Following a recent roundtable at TSAM New York, MAIA’s Ashish Shrestha has been reflecting on what it takes for asset managers to move AI from experimentation to operational reality and into trusted, scalable workflows.
AI is reshaping the asset management industry, and firms are exploring various alpha generating use cases, running pilots and incorporating AI into workflows. But there is a significant difference between demonstrating what AI ‘can’ do and embedding it into the day-to-day operating model.
For Ashish, the question is becoming less about whether AI should be part of the strategy and more about whether an organisation is operationally setup to succeed embedding AI at scale.
Moving beyond the pilot
Firms are at different stages of operational maturity, from observing the developments and experimenting with controlled use cases through to established AI strategy and integrated workflows in production.
A successful pilot can prove that a use case has value. It does not necessarily prove that the organisation has the data, architecture, controls and processes required to deploy it more widely.
Scaling AI is not simply a technology project. It requires firms to consider how AI fits into existing workflows and how its use can be governed with a ‘human in the loop’ oversight.
Firms also face practical decisions about how AI capabilities should be delivered. Larger organisations are leading the development of native, in-house capabilities, driven as much by regulatory demands as by alpha generation. Boutique firms are more likely to adopt AI embedded in partner solutions, while sophisticated hedge funds will expect those solutions to operate securely within their own AI environments. Interoperability across the firm’s wider technology and governance environment is therefore essential.
Start with trusted data
As one panellist at TSAM noted, unifying and harmonising data architecture may sound like a conversation from five years ago, but today it matters more than ever: AI is only as useful as the information it can reliably access. For asset managers, that makes the underlying data environment particularly important. If information is fragmented across systems, inconsistent, batch based or difficult to access, introducing AI does not remove those problems.
The foundations therefore need to come first: data quality, lineage and appropriate access, supported by architecture that allows information to move reliably across the operating model.
Interestingly, this is also where AI has the potential to change how people interact with investment technology. Rather than relying on another layer of custom interfaces or manual reporting, AI can increasingly act as an access layer over trusted data, helping users query information and generate insights more naturally. Trust in output ultimately comes from trust in the underlying infrastructure to deliver the right building blocks.
Governance cannot be added afterwards
As AI moves closer to production workflows, governance needs to be central to the architecture rather than an afterthought. Ownership, controls, and auditability all need to be considered alongside privacy, compliance and model risk.
The same applies to human oversight. The objective should be to create a ‘human in the loop’ framework to streamline processes. It is extremely important to understand where automation can create efficiency and where human judgement remains essential. Building that trust is paramount in an industry where decisions, data and processes operate within significant regulatory and fiduciary responsibilities.
From AI readiness to operational readiness
The next stage is creating the conditions that allow it to be deployed safely, consistently and at scale. That means trusted data. Connected architecture. Clear governance. Appropriate controls. Human in the loop. And, ultimately, confidence in the process.
The organisations that make progress will not necessarily be those that experiment with the greatest number of AI tools. They will be those that build the foundations needed to turn successful prototypes into trusted AI embedded workflows.
