Opinion · AI in fund ops

"Why can't I just use AI to do this myself?" The hidden cost of DIY AI in fund operations

It's the most common question we hear from capable engineering teams at PE, VC and hedge funds. Here's why the answer is almost never "just pipe in the OpenAI API."

menu_book Start reading
Jonty Hurwitz
Written by
Jonty Hurwitz
Founder
Read time
7 min
Published
Recent

Why do we need a platform? We have a great dev team. We can just pipe Claude into our internal processes and automate this ourselves.

It's a perfectly logical assumption. If a large language model can instantly summarise a 200-page LPA (Limited Partnership Agreement) or extract figures from an unstructured capital call notice, why pay for a third-party platform to do it?

The answer lies in the fundamental difference between intelligence and operations.

An LLM is a brilliant, stateless reasoning engine. But running a fund requires state, governance, security, and exception handling. When funds attempt to DIY their AI strategy, they usually realise — about six months and a million dollars in — that they haven't built an operational solution. They've built a highly expensive science experiment.

Wrapping an API around an LLM won't solve your fund operations. Orchestration is the missing link.

01 · The smart intern problem

AI is stateless. Fund ops isn't

An LLM is like a brilliant intern with zero short-term memory. It can parse a document perfectly in isolation, but it has no concept of time or process.

The DIY reality
Your engineering team now has to build a custom state machine. They have to code the logic that says: if the AI extracts the KYC data on Monday, wait for the background check API on Wednesday, and ping the compliance officer on Friday if the LP hasn't replied.
The orchestration fix
Next Matter handles state management out of the box. The AI is simply an actor within a larger, stateful process. Next Matter remembers where every single operation is in its lifecycle, waking up the AI only when it's needed.
02 · The hallucination liability

Maker-Checker by default

When traditional software breaks, it throws a 404 and stops. When generative AI breaks, it confidently lies. In fund administration, a hallucinated decimal point on a capital distribution isn't a bug — it's an SEC violation and a breach of fiduciary duty.

The DIY reality
To make DIY AI safe, your engineers have to build a custom front-end UI just so human analysts can review the AI's work. They have to build role-based access controls so the person verifying the data isn't the same person who approved it.
The orchestration fix
Next Matter is built on strict Maker-Checker protocols. We assume the AI will occasionally get it wrong. The orchestrator automatically routes the AI's output to the right human, at the right time, in a clean interface for verification before any data hits your core ledger.
03 · Security & audit

Your CISO is terrified of AI. They should be

Public LLMs use conversational data to train their models, and funds are dealing with highly confidential, market-moving alpha.

The DIY reality
If you build it yourself, you're responsible for proving to auditors that no proprietary LP data leaked out — and you have to build custom audit logs to track exactly what the AI did, what the API returned, and who approved it.
The orchestration fix
Next Matter provides enterprise-grade, sandboxed AI. No model training on your data. Every step of the orchestration is logged — both human and machine. When the FCA, SEC, or an LP asks for an audit trail, you export a pristine, time-stamped log.
04 · The bottom line

Be a fund, not an AI infrastructure company

To be clear: you absolutely should be deploying AI agents, and you should be using your best engineers to do it. The question is where you point them. Their edge is building agents that understand fund services, your LPs, your strategies, your data. Their edge is not rebuilding the AI infrastructure underneath those agents.

Building that infrastructure yourself is a triple tax. It is questionable from a compliance perspective, because you are now the party responsible for proving model isolation, audit trails, prompt injection defences and data residency. It burns tokens, because every team rediscovers retries, caching and guardrails the expensive way. And it burns headcount, because the people who should be shipping investor-facing agents are instead maintaining a private LLM platform.

The smarter move is to partner with a team that is AI-first and obsessed with AI infrastructure for asset services as their core product, and spend your own AI budget on the agents only you can build. Your engineering team should be focused on proprietary alpha, deploying capital faster, and improving investor relations. They should not be spending their sprints building rate-limiters, dead-letter queues, and Maker-Checker UIs for ChatGPT.

You don't need to build the infrastructure. You need to orchestrate the outcome.

With an orchestration layer like Next Matter, you get the transformative power of AI wrapped in the compliance, auditability, and human-in-the-loop governance that institutional finance demands.

Keep reading

See orchestrated AI inside a real fund ops stack

Book a working session and we'll show you what agentic orchestration looks like on your workflows.