By now, most firms have experimented with AI in some form. Some have rolled out enterprise tools. Others are testing point solutions. A few have tried to embed AI into core workflows. But for many, the results have been underwhelming.
Usage is inconsistent. Outputs are unreliable. The expected efficiency gains haven’t materialized.
This isn’t unusual. In fact, it’s predictable because most firms are skipping the foundational steps required to make AI work.
Before AI can deliver value, three prerequisites need to be in place.
1. Data Readiness
At a typical sub-200-person asset manager, critical data lives in a lot of places at once: investor reports assembled in Excel, DDQ responses buried in email threads, fund performance data sitting in a fund admin portal, portfolio company updates scattered across shared drives, and side letters filed in whatever document management system the firm was using three years ago.
That fragmentation isn’t unusual. It’s the norm. And it’s the first thing that stops AI cold.
AI doesn’t fail because it can’t handle complexity. It fails because it has no way to know which document, version, or number is the source of truth… and neither, in many cases, does your team.
Data readiness doesn’t mean you need a perfect data warehouse before you can start. But it does mean:
- Having clearly defined sources of record for your most-used workflows
- Being able to tell a new hire or an AI tool exactly where to find the right version
- Ensuring the data feeding any automated process is clean enough to trust
A simple test: if your team regularly has to reconcile two versions of the same report to figure out which one is current, AI will hit the same wall. That’s not an AI problem. It’s a data infrastructure problem that needs to be addressed first.
And for AI agents, the stakes are even higher. Agents don’t just produce outputs, they take actions. If the context they’re working from is incomplete or incoherent, those actions will be too. Data isn’t just an input. It’s the foundation every agent decision is built on.
The good news is that it doesn’t take a multi-year data transformation to make meaningful progress. Often, the first step is just picking one workflow and documenting where the data actually lives today.
2. Workflow Clarity
AI doesn’t operate in a vacuum. It enhances existing processes, which means those processes need to exist in a form that can be handed to a system.
In most firms, the real workflow isn’t written down anywhere. It lives in the heads of the two or three people who have always run the process. It varies slightly depending on who’s doing it. And it contains a lot of implicit judgment calls that have never been explicitly defined.
Ask five people how a DDQ gets assembled at your firm and you’ll get five different answers. That’s not a criticism, it’s just the reality of how most operational knowledge accumulates over time.
But AI requires clarity. Before you can automate or augment a workflow, you need to be able to answer:
- What are the actual steps, in order?
- Where does the input data come from, and who is responsible for it?
- What decisions need human judgment, and what can be rule-based?
- What does a correct output look like?
The firms that are seeing real traction with AI have done this mapping work first. Not as a lengthy consulting exercise, but as a practical exercise focused on one high-friction process at a time: DDQ production, investor reporting, portfolio monitoring updates, compliance reviews, or deal screening.
Until the workflow is clear to the people running it, it can’t be clear to an AI system built to support them.
3. Ownership
This is the most overlooked prerequisite, and arguably the most important.
In most firms, AI initiatives sit in a gray area. IT may own the platform. Operations may own the process. IR, deal teams, finance, and compliance may all touch the workflow. But if no single person owns the full transformation — the data, the workflow, the adoption, and the measurement — the initiative stays in pilot mode indefinitely.
The ownership gap is why so many firms have AI tools that people use occasionally but haven’t changed how work actually gets done.
Successful firms assign clear responsibility to someone who can bridge the operational and technical sides: someone who understands the business workflow deeply enough to redesign it, and understands the technology well enough to know what’s feasible.
This is not a purely technical role. It’s not a purely operations role either. It sits at the intersection, and it’s the profile that’s hardest to hire for right now, because most candidates have one side but not both.
Without this person in place, adoption stays optional. And optional adoption means AI remains an experiment rather than an operating improvement.
A Simple Diagnostic
If AI isn’t delivering value at your firm yet, three questions are worth asking honestly:
- Do we trust the data AI is using, and do we have a clear source of record for the workflows we’re trying to improve?
- Have we mapped how those workflows actually function today, including who owns each step and what decisions require human judgment?
- Is there one person explicitly accountable for driving adoption, measuring results, and closing the gap when it stalls?
If the answer to any of these is ‘not really,’ that’s most likely the root of the issue — not the tool itself.
The Takeaway
AI is not the starting point. It’s the accelerator.
The firms starting to see meaningful results from AI aren’t necessarily the ones with the biggest technology budgets. They’re the ones that took the time to get the foundations right first: cleaner data, documented workflows, and clear ownership of the outcome.
That foundation work isn’t glamorous. But it’s the difference between AI that changes how your team operates and AI that sits in a demo environment that nobody uses.