How Asset Managers Are Using AI in 2026

A portfolio manager at a large firm can now open BlackRock's Aladdin, ask it in plain English to stress-test a position and get the answer before her coffee cools. Then she decides whether to buy or sell herself, the way she did in 2015. That small gap, between what the machine hands her and what it is allowed to decide, is the whole story.

So how are asset managers using AI? Almost everywhere, and at two very different speeds. Fast and freely in research, operations, distribution and compliance. Slowly, and on a short human leash, inside the investment decision itself. Adoption is settled. Mercer's 2026 AI in Asset Management Survey found that 55 percent of managers already run AI in at least one part of their investment process, and 91 percent plan to use more of it. The open question is where the value is real. Here is where AI is genuinely working, where it is mostly a marketing line, and the one place the industry will not let it go.

Accelerating Investment Research and Synthesis

The first place AI earned its keep inside the investment engine is research, where it reads faster than any analyst and never gets tired.

The core job is turning an ocean of unstructured material into something a person can act on. Models summarize earnings calls and regulatory filings, compare a portfolio's holdings against its investment guidelines, digest alternative data and rank the signals worth a human's attention. The result is a shorter research cycle. Take the junior analyst who used to lose a day and a half to a two-hundred-page biotech filing dropped after the close. A model now surfaces the buried risk factors in minutes, and he spends the afternoon deciding what they mean instead of hunting for them.

The marquee firms have built this into named tools. Morgan Stanley extended its AI from the wealth arm into its institutional securities division through AskResearchGPT, a GPT-4 assistant that lets staff surface and summarize insights from a research library of more than 70,000 reports a year. JPMorgan has been building tools to analyze and select securities since it filed for IndexGPT in 2023.

One rule runs under all of it, and it will run under every engine section that follows. AI drafts the research; the analyst owns the conviction. A model can find the number and even suggest what it means, but the moment a portfolio manager forwards an unverified figure, he owns it as surely as if he had typed it himself. The speed is real. It does not transfer the responsibility.

Building and Stress Testing Portfolios

One level deeper sits the work closest to the fiduciary core, using AI to help build portfolios and pressure-test them, and it is where the human leash is shortest. This is the territory of portfolio construction, optimization, risk analytics and scenario analysis, and no platform embodies it more than BlackRock's Aladdin. Aladdin is the risk and portfolio engine behind BlackRock and a long roster of other institutions, and its whole pitch is speed from insight to action. Its head of product management describes Aladdin as using AI to help investors move from insight to action faster by cutting manual work and widening access to information. Its newer generative layer lets a portfolio manager query positions and run a risk scenario by typing a question in plain English, rather than waiting on a quant to build the report.

Powerful as that is, the discipline around it is the point. Vanguard runs proprietary AI models in its quantitative strategies and still puts human experts between the model's output and the portfolio. Trading and portfolio optimization top the surveys of high-return uses, and yet, as the returns data already shows, the edge is hard to hold. So the clear-eyed firms treat AI here as a tool that widens what a manager can see, not one that decides what she does about it. Models propose. Humans dispose. That is a choice, not a technical limit, and it is the choice a fiduciary is paid to make.

Sharpening Risk and Compliance Surveillance

AI is a natural fit for surveillance, the unglamorous work of watching everything at once and flagging the few things that matter.

Trade surveillance, communications and email review, behavioral analytics: these are pattern problems at a scale no team can cover evenly, and models are good at triaging them and escalating the outliers to a person. A model reads the ten-thousandth message with the same scrutiny it gave the first, and it does not tire late on a Friday. McKinsey notes that automated compliance monitoring is one of the clearer ways AI is lowering operational risk, alongside codifying institutional knowledge so it survives a key departure.

Compliance is also where AI meets its own rules, and SEC compliance does not ease up because a model did the work. The Investment Advisers Act bars false or misleading claims about a firm's capabilities, its AI included, and the SEC's marketing rule requires a fair, substantiated presentation of performance that reaches AI-generated materials and any forecast a model helps produce. None of this is a reason to keep AI out of the compliance function. It is a reason to keep a human on the escalation and hold AI's output to the same standard as everything else the firm produces, the standard that keeps a manager clear of breaching fiduciary duty. The payoff is coverage and consistency a manual review could never match, with a person still making the judgment call at the end.

Scaling Distribution and Client Personalization

Here the piece crosses from the investment engine to the machine that gathers the assets, and the economics change fast. Distribution is where AI stops being a leashed assistant and starts acting like a growth lever. On this side of the firm, models extend a wholesaler's coverage to more advisors than a person could ever call, tailor commentary to what a given client holds, prioritize the accounts worth a human's time and prompt the next best action for a relationship manager walking into a meeting. Modeling the upside, BCG found that AI works as a force multiplier in distribution and that an AI-first manager could add incremental annual inflows of roughly 0.5 to 1 percent of AUM within three to five years through distribution alone. For a firm running tens of billions, that is not a rounding error. It is a growth line. McKinsey points to sharper client targeting as one of the earliest uses that generate real revenue rather than just savings.

There is a catch, and it sets up everything that follows. Personalization at scale is only as good as the firm's record of what each client said and needs, and that record is thinner than most managers admit. The moment a wholesaler learns that an advisor has soured on a fund, or an institutional client hints that a mandate is under review, the signal that should feed the personalization engine is spoken aloud in a meeting and then, most of the time, lost. You cannot personalize around a conversation nobody wrote down.

Turning Client Meetings Into Structured Records

The machine that gathers assets runs on meetings, and most of what happens inside them is gone by Friday.

Think about where the client-facing work of an asset manager lives. A wholesaler sits with an advisor. An institutional relationship manager sits with a pension board or an endowment's investment committee. A client-service team fields the call after a rough quarter. Every one of those conversations carries the signals that decide whether the assets stay and grow: a worry about a strategy, a hint of redemption risk, an opening to bring over a held-away assets, the real objection under a polite no. And every one of those signals tends to die in a notebook nobody reopens.

This is the work a growing category of AI assistants was built to carry, the ones designed for client-facing financial professionals rather than the trading desk, the same shift that has advisors handing AI the work that happens after the client leaves the room. Jump, AI for asset managers and the client-facing side of the business, joins the meeting, writes the note, files the details into the CRM and drafts the follow-up, then builds the next agenda, a review or a discovery meeting agenda for a new prospect, from what the last one surfaced. The relationship manager who used to rebuild six meetings from memory on a Friday night walks out of each one with the record already done. Nothing important waits on someone remembering to type it up.

What matters is what the software protects. A relationship manager can carry more accounts without dropping the thread on any of them, which is as true for attracting high net worth clients as it is for servicing institutional ones. The offhand comment that signals a mandate review in March becomes a scheduled conversation in April instead of a regret in June. And because the interaction is captured as it happens, the record stays current for the books-and-records and marketing-rule obligations from a few sections back, rather than getting rebuilt under deadline before a review. The distribution engine you just read about is only as smart as this, the plain discipline of turning what a client said into something the firm can find later.

Codifying Knowledge and Automating Operations

The least visible use may be the most durable: pointing AI at the firm's own institutional memory so it stops walking out the door every time someone resigns. Most of the value large firms report from AI so far is this kind of internal productivity, not a flashy customer-facing product. The dominant pattern is an internal assistant indexed against the firm's own research, policies and history, a retrieval setup that answers from what the firm already knows rather than from the open web. The scale is striking. Goldman Sachs began rolling its GS AI Assistant out firm-wide in 2025, starting with around 10,000 employees and aiming at every knowledge worker, and by early that year more than 200,000 JPMorgan employees had access to in-house generative AI tools of their own. These assistants draft, summarize, translate code and answer questions against a governed body of knowledge, and they quietly automate the back office too, the reconciliations and reporting that used to eat junior hours.

McKinsey makes the sharper point that codifying institutional knowledge this way guards a firm against the losses that come when a senior person leaves and takes the context with them. Some of that context is technical. A lot of it, on the client-facing side, is relational, the history of what a given client has worried about and asked for across years of meetings, which is precisely the memory an assistant like Jump builds as a byproduct of capturing the conversations. Run at the level of the whole firm, this is the machine getting cheaper and more durable at once, and it is where the honest return on AI has been easiest to find.

The Conversation is the Asset

Ask how asset managers are using AI and the honest answer is a map with two speeds. Quick and confident where the work is safe, slow and supervised where the stakes are a client's money. The firms getting real value are not the ones chasing an AI that beats the market. They are the ones using it to run the research engine faster, the compliance function wider and the distribution machine cheaper, while keeping the investment decision exactly where it has always been, in human hands.

The hardest part of that, on the fast side, is memory. The signals that drive distribution and retention are scattered across hundreds of client meetings no one can hold in their head, and they fade a little more every week they sit uncaptured.

That is the gap Jump closes for the client-facing side of the business. It sits in the meeting, writes the note, files the details into the CRM and drafts the follow-up, so the thing a client said once, in passing, becomes something a wholesaler or a relationship manager can find months later, across the whole book, on the day it finally matters. Jump reports that its users save around 10 hours a week once the notes, follow-ups and CRM updates run on their own, and that firms using it see a 42 percent drop in outstanding client-service tasks; more than 35,000 advisors and their teams already run on it, most reaching full adoption within days. Point those reclaimed hours back at the relationships that gather and keep assets, and the client work stops slipping through the cracks. See how Jump works in a demo.