AI in Asset Management: The Gap Between Having It and Operationalising It

AI in asset management is the use of machine learning and, increasingly, agentic systems to support investment research, portfolio construction, risk management and operations across an investment firm. That is the working definition, and it is deliberately narrow: it is about applying these techniques to the business of investing money, not about managing physical equipment.

The distinction matters because the same three words describe two unrelated fields. In investment asset management, "assets" are securities, funds, private holdings and the portfolios built from them. The job is allocation, selection and risk. AI here reads filings, extracts signals from unstructured data, stress-tests portfolios and automates reporting.

Most of what searchers want sits in the first field, even though the second one ranks for the same phrase. The rest of this page stays firmly in investment asset management: how an investment firm uses AI to research, build, monitor and operate portfolios, and what separates a firm that runs that capability from one that merely advertises it.

Investment AM vs enterprise AM: the same words, two different fields

Enterprise asset management is a different discipline entirely. There, "assets" are physical: pumps, turbines, vehicles, plant. AI is used for predictive maintenance, sensor analysis and asset-lifecycle planning on the factory floor or across an infrastructure estate. It is a legitimate and valuable use of the technology, but it has nothing to do with managing capital.

Confusing the two leads readers somewhere unhelpful, into industrial maintenance content when they came to understand investing. For clarity: when this page says asset management, it means the management of investment portfolios. The machine-learning techniques may rhyme across the two fields, but the objectives, the data and the people involved do not.

AI is not one capability inside an investment firm; it is several, applied to different parts of the process. The cleanest way to see the field is to map it to the work itself. Five areas account for most of what is in production today, and a private-markets line runs through each of them.

• Investment Research. This is where AI earns its keep first. Models read and summarise filings, transcripts, news and other unstructured data at a scale no research desk could match, extracting signals and surfacing the document that matters from a pile of thousands. In manager research and due diligence, the same techniques compress weeks of reading into a working brief. The constraint is not capability but judgement: the output is a faster starting point, not a decision.
• Portfolio Optimisation. AI supports construction, rebalancing and scenario modelling, testing how a portfolio behaves across many possible futures rather than a handful of hand-built cases. Used well, it widens the range of scenarios a team can consider and sharpens the trade-offs; it does not replace the mandate, the constraints or the allocator's call on risk.
• Risk Management. Here the work is monitoring and measurement: supporting value-at-risk estimates, flagging concentration and exposure drift, and watching positions continuously rather than at month-end. Model risk itself becomes a live concern, because a risk system built on a model is only as trustworthy as that model and the oversight around it.
• Operational Efficiency. The least glamorous use case is often the most immediate. Reporting, reconciliation and other back-office workflows are heavy with repeatable, rules-based tasks, and that is precisely where automation removes cost and error. Much of the early, measurable productivity gain in the industry comes from here, not from the research desk.
• Personalisation. On the client side, AI tailors reporting and communication, shaping how a portfolio is explained to the person who owns it. For an institutional allocator this matters less than research or risk, but it is a real and growing part of how managers differentiate their service.

Across all five, the same pattern holds. AI compresses the work and widens what a team can consider; people still exercise the decision. Technology produces the package; judgement signs it off.

The private-markets use cases the public-markets literature skips

The published explainers on this topic are almost entirely public-markets framed, written for and about large managers of listed securities. Private markets are where the more interesting AI work is now happening, and where the literature is quietest.

Deal sourcing is the first. In private markets, the edge is proprietary: seeing opportunities other allocators do not. AI helps surface and screen off-market situations from a far wider information field than a deal team could canvass by hand. Diligence and underwriting are the second, and the most consequential. A private transaction carries weeks of document review, where a public trade carries a screen and a price; that is exactly the kind of work agentic systems are built to compress, taking a diligence cycle measured in weeks down toward hours. Illiquid-asset monitoring is the third: private holdings do not reprice on a public tape, so the discipline is to read the signals that do exist, in operating data, covenants and counterparties, continuously rather than at the next valuation. This is native territory for a private markets investment group. IMS Group runs its diligence and underwriting on exactly this premise, a subject taken up in detail in where AI actually changes underwriting.

The shape of adoption has changed in the last two years. The first wave was conversational: chatbots and copilots that answered questions and drafted text, useful but bounded, a demo of what the technology could say. The second wave is agentic: systems that do not just answer but act, executing multi-step tasks across tools and data with limited supervision. The shift toward agentic AI as an operating system rather than a chat window is the single most important development in how investment firms use AI, and it is the development the industry is now racing to absorb.

The headline figures point in the same direction, though they should be read as a range rather than a single number. McKinsey has estimated that AI and agentic adoption could improve the cost base of asset management by roughly 25% to 40% over time (McKinsey, 2025); the figure is an estimate of potential, not a measured result, and it engages the scale of the opportunity without prescribing a methodology. For a wider frame, J.P. Morgan Asset Management estimates that generative AI could lift developed-market labour productivity by roughly 1.4% to 2.7% per year over a decade (J.P. Morgan Asset Management, 2025); this is a whole-economy estimate rather than a figure measured inside asset-management workflows, and it is included to set the macroeconomic backdrop, not as a directly comparable asset-management result. Grant Thornton's 2025 survey of around 500 senior executives globally (Grant Thornton, 2025) provides the breadth: at the level of firm-wide adoption, AI in investment management is broad in intent, uneven in execution, and concentrated for now in operations and research rather than in core investment decisions.

The numbers support two readings at once. Direction is clear: adoption is accelerating, and the agentic shift is real, not hype. Magnitude is contested: McKinsey's 25% to 40% is an estimate of potential, and the distance between that potential and what most firms have so far made AI do is precisely the gap that matters. That gap is the subject of the next section.

This is the distinction that matters most, and it is rarely named. Almost every firm now has AI ambition; far fewer have AI capability that runs. The difference between appetite and operating reality is the defining feature of where the industry stands today.

The reason is structural. A model is not an operating system. A capable model, bought or accessed off the shelf, is the easy part; making it useful inside an investment firm requires four things the model itself does not provide. It needs data: clean, connected, proprietary information for the model to work on, which most firms do not have in usable shape. It needs workflow: the model has to sit inside how the firm actually researches, decides and reports, not alongside it as a novelty. It needs oversight: human review, model-risk governance and the authority to overrule the machine, built in rather than bolted on. And it needs integration: the connective engineering that turns a clever demo into a system the firm relies on every day. Assemble those four and AI operates; leave any of them out and the firm has a capability it owns but cannot run.

This is why "we use AI" and "we operate AI" are not the same claim, and why the distance between them tends to be measured in years of engineering rather than a procurement decision. Industry research has repeatedly found that only a small minority of organisations have scaled an AI agent into genuine production in any function; the ambition is near-universal, the operating capability is not. Building it is a question of appetite versus capability, examined directly in appetite vs capability — the AI skills gap, and of why a firm might choose to engineer the capability in-house at all, the subject of why an investment group runs its own AI lab. The point to hold at the category level is simpler: production beats demonstration, and the gap between the two is where credibility is won or lost.

Why "we use AI" and "we operate AI" are different claims

A firm that uses AI has access to a tool. A firm that operates AI has rebuilt part of how it works around that tool, with the data, workflow, oversight and integration to match. The first is a statement about software the firm can reach; the second is a statement about how the firm runs. Most marketing language blurs the two deliberately, which is exactly why an allocator needs a way to tell them apart.

"AI washing" is the practice of marketing AI capability a firm does not genuinely have, or materially overstating what it does. It is the predictable consequence of a moment when claiming AI is rewarded and verifying it is hard, and it is a live enough risk that the term now appears in the industry's own list of challenges and in the attention of financial regulators. The useful response is not cynicism but a test.

An allocator can apply four practical checks before taking an AI claim at face value. The first is production evidence: not "we are exploring AI" but what is running today, in which workflow, with what observable effect. A firm that operates AI can describe the system; a firm that washes can only describe the ambition. The second is human oversight: a credible operator can explain exactly where a person reviews, overrules and signs off, because unsupervised models in an investment process are a risk to be governed, not a feature to be advertised. The third is governance: model-risk controls, validation and documentation, the unglamorous scaffolding that separates a system you can rely on from one you merely hope works. The fourth is measurable workflow change: a real capability shows up as a process that demonstrably runs differently, faster diligence, continuous monitoring, automated reporting, not as a logo on a slide.

This test sits comfortably alongside the governance stance taken by bodies such as the CFA Institute, whose work on AI in asset management stresses transparency, oversight and the management of model risk as the conditions for using these tools responsibly (CFA Institute, 2025). The credibility floor is the same one a careful allocator would set anyway: human oversight and model-risk governance are not optional extras on top of an AI capability; they are the evidence that the capability is real. No fee comparison and no named competitor is required to apply the test. The questions do the work.

For an allocator in private markets, the operationalisation gap is not an abstraction; it is a selection criterion. Most of the published guidance on AI in asset management is written for public-markets managers, so the questions that matter to private capital fall to the allocator to ask directly.

The places AI changes the private-markets process are specific. Deal sourcing widens the aperture on off-market opportunities. Diligence and underwriting compress weeks of document review toward hours, which changes not just cost but the number of opportunities a team can credibly assess. And monitoring turns illiquid holdings from a quarterly valuation exercise into a continuous read of operating signals. At each of these points, a firm that operates AI does the work differently from one that does not, and an allocator evaluating managers can ask to see exactly that difference. The same dynamic runs through adjacent private-markets strategies, including private credit, where underwriting quality and speed are the whole competitive question. It also connects to the infrastructure layer, where tokenised assets need an authentication layer and AI sits behind the verification of what an on-chain instrument represents. All of this is one part of the firm's wider reading of the next generation of private capital.

Over the next cycle, the dividing line in asset management is unlikely to be whether a firm uses AI; that question is already settled. It will be whether a firm operationalises it, turning a capability it owns into a system it runs, with the data, workflow, oversight and integration that distinction requires. Adoption will keep accelerating and the agentic shift will keep maturing, though the realised gains should still be read as a range rather than a single promised number.

The firms that close the gap between having AI and operating it will compound that advantage across cycles, particularly in private markets, where the published field has barely begun to look. To see how an operator approaches this in practice, explore how IMS Labs builds and runs the firm's AI capability in production (imslabs.ai), and read across the cluster, from the agentic shift to the skills gap, for the detail beneath this category view.

This article is provided by IMS Group for information purposes only and does not constitute investment advice, an offer, or a solicitation. Figures are sourced and dated where cited; forecasts are projections and stated as ranges.

This article is provided by IMS Group for information purposes only. It does not constitute investment advice, an offer, or a solicitation. Figures are point-in-time and projections are estimates.