EY published its 2026 Global Private Markets Survey this week, and two numbers in it belong side by side. Nearly half (48%) of institutional investors expect fund managers’ use of AI to lead to lower fees. Meanwhile, 45% of managers report only modest returns on their operational investment in AI so far.¹
That gap is going to shape a lot of conversations in 2027. The same survey found that just 21% of institutional investors favor the traditional “2 and 20” fee model, compared with 45% of managers.¹ If LPs believe AI is making fund operations cheaper, and managers can’t show where the savings went, fees are where that disagreement gets settled.

We have a stake in this question, and not only as advisors. We run OakTech on an internal AI operating system we call ACT-EV. It reads our mailboxes, calendars, CRM, task lists and meeting notes, turns what it finds into owned and dated actions, and reports what it did and what it couldn’t do. Here is what that has actually delivered, with real numbers and no client data.
Lesson 1: Speed Comes From Push. Trust Comes From the Sweep.
Our original design checked each source on a schedule. Reliable, but slow. By our own measurements, the old path took a median of 138 seconds from a new email arriving to the system acting on it.
We then moved to real-time notifications. The mail platform tells the system a message has arrived, and the system processes that single item immediately instead of re-reading the mailbox. On real mail in our development environment, the median fell to 10.7 seconds, with a range of 9.1 to 15.1 seconds across 11 messages. Our target had been under 30 seconds.
What we didn’t do is switch off the schedule. Push notifications are best effort. Google’s Gmail API documentation says that “in rare situations, notifications might be delayed or dropped,” and tells developers to keep syncing anyway.² So a sweep still runs every 15 minutes and reconciles anything the fast path missed. Without that, a single lost notification could mean a lost LP email, and nobody would know.
Lesson 2: The Most Valuable Number Is Often Zero
When we turned on hourly action intake in production, the first live run created 37 new actions. It also held back 44 items because they duplicated work already being tracked, and it sent 70 items to a review queue instead of acting on them. A rerun minutes later created zero new actions.
That zero is the number we were proudest of. Every item is keyed to its source, its native ID and the specific change, so seeing the same thing twice changes nothing. Ask any AI vendor what happens when their system sees the same email twice. If the answer isn’t “nothing,” keep looking.
Lesson 3: No Owner, No Automation
Of those 70 items sent to review, 68 had no clear owner. The system could have guessed. It didn’t.
Our rule is that an action goes live automatically only when the system knows who owns it. Everything else is indexed and queued for a person to decide. In a small GP team, a task that lands on a partner’s list without context costs more attention than it saves.
Lesson 4: Real-Time Can Cost Less
We assumed real-time processing would cost more. It didn’t. Handling one changed item takes one to three connector calls, against the 10 to 20 a full scheduled pass needed for each batch of events. We estimate the added infrastructure cost of the push notifications themselves at roughly $0 to $2 a month.
This is the part of the EY gap we find most interesting. Many AI workflows are expensive because they re-read everything to find one change. Design for the smallest unit of work and the economics change.
Lesson 5: Reliability Is Something You Measure
We track how often each way into each source actually works, and we flag any access path whose seven-day success rate drops below 95%. We alert three days before any login token expires. And we make sure a backup path never shares the same sign-in, machine or method as the primary, because a backup that fails for the same reason isn’t a backup.
None of this is visible to the people the system serves. All of it determines whether they trust it.
So, Should AI Lower Fees?
Here is our opinion, for what it’s worth. AI will lower the cost of some fund operations, and LPs are right to expect part of that benefit. But the managers who handle this conversation well won’t be the ones who promise savings. They will be the ones who can show their LPs, specifically, what the AI does, what it costs, what it saves and how it is controlled.
The 45% of managers reporting modest returns are not wrong about their results. In our experience, modest returns usually come from bolting AI onto a few tasks instead of rebuilding the process around it, with guardrails, measurement and a clear owner for every action. That rebuild is less glamorous than a new model, and it is where the returns are.
Where a Fund Manager Can Start
- List every source your pipeline actually runs through: inboxes, calendars, CRM, notes.
- For each, write down what happens today when an important message arrives late on a Friday.
- Pick one workflow, measure it for a month, and only then decide what AI should do in it.
This is the work we do with clients in Business Process AI (BPAI), and it underpins the investor operations we build for emerging managers in CRS Launch. OakTech has helped clients raise more than $235 million. We now run our own firm on the same discipline.
See What an AI Operating System Could Do for Your Firm
If your fundraising or deal flow runs through a crowded inbox, we can map where the gaps are and what an AI operating layer with real guardrails would look like for your team.
References
- EY. “As private markets continue to grow, so do expectations, with AI raising the stakes, EY survey finds.” 2026 EY Global Private Markets Survey press release, November 18, 2026. https://www.ey.com/en_gl/newsroom/2026/11/as-private-markets-continue-to-grow-so-do-expectations-with-ai-raising-the-stakes-ey-survey-finds
- Google for Developers. Gmail API: Push Notifications, Reliability section. Accessed November 2026. https://developers.google.com/workspace/gmail/api/guides/push
The next question is which work to redesign first. Our business automation guide covers the building blocks; BPAI starts with an AI assessment of how your business actually runs.