Why the fund administration business model is about to change shape
By Kumar Ujjwal, Founder and CEO, DwellFi
The fund administration business model has run on labor arbitrage for three decades. Win a mandate, staff it with people, keep the margin left after their salaries. That model is not collapsing. It is being rebuilt, because AI is the first technology that does the work itself instead of making the worker faster, which turns the cost of the next fund from a hiring decision into a software decision.
This is not a fund-ops curiosity. Gartner reports that 71% of CEOs say their IT operating models are not fit for the age of AI. The same diagnosis lands harder in fund administration, where the operating model is almost entirely people and the people are almost entirely doing work a machine can now do. The firms that read this as an efficiency story will optimize the old model. The ones that read it correctly will rebuild it, and come out the other side with a better business and a better job for the people in it.
Key takeaways
- The fund admin margin has always been set by labor cost, not by technology.
- Every prior software wave made your teams faster without changing the model.
- AI changes the unit of production from a headcount to an agent.
- The head of fund services who rebuilds the cost structure, not just the speed, owns the next decade.
How has fund administration made money for the last 30 years?
For three decades, fund administration has earned the spread between what a client pays to service a fund and what it costs to staff that service. The mechanics are simple and they have barely moved. A client signs on, you assign three to five people to the account and profit is whatever survives those costs.
So the business runs on two variables. How you spend to hire talent. How long you can keep them before they cross the street for a raise. Everything else is detail.
This is also why the industry consolidated into a handful of large players. Scale spreads overhead and lets you recruit in volume, which protects fund admin margins at the edges. It does not change the arithmetic underneath. The cost of servicing a fund is still, at bottom, the cost of the humans servicing it.
Why didn't earlier software change the fund administration business model?
Earlier software automated steps, not judgment, so it made administrators faster without ever removing the people who carried the margin. Fund admins have bought technology for decades. General ledgers, fund accounting platforms, investor portals, document management, all of it genuinely useful, none of it load-bearing on the model.
The reason is structural. Traditional software hands a person a quicker way to do a task they were always going to do themselves. The reconciliation still needs someone to run it. The report still needs someone to assemble it. Software changed how many accounts one person could carry, and the headcount-per-mandate ratio compressed slowly across thirty years. It never collapsed, because the person never left the loop.
The model held because the technology only ever touched speed. It never touched the work.
What makes AI different for fund operations?
AI does the work. That is the entire difference, and it is easy to underrate, because the early demos look like the old software with a nicer interface.
A deterministic AI agent does not hand a faster tool to the reconciler. It runs the reconciliation, surfaces the breaks, and waits for a person to judge the exceptions. The work shifts from doing to checking. Someone who used to carry ten accounts can oversee the agents carrying many times that, and the second number is not a productivity gain. It is a different job.
Here is the part that matters for AI fund operations specifically, and it is also the part generic chat tools get wrong. The work is regulated. A reconciliation that produces a different answer on a second run, or that cannot show how it reached the first one, is not faster. It is a liability with a shorter fuse.
This is where the cost question bites: Gartner estimates that by 2028, the cost of governing AI will offset 60% of the savings agentic AI generates. Read that as a warning about the wrong kind of AI. Bolt governance onto a system that guesses and you spend the savings policing it. Build on a system that is deterministic and auditable by construction and the governance is the architecture, not a tax on top of it.
When an agent clears that bar, the cost of a new mandate stops being a hiring question. You do not staff up to win the account. The marginal cost of the next fund approaches the cost of the software running it. That is a different business, with a different margin structure and a higher ceiling on how many funds a single team can hold.
Will AI replace fund administrators?
No. AI will not replace the fund administrator. It replaces the manual processing inside the role and moves the administrator into oversight, exception handling, and the client judgment that was always the valuable part.
The exposure is uneven, and it is worth being precise about where it falls. Roles that are purely manual data handling, the keying and matching and assembling, are aimed at directly. Roles built on judgment and relationships are not replaced. They are freed, because the person doing them stops spending the week on processing they should never have been doing by hand.
For the head of fund services, that is the opportunity hiding inside the disruption. The team you manage stops being a processing line you defend on cost and becomes an oversight function you grow on value. Same people, better work, and a service you can scale without the linear hiring that used to cap it.
Who is most exposed?
The administrators whose entire advantage is cheap, well-managed labor. If your edge is that you recruit and retain operations staff better than the firm across the street, AI is pointed directly at the thing you are good at, and being good at it will not save the model.
The best-positioned firms treat AI as a cost-structure rebuild rather than a productivity perk. Saving an analyst an afternoon is a rounding error on the P&L. Changing how many funds a team can administer is the whole game, and the advantage compounds, because the firm that starts now builds proprietary structure around its own data that a late starter cannot buy back quickly.
What should a head of fund services do about it now?
Stop framing AI as efficiency and start treating it as the substrate the next decade of margin runs on. The practical first move is narrow. Pick the highest-volume, most rules-bound workflow you own, which is almost always reconciliation, and run it on deterministic, auditable AI inside your own environment. Then measure the new ratio. Accounts per person, not minutes saved.
That distinction is the tell that separates a leader who gets this from one who does not. Minutes saved is the old model, optimized. Accounts per person is the new model, started.
Frequently asked questions
Have more questions about how AI is transforming fund operations? I’ve got them answered.
Will AI replace fund administrators?
Not the function. The function becomes oversight of AI agents rather than manual processing. The roles most exposed are the purely manual data-handling ones. Judgment, client relationships, and exception handling stay human, and arguably become the entire job.
What is deterministic AI in fund operations?
It is AI that produces the same output from the same input every time, by an explicit rule you can read and reproduce, with a record of how it got there. It is the opposite of a general chatbot, which predicts a plausible answer and can give you a different one on the next run. For regulated fund-ops work, determinism is not a nice-to-have. It is the condition for using AI at all.
Is this just hype?
The distinction that matters is between AI that demos and AI that ships into a regulated workflow. Deterministic, auditable systems are already running production fund-ops tasks. The hype lives in the consumer-chatbot framing, not in the back-office application.
How fast is this happening?
Faster than most boards assume, because the advantage compounds. A firm that starts now builds proprietary structure around its own data, and a late starter cannot buy that time back.
The future of fund operations: the operator moves up the value chain
The operating model that is changing shape was never the rewarding part of this business. It rewarded firms for hiring cheaply and running analysts through reconciliation a machine should have been doing all along. What comes next points the head of fund services and their team at the work that was always the point: the exceptions, the client relationships, the judgment calls, and the revenue-generating services that the old cost structure never left room to build.
DwellFi runs deterministic AI agents inside fund operations, in your own environment, on the work that has to be right. Reconciliation, NAV close, capital calls, investor reporting.
Not a faster version of the old model.
The substrate for the next one.