Much of the writing on AI in asset management is about research and strategy. The dependable gains are in operations: settlement, onboarding, notices, documents, RFPs and investor queries, with controls that hold up.
AI in asset management pays back most reliably in the middle and back office, not in picking investments. Settlement exceptions, investor onboarding, capital call and distribution notices, document classification, RFPs, due diligence questionnaires and investor queries are all high-volume, rule-bound work where AI can prepare and a person can approve. Much of what is published on the subject is about research, distribution and strategy. This is the operations view, and it applies just as much to private equity.
Where AI in asset management earns its keep
- Trade settlement. Matching trades against confirmations, investigating breaks, gathering the evidence for each exception and preparing the fix. The agent does the legwork; a person approves anything that moves cash or securities.
- Investor onboarding. Reading subscription documents and identity packs, extracting the details, checking them against your requirements and listing what is missing for the KYC and anti-money laundering review. The decision to accept an investor stays with your team.
- Capital call and distribution notices. AI drafts the notice text and covering email for each investor. The amounts come from your fund accounting system and are checked in code, never typed or calculated by a language model.
- Document classification. Statements, notices, valuations, legal documents and correspondence land in shared inboxes all day. Classifying and routing them, and pulling out the key fields, is often the first win.
- RFPs and due diligence questionnaires. Drafting answers from your approved library of past responses and policies, with each answer pointing to its source. A subject-matter expert approves every answer before it leaves. RFP automation was part of my work at Alter Domus.
- Investor queries. Drafting replies to routine questions about statements, payments and documents from your own records, for a person to check and send.
What multi-agent trade settlement taught me
At Alter Domus, where I was Head of AI & Automation, one of the systems we ran was multi-agent trade settlement doing the work of roughly 25 people, with a person approving every irreversible step and a full audit trail.
The two conditions at the end of that sentence are the point. In any settlement workflow, most steps can be undone: gathering data, matching records, drafting an instruction. A few cannot: releasing a payment, instructing a counterparty, closing a break in the books. Agents can do the first kind at speed. The second kind waits for a named person, who sees what the agent found and why it proposes the action. The audit trail records both, so you can answer an auditor or a regulator months later. The rest of that track record is on the AI for financial services page.
The controls that make it shippable
In a regulated firm, controls are what get a tool into production. These are the ones I agree before any data source is connected on an agents and automation project:
- A person approves anything irreversible. Payments, instructions to third parties, changes to investor records and anything sent outside the firm.
- Numbers are checked in code. Amounts, allocations and dates come from your systems and are verified by fixed rules. A language model can draft around a number; it should never be the source of one.
- A full audit trail. Every input, every step the agent took, every approval and every override, kept with the record.
- Least privilege. The agent can read more than it can write. Read access to a register is low risk; permission to change it is not.
- Private tenancy. Models run in your own cloud environment or under business terms that keep your data out of training.
- An evaluation set. Real past cases, including the awkward ones, run against every change to the prompts or the model before it goes live.
- A way to stop it. A simple switch back to the manual process, and people who still know how to run it.
If you are FCA-regulated, the FCA’s AI Update points to rules you already follow: governance, systems and controls under SYSC, outsourcing, and, for firms they apply to, the operational resilience rules where AI supports an important business service. For retail business the Consumer Duty applies too. This is practical guidance, not legal advice.
AI in private equity: the same work, more paper
Private equity operations run on documents: limited partnership agreements, side letters, subscription agreements, capital calls, distribution notices, quarterly reports and investor due diligence questionnaires. That makes AI in private equity mostly a document problem. Good starting points:
- Extracting key terms from partnership agreements and side letters into a structured record your team can search.
- Tracking side letter obligations, such as reporting commitments and notice requirements, so none is missed.
- Drafting capital call and distribution notices around figures produced by the fund accounting system.
- Preparing first drafts of investor questionnaires from previous approved answers.
The same rule applies as in settlement: AI reads and drafts, the numbers come from the system of record, and a person approves what goes to investors. The guide to intelligent document processing explains how the reading part works and why it fails.
Where to start
- Pick one process with high volume and frequent exceptions, such as settlement breaks or routing incoming documents.
- Map it as it runs today, and mark every step that cannot be undone.
- Collect real past cases for an evaluation set, including the ones that went wrong.
- Run the agent in shadow alongside your team and compare results before it touches anything live.
- Widen its scope only when the evidence supports it, and keep the irreversible steps with people.
If agents are new to your team, what an AI agent is explains the difference between an agent, a chatbot and a fixed automation.
Who should not start here
If your operations depend on spreadsheets that change shape every month, or nobody clearly owns the process, fix that first. AI on top of an unowned process produces unowned errors. And if you outsource fund administration entirely, start by asking your administrator how it uses AI and which controls sit around it, rather than building something yourself.
If you run operations at an asset manager, private equity firm or fund administrator and want a second opinion on where to begin, book a free 30-minute AI consultation, and we can look at which part of the operation to start with.