AI value creation in the lower mid-market: why access is not a strategy
by Scott Sheldon
Ask an owner-managed business in the UK whether it uses AI and the answer is almost always yes. Someone drafts emails with it, someone summarises documents, the finance team has it running inside Excel. Ask what AI is doing for the business as a whole and you tend to get a longer pause. For a private equity investor, that pause is where AI value creation now sits. Closing the gap has surprisingly little to do with the technology itself; it is a decision about how the business uses AI as a team.
We have spent the past year working through exactly this with our own investment team at Rockpool. What follows is what we learned along the way, and where we think it carries over to the wider lower mid-market.
Phase one: everyone gets access
The first step every firm takes is the obvious one: buy the licences, run a session on prompting, let people get on with it. And it does work. Individuals get quicker at first drafts and research, and the keener users build habits that genuinely save them time each week.
Step back and look at how the tool is actually being used, though, and the picture is less impressive. Each person is working alone in a private session. The prompt one person wrote to pull the numbers out of an information memorandum sits in their chat history and nowhere else, so a colleague doing the same job the next day starts from a blank page and produces a different answer in a different format. In effect, the firm bought one tool and ended up with dozens of highly personalised versions of it. There is nothing wrong with that as far as it goes. It just doesn’t scale.
This is where most businesses we see have got to, and it is where most of them stop, because the individual gains are visible enough to feel like the job is done.
Phase two: AI workflows the whole team runs
The second phase starts when the question changes from “how do I use AI for this task?” to “how should AI fit into this workflow?” Once the answer is shared, written down and repeatable, you are no longer talking about personal productivity. You are talking about how the firm operates.
Reviewing new investment opportunities is a good example, partly because every private equity firm does it constantly and partly because no two people do it quite the same way. How thorough a review is often depends on whatever else is competing for attention that week. So we rebuilt it as a defined workflow. Deal materials go into a shared environment the AI can access. A standard first-pass review extracts the facts with page and document references, raises flags against thresholds we have agreed as a firm and, just as usefully, points out what the document does not say. That review feeds a desktop market check run to the same standard every time. The IM financials seed a structuring model, with the market check shaping the assumptions behind it. By the time a deal is discussed internally, the numbers, the market read and the open questions are already in a format everyone recognises.
A strong analyst working alone could produce any one of these components. What they could not produce is consistency across the team: output that can be audited, and Monday’s work sitting there for whoever picks the deal up on Wednesday. Our judgement about what matters, what gets flagged and where our thresholds sit is now written into the workflow rather than carried around in people’s heads.
Phase three: connecting AI to the systems where the work lives
Phases one and two share a weakness: somebody still has to take the work to the AI. Download the document, upload the document, paste the email, copy the numbers across. Every task starts with a small manual step, and that friction matters more than it sounds. People who have both the tool and the workflow will, on a busy afternoon, still do the task by hand, because assembling the inputs in an AI-friendly format is a chore.
The third phase removes the chore by connecting the AI directly to the email client, the document store, the CRM and the calendar. The instruction changes from “here is a document, review it” to “review the document that arrived this morning, check whether we have seen this business before, and set up the project.” The work no longer has to be carried to the tool.
We are in the middle of this step ourselves, and it is the one we expect to matter most, because at this point AI is simply part of how the firm runs. It is also the step that forces a governance conversation many businesses have been putting off. When the tool can read a shared drive, who decides which folders? When it can see a mailbox, whose? Every firm will land somewhere different here. Our position is that the AI inherits the permissions of the individual using it, within a firm-level view on which data we are comfortable exposing at all.
What AI value creation looks like in a portfolio company
None of this is specific to private equity. A specialist manufacturer reviewing supplier quotes, a care provider handling regulatory correspondence, a software business triaging support tickets: each has a handful of recurring tasks that are done often, done inconsistently, and rely on inputs already sitting in a system somewhere.
The strategic question for a management team is which three or four of those tasks, done the same way every time and fed directly from the systems the business already runs, would change the week for the people responsible for them. Mostly what gets removed is toil: preparing data, moving information from one system into another, chasing inputs before the real work can start. What comes back is capacity for the work that actually needs judgement.
The practical route in is unglamorous. Write down the workflow as it is done today, how long it takes, what the output looks like, which tools it touches. Decide with the team which steps can be automated and where a person still needs to weigh in. Afterwards, measure whether the task takes less of an employee’s time and whether the output is more consistent than it was. If both are true, the initiative worked.
That is operational value creation in a very plain sense. It shows up first as capacity and consistency, and over time in margin. What it needs is someone senior prepared to look at AI organisationally rather than personally.
Where a lower mid-market business should start
If your business is in phase one, the next step is not more licences or a newer model. Pick one recurring task, put the people who do it in a room and agree how it should be done. Build that into a workflow everyone uses, then trace where the inputs come from and ask whether the tool could reach them itself.
Almost every business is using AI now. Over the next few years, the difference will be made by the ones that decided how.
