A portfolio company should start using AI where the business is most constrained, in the one workflow that limits revenue, margin or capacity today. Fix that workflow first, then apply a proven commercial tool to it and measure a single result before expanding anywhere else.
That order matters more than the tool. Most companies are not starting from zero with AI. Their people are already experimenting with it, and boards and operating partners want to see it in the value creation plan. That pressure pushes leadership teams toward the use case that demos well instead of the one that moves the business.
Why is "where can we use AI" the wrong first question?
Because the honest answer is "almost everywhere," and that answer is useless for planning.
Most leaders ask where they can use AI, but the better question is where is the business constrained? The first question produces a long list of pilots. The second produces a short list of problems worth solving.
A long list of pilots spreads budget and attention thin. Each one can prove a tool works in isolation. None of them proves the business is ready to scale it, and few connect to a number the board tracks.
Starting from the constraint changes that. You begin with a problem that already costs the company something, so the improvement is easy to see.
How do you find the real constraint in the business?
Look for the place where work waits. In a growing, acquisitive company, the constraint is rarely where leadership first points. It usually sits in a handoff between teams, sites or systems.
Common signs include:
- Work that queues up behind one person or one team before it can move
- Volume growth that requires adding headcount at the same rate
- Rework caused by the same data being entered in more than one system
- Decisions that stall because the numbers take days to assemble
- Office and field or plant teams working from different versions of the truth
Ask the people closest to the work, not only the leadership team. A COO may see a slow quoting process. The estimator doing the quoting can tell you it is slow because they are reconciling three spreadsheets that disagree.
Then put a cost on it: hours lost, orders delayed, margin left on the table. If no one can describe what the constraint costs, it is not the right place to start yet.
Why fix the workflow before automating it?
AI amplifies whatever it is deployed into. A clean workflow gets faster. A broken one gets broken faster, and at higher volume.
AI doesn't create ROI. Better operations do. The most common gap in AI plans is a tool that has been chosen for a process no one has mapped.
Before adding a tool, the workflow needs:
- A clear owner who is accountable for the result
- Defined steps, so you know what the tool is replacing or speeding up
- Usable, secure data in a place the tool can reach
- Access controls, so an AI assistant only sees what each user should see
Sometimes fixing the workflow solves most of the problem on its own. That is a good outcome. It means the AI investment, when it comes, starts from a stronger base and has a cleaner baseline to beat.
What makes a good first AI use case?
The first use case sets the tone for every one after it. It should be narrow, visible and measurable. Run each candidate through these five tests:
- It sits on a real constraint. It addresses the bottleneck you identified, not a nice-to-have.
- The workflow is already sound. Ownership, steps and data are in place.
- A commercial platform fits. Proven tools applied to a specific problem beat custom builds on speed and cost.
- The result is a number. Hours returned, cycle time reduced, capacity added without new hires. Set the baseline before you start.
- It can be judged within a quarter. A result the leadership team can see in 90 days builds confidence. A result two years out builds skepticism.
If a use case fails two of these tests, move it to the later list. It may be a good idea. It is not a good first idea.
When is it time to scale?
Scale after the first result is proven, measured against the baseline and owned by someone in the business. Not before.
Moving from one team using a tool to the whole company using it is where most organizations struggle. That step requires clear ownership, governance on how AI is used and a repeatable way to measure value. Clear policies are what make responsible adoption possible at that scale.
For a platform with bolt-ons, this is also the moment to set a standard. If the first use case lands on one set of tools, data practices and controls, each new acquisition can adopt it instead of solving the same problem again.
Then go back to the original question and find the next constraint. Strong AI programs in portfolio companies look less like a single initiative and more like a habit: find the bottleneck, fix the work, apply the tool, measure, repeat.
The principle
The best place to start with AI is the place the business is already stuck. Fix the work, prove one result, then scale what earned it.
Frequently asked questions
Should a portfolio company build its own AI model?
In most cases, no. Custom builds are slow, expensive and hard to tie to a measurable outcome. Proven commercial platforms applied to a specific business problem usually deliver results faster and with less risk.
Who should own AI in a portfolio company?
Each use case should have a business owner who is accountable for the result, not only an IT owner. Leadership should also assign clear ownership for AI governance, covering which tools are approved, what data they can reach and how value is measured.
Does the IT foundation need to be ready before starting with AI?
Yes. AI tools inherit the access controls, data quality and security of the environment they run in. Gaps such as unenforced multi-factor authentication or unclear data permissions become more expensive once AI is layered on top of them.
How do you measure ROI on a first AI use case?
Record a baseline before the tool goes live, such as cycle time, hours spent or volume handled per person. Measure the same figure after a set period and compare. If the result cannot be expressed as a number tied to the constraint, the use case was not defined tightly enough.
What if employees are already using AI tools on their own?
Treat it as a signal, not only a risk. It shows where people feel the friction in their work. Bring that use under clear policy, approve the tools that fit and point the effort at the constraint the business has chosen to fix.
