Almost every AI conversation inside a portfolio company starts with the same question. What can we use AI for. The operating partner asks it. The CEO repeats it. A working group forms and comes back with a list of use cases ranked by effort and impact.
That question is not wrong. It is just shallow. It treats AI as a feature you bolt onto the business you already run.
The deeper question is the one almost nobody asks. What is our operating model when AI is orchestrating it. Not which tasks can we hand off, but how does the company decide, learn, and act once AI sits in the middle of the work. Answer that and the use cases sort themselves out. Skip it and you end up with a portfolio of pilots that demo well and change nothing.
The question under the question
When you ask “what can we use AI for,” you are looking for places to insert a tool. The frame is additive. You keep the existing workflow and you find spots where AI saves time.
When you ask “what is our operating model when AI orchestrates it,” you are looking at the whole system. How decisions get made. Where judgment lives. What data the company relies on to know it is right. The frame is structural.
The two questions produce very different roadmaps. The first gives you a faster version of what you already do. The second gives you a different company. Only one of those moves a multiple.
I have written before about why your portfolio company’s AI initiative is stalling. The data underneath is usually the proximate cause. But there is a strategic cause that sits above the data problem, and it is worth naming on its own. Most companies are asking AI to do the thinking, when the defensible move is to make AI amplify thinking the company already owns.
The risk nobody prices
Here is the part that should worry anyone responsible for a hold thesis.
A thousand companies have access to the same models. If a thousand companies ask the same model the same question, a lot of them get the same answer. The model is trained on the public record. It is very good at producing the consensus view of any problem you hand it.
So when your team prompts a general model for a market entry strategy, a pricing framework, or a growth plan, what comes back is competent and generic. It is the answer the model would give your competitor too. It is the answer it would give the firm bidding against you.
A strategy any competitor can also generate is not an edge. It is table stakes dressed up as insight.
This is the trap hiding inside generic AI strategy. The output feels like progress because it is articulate and fast. But if the same prompt produces the same plan across the industry, you have not built a moat. You have rented the average. And the average does not expand a multiple, because the buyer’s diligence team has the same model and can see the plan was never proprietary.
What actually defends a position
The edge is not in having access to AI. Everyone has access. The edge is in what you feed it and what you do with what comes back.
Two things are proprietary to your portfolio company and to no one else. The first is your data. The actual record of your customers, your transactions, your operations, your failures and recoveries. No competitor has it. No general model was trained on it. The second is your judgment. The accumulated sense your operators have for what works in your specific market, with your specific customers, under your specific constraints.
Defensible AI use amplifies both. It takes the proprietary data the company already holds and the judgment the operators have already earned, and it makes them faster, sharper, and more consistent. It does not replace the thinking. It compounds it.
A generic model asked “how should we price this” gives you the textbook. The same model, grounded in your pricing history, your win-loss data, your margin by segment, and steered by an operator who knows which accounts will tolerate an increase and which will walk, gives you something no competitor can reproduce. Same tool. Completely different output. The difference is the proprietary inputs and the human judgment in the loop.
That is the whole game. The model is the commodity. Your data and your judgment are the assets. Decision sovereignty means the company keeps ownership of the decision and uses AI to make it better, rather than handing the decision to a system that gives everyone the same answer.
Why this is a data problem before it is a strategy problem
You cannot amplify proprietary data you cannot access. This is where the strategic ambition meets the unglamorous reality on the ground.
If your customer data is fragmented across four systems that do not agree, you cannot ground a model in it. If your revenue does not reconcile, you cannot trust what the model returns. If nobody can produce a clean view of margin by product line in under a week, then your proprietary advantage is locked in a vault you do not have the key to.
This is why AI readiness starts with data readiness. The companies that win with AI are not the ones with the best prompts. They are the ones whose proprietary data is clean, reconciled, and accessible enough to feed the model. Your competitive edge is only as good as your ability to get your own data into the decision.
You can test where you stand with the AI readiness assessment. It will not tell you which model to use. It will tell you whether the data that holds your actual edge is in a state you could ground a model in tomorrow.
What it looks like in practice for a portfolio company
Skip the all-portfolio AI strategy deck. Start with one decision that matters and is made often.
Pick a recurring, high-value decision the company already makes on judgment and incomplete information. Pricing a renewal. Prioritizing the sales pipeline. Deciding which customers to invest retention effort in. Forecasting demand for a product line. These are decisions where your operators have real instinct and your systems have real history.
Map how that decision is made today. Who makes it. What they look at. What they wish they could see but cannot get to fast enough. Where their judgment adds value and where they are just doing manual lookup that a machine should handle.
Then ground a model in the proprietary data behind that decision. Not the public record. Your record. The win-loss history, the cohort behavior, the operational signals specific to your business. The model’s job is to surface the pattern and present the options. The operator’s job is to apply the judgment the data cannot capture.
Keep the human in the decision. The model proposes. The operator decides. Over time the model gets better at surfacing what matters because it is learning from your data and your operators’ choices, not from the internet’s average opinion. The decision stays yours. The work behind it gets faster and sharper every cycle.
That is one decision, made measurably better, on a foundation no competitor can copy. It is worth more than ten generic pilots, because it deepens an edge instead of renting one.
A note on the reorg reflex
When the AI plan stalls, the instinct is to restructure around it. Stand up an AI function. Hire a head of AI. Redraw the org chart so somebody owns the mandate.
That rarely touches the actual constraint. I have made this case at length in the reorg that fixes nothing. The constraint is almost never the org chart. It is that the proprietary data which would make AI defensible is not accessible, and the decisions you would want to amplify have never been mapped. A new box on the chart does not fix either. It just gives the problem a nicer reporting line.
The bottom line
The wrong question is what can we use AI for. The right question is what is our operating model when AI orchestrates it.
Get that right and the rule that follows is simple. Use AI to amplify what only you have, your proprietary data and your operators’ judgment, and keep ownership of the decision. Use it to outsource the thinking and you will get the same answer your competitor gets, because you are both asking the same model the same question.
The firms that build durable value in their portfolio over the next few years will not be the ones with the most AI tools. They will be the ones whose companies kept decision sovereignty, fed the model what no one else could, and turned a commodity into an edge.
For a weekly brief on AI deployment, data strategy, and PE portfolio operations, subscribe to Inside the Data Room.