For a long time, there have really been two options for fund operations teams as workloads grow. You either build the capability internally, adding more people, technology and process as complexity increases, or you outsource parts of the work to a third party.
Both models can work. But the managed service model in particular hasn’t changed all that much over the years. Traditionally it has been based on labour arbitrage. Take an existing process, move it to a lower-cost location, put a team around it and drive efficiencies through scale and standardisation. AI gives us the opportunity to do something different, and that’s a big part of why we set up Fund Recs AI Ops.
The opportunity, as we see it, is to move away from labour arbitrage and towards efficiency arbitrage. Rather than simply asking where a process can be run more cheaply, we can ask which parts of that process actually need a person, and which are routine steps that software and AI can handle. By combining experienced operations people, purpose-built software and AI we think we can materially reduce the amount of manual work involved while still keeping the right level of human judgement, control and accountability.
That’s Fund Recs AI Ops: our AI-enabled managed service.
This thinking isn’t unique to fund operations. Sequoia Capital recently published an interesting piece on the changing relationship between software and services, exploring how AI is creating opportunities for software companies to move beyond providing tools and increasingly take on the work itself. It’s a broader shift that aligns closely with what we’re trying to do with Fund Recs AI Ops. (https://sequoiacap.com/article/services-the-new-software)
The gap between software and the end result
Fund Recs has always been a software company. We build the Fund Oversight Layer: technology to automate data workflows, reconciliations, oversight processes and exception management. We’ve seen first-hand how much operational work can be removed through good automation. But one thing becomes pretty obvious when you spend enough time with operations teams: having access to great technology doesn’t necessarily mean you have a fully efficient process.
A reconciliation might be 100% automated, but someone still needs to investigate and resolve the breaks. A NAV oversight process might be highly structured, but somebody still needs to gather the supporting information, investigate exceptions and prepare everything needed for the final review. A delegate oversight review might follow a clear framework, but the team is still assembling evidence from multiple administrators before it can begin.
A lot of operational work still sits in that space between automation and judgement. Historically, that has been difficult to automate because it involves context, investigation and a degree of interpretation rather than a simple rules-based task. AI is starting to change that.
Why build the managed service ourselves?
There is another important reason we wanted to build the team internally: it allows us to take the idea of eating our own dog food a step further.
We already use our technology extensively ourselves, particularly through client onboarding and support. AI Ops gives us the opportunity to go further by using the same platform and AI capabilities to actually run operational processes day in, day out. That puts us much closer to the work and gives us a different level of insight into where the technology is working well, where there is friction and where the next opportunities for automation are. There’s only so much you can learn by building AI functionality from a product backlog. You learn a lot more when your own operations team is depending on it to deliver a service for a client.
If an agent saves an analyst two minutes on a task that happens hundreds of times a day, we will see the benefit very quickly. Equally, if something looks impressive in a demo but isn’t actually useful in an operational process, we will find that out just as quickly. And if an experienced analyst instinctively looks for a piece of information or applies a judgement that we haven’t accounted for, that becomes direct feedback into the product. As we prepare for our first engagements, this is the operating principle we’ve designed around.
AI Ops is not separate from the product strategy; it is core to it. Running these processes ourselves gives us another source of real-world product feedback: we can see the repetitive work, edge cases and friction first-hand, improve the relevant workflows and AI capabilities, and bring those improvements back into the Fund Oversight Layer. Those same improvements flow back to every SaaS client, whether or not they use AI Ops.
So why do this now?
Operations teams are being asked to do more. There is more data, more oversight, more controls and more pressure to do everything faster, and the traditional answer has often been to add more people. That isn’t particularly scalable.
At the same time, AI has moved on significantly. There has been huge momentum around AI over the last few years, but we are now getting to the point where the conversation is moving beyond experimentation. In fund operations, there is a big difference between being able to demonstrate that AI can perform a task and being able to make it a trusted part of an operating process.
For us, that is the important shift. The challenge is no longer simply proving that AI can do something. It is designing an operating model where AI can take on the right parts of the work, with the right data, controls and oversight around it, and then constantly reviewing that model as AI capabilities evolve and the line between what AI can do and where human judgement is needed continues to move.
There is a huge amount of activity in fund operations that lends itself to that model. Gathering information, comparing data, reading documents, investigating exceptions, looking for patterns and preparing a decision are all areas where AI is becoming increasingly capable. Judgement, accountability and control still need to sit around that work, but that doesn’t mean a human needs to manually perform or approve every individual step. Instead, the oversight is automated and people review by exception.
It led us to a fairly simple principle: AI prepares the work. Humans stay in control.
Exactly where that human intervention sits will evolve. Lots of activities will require review or approval. Others can increasingly run automatically within clearly defined controls, thresholds and escalation points, with a full audit trail so the accountable person can see exactly what ran, what it found, and what it did. As AI becomes more capable and more reliable, we want the operating model to be able to move with it, allowing AI to take on more of the work and making the process more efficient, while changing those control points in a deliberate and controlled way.
That means we aren’t designing the operating model around what AI can do today. We are building it so that the balance between people, software and AI can continue to change as the technology improves.
That changes the starting point completely. Rather than asking, “How many people do we need to run the client’s existing process?”, we can ask, “What should this process look like if we designed it today using the best combination of software, AI and people?” And increasingly, we can keep asking that question as the technology improves.
Operations teams need a more scalable model, while AI is becoming capable of playing a meaningful role in parts of operational processes that have historically required significant manual effort. That is why we think the managed service model itself is ready to change.
Where we go from here
We’re still early, and we’re deliberately starting with a small number of launch engagements. The aim is to get deep into the underlying processes, understand how the work is actually being done and redesign those processes around the best combination of our platform, AI and experienced operations people.
Some parts of the process will automate extremely well. Some will continue to need experienced people for a long time. There will also be areas where our assumptions turn out to be wrong, and that’s part of the journey. We’d much rather learn by running real operational processes than by guessing from the outside.
Over time, I think the line between software and services is going to blur. The best operational platforms won’t just give people tools to use; they’ll increasingly take on parts of the work themselves.
We’re taking on a small number of launch engagements. If you want to redesign a process with us, get in touch: brian@fundrecs.com
