Every operating partner has by now delivered some version of the same mandate to portfolio company leadership: adopt AI, and do it now. Delivering the mandate is the easy part. The hard part, the part that separates an operating improvement from a slide with AI on it, is what happens six months later: does the operating partner know which portcos actually adopted it, what it costs, whether the data governance underneath it is sound, and whether any of it produced something measurable at all. This is a guide to answering all four.
The mandate is not the plan
Telling a portfolio company to use AI produces uneven results almost by design: one portco's IT lead licenses a single chat tool and calls it done, another lets every team pick its own subscription, and a third does nothing because nobody owns the follow-through. None of these outcomes is visible from the operating partner's seat unless someone builds visibility in from the start. A mandate with no way to check on it isn't governance; it's a hope.
Get visibility into adoption and spend, per portco, not just anecdotes
The natural reporting cadence for AI at most portfolio companies today is an anecdote at the quarterly review: someone mentions a chatbot that saved the support team time, or a deck that got built faster. That's a data point, not a picture. What an operating partner actually needs is the same view finance already expects for any other cost line: what each portco spends on AI, which teams use it, which models, and whether usage is growing in a way that matches the value it produces. Without that, we rolled out AI and AI is working are two very different claims being treated as one.
Shadow AI is a data-governance risk wearing a productivity story
Where a sanctioned tool doesn't exist, or is worse than what people can get on their own, employees route around it, often on a personal card, often with sensitive customer, financial, or clinical data pasted directly into a tool nobody at the company approved or can see. This is not a hypothetical sitting at the bottom of a checklist; it is the default outcome of an unmet mandate at a company full of people who already have a phone and a personal AI subscription. An operating partner who doesn't know whether a portco has this problem should assume it does, until shown a governed, adopted alternative.
Benchmark across the portfolio, not just within one company
A single portco's AI spend and adoption numbers, viewed alone, tell you very little about whether they're good or bad. The value of an operating partner's seat is being able to look across a dozen companies at once and ask which ones are getting real value for their AI spend, which ones are overspending on the wrong models for routine work, and which ones haven't gotten past the pilot stage at all. That comparison is only possible when every portco reports on AI usage the same way, which almost never happens on its own and has to be designed in.
Tie usage to outcomes, not tool sprawl
A portco with five different AI subscriptions and no way to say what any of them delivered has not adopted AI; it has adopted vendor sprawl wearing AI's name. The metric that matters is not how many licenses were purchased or how many employees logged in once. It is whether specific workflows, a faster monthly close, a lighter support queue, fewer hours drafting a first version of a document, actually changed. Getting to that answer means connecting spend and adoption data to the workflows they touched, which is a reporting discipline, not a one-time audit.
What to ask for at the next portfolio review
- Total AI spend by portco, broken down by team and by model, not one company-wide number.
- Which models are approved at each portco, and for which categories of data.
- Evidence of at least one workflow where AI usage produced a measurable operating improvement.
- Confirmation that provider credentials are held centrally rather than scattered across individual logins.
- A comparison across portcos on spend per seat and adoption, not just each company's own trend line.
How Switchboard helps
Switchboard gives an operating partner the visibility this guide describes without asking every portco to build its own reporting from scratch. Each portfolio company runs on the same governed layer: one login to frontier and open-weight models, per-team budgets and model allowlists set once and enforced automatically, and provider keys held centrally instead of scattered across personal accounts. Every request is attributed to a person, a team, and a portco, so an operating partner can pull one view across the whole portfolio, spend, adoption, and model mix side by side, instead of chasing a different answer out of every company on the quarterly call. The mandate to use AI becomes something the operating partner can actually verify, portco by portco, rather than something each company reports back on its own terms.