Despite rapid advances in AI models, agentic systems and automation tools, many AI initiatives in investment management struggle to move beyond the demo. One reason is simple – most investment organizations do not maintain a comprehensive catalogue of the entities, logic and workflows that define how they operate. This leaves AI delivering inconsistent results and too open to probabilistic workings. For institutional asset managers, that’s a key line in the sand that leads to thinking that AI cannot be relied upon.

AI needs structure, guardrails, governance… and a catalogue of everything

AI performs best when it can operate within a clearly defined representation of a business. In technology this is often called an ontology, a concept advanced in other industries. Think of this as a structured, governed and transparent catalogue of everything that an investment manager knows and does.

At Jacobi, this is something we think long and hard about. Our heritage in investment management has been helping institutional managers to industrialize their unique investment workflows. Doing that in a scalable way across many clients has required focus on how best to break down workflows into modular elements. These extend beyond data, but also include user workflow steps, system actions and analytic calculations, spanning from simple through to computationally complex.

What does a ‘catalogue of everything’ look like for investment managers?

The starting point is a comprehensive investment data model – but not one predicated on well-established constructs such as IBOR and security masters. These are still essential, but firms must also look much wider. For example, clients, model portfolios, transactions, cash flows, investment policy statements and management agreements, products, research signals – each of these are key entities in need of cataloging, structure and governance.

The next catalogue layer is your investment logic and calculation methods. Analytics are the lifeblood of investment managers, with every firm reliant on a sea of measures such as performance, risk and attribution. Yet a seemingly straightforward metric like volatility can be calculated in many different ways – each may be technically correct, just a different approach. So a robust analytics library is a must, yet is still disjointed in many investment teams.

The same cataloguing discipline must also be applied to other aspects of the investment process, such as inputs into portfolio construction calculations like constraints, objectives, research inputs, as well the construction models themselves (optimization etc).

Crucially, cataloguing also needs to be hierarchical. Extending on the analytics example, the catalogue cannot stop at individual calculations. It must account for the relationships between them, how they are combined and visualized, and account for complex multi-dimensional constructs such as simulations, analytic decompositions and scenario analysis.

From this point you can then start to catalogue broader level workflow inputs and outputs. These are elements repeatedly combined into workflows. Examples may include analytic dashboards and reporting templates that are consistent inputs and outputs for investment committee meetings, client portfolio reviews, analyst research meetings and risk oversight processes.

The final layer are the workflows themselves, which may be fully-agentic or human-driven applications and tools. While agentic workflows receive a lot of attention, regulated investment firms need high levels of consistency, explainability and auditability. That means the future is unlikely to be entirely agent-driven. Instead, firms will combine deterministic workflows, specialist applications, human oversight and AI-assistance in carefully controlled ways.

Jacobi’s ‘Catalogue of Everything’ for Investment Teams

Don’t mistake an organization challenge for a technology barrier

Perhaps the most interesting observation is that none of this is fundamentally new. Investment managers already have the capabilities in place, but meticulous cataloging, organization and planning is. AI has simply raised the bar. Processes that were previously tolerated as fragmented, undocumented or dependent on institutional memory are now barriers to innovation.

And as we look to the future, the catalogue of everything must evolve to include new concepts – such as AI models, AI agents, MCP servers and all the skills and rules that provide context to agentic processes. Entirely new product and service lines will also need to be accommodated.

The firms whose AI trials succeed that succeed with AI will not necessarily be those with the most advanced models. They will be the ones that have a clear catalogue of their investment management business: their data, their investment logic, their content and their workflows. Once that foundation exists, AI becomes far more than an experiment, it becomes operational.