Guide
Your people have the licences. Here is why they do not use them.
The usage report says forty people out of two hundred opened the tool this month, and the forty are the same forty as last month. That is the shape of AI adoption at most investment firms a year after the purchase, and it has one cause: a licence was treated as a rollout.
This guide says what adoption actually looks like inside a private equity, credit or hedge fund firm, which number to watch instead of utilisation, and the order the work goes in. It is written from over 100 workshops at more than 50 financial services firms, and from the months on retainer that follow them.
The number
how to increase ai adoption in my firm
Licence utilisation is the wrong figure.
Utilisation counts people who logged in. The number that matters is how many people in the firm use these tools well: on their own work, to a standard a partner would sign, without a colleague beside them. That figure is usually far lower than utilisation, and it is the only one that moves the business.
The forty who show up in every report are the early adopters. They needed the least help and they are already counted. The firm’s return sits with the hundred and sixty who tried it once, got a confident wrong answer on a document they knew well, and went back to the way they worked before. They are the number, and they do not move on their own.
Why they stopped
employees not using ai tools after training
A confident wrong answer, on a document they knew.
Ask the analyst who stopped and the story is the same at every firm. They pasted a CIM or a credit agreement into the tool, asked for a summary, and got back something fluent and subtly wrong: a covenant misread, a number from the wrong year, a definition the document negotiates differently from the market. They caught it because they knew the document. They concluded that the tool cannot be trusted on anything they do not already know, which is the only kind of work worth delegating. So they stopped.
The failure was in how the question was asked, and nobody had taught them that. Generic training teaches prompt tricks on generic material. The judgment an investment professional needs is specific: what to give the model, what to tell it about the firm’s own process, how to check the output. Nobody was taught that, because the course was bought before anybody asked what the work was.
What has to be true
ai enablement framework for financial services firm
Five things, before the question has an answer.
“How do we get people to use AI?” has no answer until five conditions hold. They are checkable from the inside in an afternoon, and at most firms two or three of them fail.
- Access nobody has to ask for
- The tool is open on the desk where the work happens, inside the firm's data terms, with no ticket, no pilot list and no approval per use. A tool you have to request is a tool you use twice.
- A line short enough to read
- One page that says what may go into the tool and what may not, in the words of the desk rather than the vendor's. Nobody reads a forty-page policy, so in practice they assume the answer is no.
- Patterns from your own work
- Worked examples on the firm's own documents and the firm's own process: how this desk reads a CIM, how this credit team checks a covenant. A prompt library built on somebody else's work is a library nobody opens.
- Somebody to ask
- A person, named, who the analyst can turn to when the output is wrong and they cannot see why. In the first months this is the single thing that separates the firms where adoption spreads from the ones where it stalls.
- A way to see what stuck
- A measure of who uses the tools well, by desk, read monthly, in a form somebody who was not in the room can understand. Without it the firm is back to the utilisation report.
These five are set out in full on AI Enablement, with what we do about each.
The order
ai rollout plan private equity firm
Workshop, then the desks that stalled, then one workflow.
The work goes in this order because each step tells you whether the next is worth buying. A firm that starts with a platform build has bought the last step first.
- Do real work once, with someone watching
- A session of ninety minutes to two hours per team, on the documents they already have open. The deal team on a live CIM, the credit team on an agreement, IR on the inbox. Each person leaves having done one piece of their own work correctly, which is the experience the generic course never gave them.
- Go to the people who stopped
- Month to month: prompting practice across the whole employee base rather than the early adopters, and direct work with the teams who tried it once and went back. This is where the number moves, and it moves one desk at a time.
- Design one workflow with AI at its centre
- Pick the process with real volume and real cost. Interview the people who run it, document it as it actually operates, and ask which steps remain when a model does the reading, drafting and reconciling. Then the firm's own people build it, so it is theirs to run on the first day.
- Read the number monthly
- A plain account of what changed, by desk, written so somebody who was not in the room can read it. When the figure stops moving, the account says where and why, which is the start of the next month's work.
What it looks like
what does good ai adoption look like at an investment firm
The owner confirms rather than retypes.
At a firm where this has worked, the read that used to take a week per contract is a structured first pass across the whole book, and the analyst spends the week on the exceptions. Investor emails that stayed in inboxes reach the record, triaged, in minutes rather than days. The person who owns a number confirms it instead of retyping it from three sources. None of this is a chatbot. It is the same work, with the reading done by a model and the judgment kept by the person whose name is on it.
Questions
The questions that follow.
What a partner, a COO or a head of technology asks once they accept that the utilisation report is the wrong number.
How long does it take to get a private equity firm's employees using AI?
A desk changes in one session if the session is on its own documents, and that change holds only if somebody follows up with the people who stalled. Across a firm, the number of people who use the tools well moves over months rather than weeks, which is why the retainer is month to month rather than a project with an end date.
Should we train everyone or start with one team?
Start with the teams whose work has the most reading in it: deal teams, credit, investor relations, fund finance. Train each on its own material rather than all of them on one curriculum. A firm-wide session on generic examples is the training most of your people have already had, and it is why they stopped.
Do we need a custom AI platform before adoption will work?
No. Adoption fails at most firms on the five conditions above, none of which is a platform. Build one workflow after the people who will run it have learned to use the tools on their own work, and build it with them, or it is a system nobody trusts.
What should we measure instead of licence utilisation?
How many people use the tools well, by desk, read monthly. Well means on their own work, to a standard a partner would sign, without help. For a single workflow, measure the time and the error rate on the firm's own volume before and after.
Our people say AI is inaccurate on our documents. Are they wrong?
They are describing a real failure, and it has a cause that can be fixed. Most inaccuracy on investment documents comes from how the question was asked and what the model was given, which is teachable. The guide on AI accuracy on this site goes through the five things that cut it.
Who should own AI adoption inside the firm?
Somebody who performs the work, named, with time set aside. At most firms it has been handed to technology or to the committee that approved the purchase, and both are too far from the desk. The role is to be the person an analyst can ask when the output is wrong.
Ask what your number is.
Most firms know their licence count and not their adoption. Tell us which tools your people have and what you believe they are doing with them, and we will tell you what we would look at first.
Subject line: Adoption