massive GPU clusters. Cloud and model vendors handle that. The differentiator is what sits inside the enterprise:
shared data access and permissioning standardised tooling and deployment patterns reusable components (retrieval, evaluation, monitoring) repeatable delivery methods
Banks have done versions of this for years for analytical AI (fraud, credit decisioning). Now the same pattern is expanding across industries and across AI types — analytical, generative, and agentic. Intuit’s “GenOS” is a clear signal of where this is going: an internal operating system for building AI at speed, not a collection of disconnected pilots.
Our View — “AI factories” are the missing layer for agentic AI in asset management
We believe most AI roadmaps in asset management fail for one reason: firms try to scale outcomes without scaling the underlying production system.
1) Asset managers don’t have an AI delivery engine yet
In a typical AM firm, every team still has to re-solve the same problems every time they start an AI project:
Which tools do we use? Where is the data and who can access it? How do we evaluate output quality? How do we prove compliance and control? How do we monitor drift and production performance?
That creates two predictable outcomes: AI becomes slow, and AI becomes expensive. Worse, teams lose confidence because delivery is inconsistent.
An AI factory changes the economics. It replaces bespoke work with a production line.
2) AI factories make agentic AI practical — not just impressive demos
Agentic AI for asset managers is not “one agent that does everything.” In production, it’s multiple agents performing narrow tasks inside controlled workflows: extracting intent, retrieving policy, checking constraints, producing evidence, logging actions, routing exceptions.
Without factory infrastructure, agentic systems break immediately on the enterprise realities that matter in asset management: auditability, explainability, segregation of duties, model risk, regulatory proof.
With factory infrastructure, agents become safe and deployable because the surrounding controls are standard and reusable.
3) The real win is speed compounding across workflows
A factory unlocks the compounding effect most firms miss.
If you speed up only one step (for example, research summarisation), you get a local improvement. But if you speed up connected steps (summarisation → draft recommendation → compliance checks → reporting), the whole cycle accelerates.
This is the point: the total benefit of AI in asset management is greater than the sum of individual automations when workflows are interconnected. AI factories are the mechanism that makes those connections cheap and repeatable.
Implications — What this means for CEOs, CIOs, CTOs, and Chief AI Officers
If you want AI agents for asset management to move from “interesting” to “strategic advantage,” treat the factory as the product.
Build the factory first, then scale use cases
Prioritise these building blocks:
Data layer: approved datasets, lineage, permissions, feature stores where relevant Model access layer: standard interfaces for LLMs, embedding models, classifiers, routing Reusable components: retrieval pipelines, evaluation harness, prompt patterns, tool calling Governance layer: logging, evidence capture, policy-as-code, model risk documentation Deployment layer: CI/CD, runtime monitoring, fallback logic, cost controls
This directly supports AI governance for asset managers in Europe, where audit trails and control frameworks are non-negotiable.
Measure success like a factory, not a pilot
Replace vanity metrics with operational ones:
time from idea → production percentage of use cases built from reusable components evaluation coverage (accuracy, completeness, hallucination rate, policy adherence) incident rate + time to remediation cost per workflow run and cost per team adoption
Don’t let each team choose its own stack
This is the silent killer of scale. If every desk invents its own tooling and patterns, you end up with fragmented prototypes and no enterprise learning curve.
Factories create standardisation without blocking innovation: teams still build use cases, but they build them on shared rails.
Design for multiple forms of AI from day one
Your factory must support:
Analytical AI (forecasting, risk, anomaly detection) Generative AI (summaries, drafting, synthesis, explanation) Agentic AI (workflow automation, monitoring, escalation handling)
If the factory only supports one category, you’ll rebuild it later. If it supports all three, you accelerate everything.
Conclusion — AI factories are how AI in asset management becomes a durable edge
The next phase of AI in asset management will not be decided by who demos the smartest chatbot. It will be decided by who builds the internal capability to deliver AI repeatedly, safely, and cheaply.
AI factories are the operating system of modern AI adoption. They reduce duplicated effort, compress delivery cycles, and make agentic AI for asset managers real in production — with governance, evidence, and controls baked in.
Firms that build the factory early will ship faster, learn faster, and compound advantages across the investment and operational lifecycle.
