AI in portfolio companies: from pilots to impact

AI is often introduced into a process that was already unclear, inconsistent, or poorly governed.

The technology may make parts of that process faster without making the overall result better. It can also scale existing errors and ambiguity more efficiently.

Before automating work, the company should understand how the work is performed today, where decisions occur, what information is required, and who owns the outcome.

This does not require a lengthy process-redesign program. It requires enough operational understanding to ensure that the AI capability fits into a workable system.

The model is only one part of the solution. Data, workflow, controls, adoption, and accountability usually determine whether the solution becomes useful.

Companies often respond to AI interest by concluding that they first need a comprehensive data transformation.

That may be necessary for some ambitions, but it should not become the automatic starting point.

Data readiness should be evaluated against the specific use case.

What information does this solution require? Is it accessible? Is it sufficiently accurate and complete for the decision being supported? Can the company maintain it? Are there privacy, security, contractual, or regulatory limitations on its use?

Some opportunities can begin with a narrow and controlled data set. Others depend on information scattered across systems, documents, and individual employees. The difference has a significant effect on cost, risk, and time to value.

The goal is not perfect enterprise data. It is data that is appropriate for the intended outcome.

A pilot should not exist merely to prove that the technology works.

It should answer a business decision:

• Did the solution improve quality, speed, revenue, cost, or customer experience?
• Can employees use it effectively?
• Can the company operate and support it?
• Are the risks understood and manageable?
• Does the economic value justify further investment?

That requires a baseline and a defined standard for success.

Without them, an interesting demonstration can continue indefinitely because no one has established what evidence would justify expanding it or stopping it.

The discipline to end an unsuccessful pilot is as important as the ability to scale a successful one.

Many pilots stall because the path to normal operations was never considered.

A production capability needs ownership, support, security, monitoring, change management, and a process for evaluating performance over time. It also needs a clear understanding of how human judgment remains involved.

The company should know:

• Who owns the business outcome
• Who manages the underlying technology
• How output quality will be monitored
• What happens when the solution produces an incorrect or unexpected result
• How access and sensitive information are controlled
• How the process will change as the capability improves

These questions do not need to prevent experimentation. They should shape experiments that can become real operating capabilities when the evidence supports them.

Private equity firms have an opportunity to accelerate learning across the portfolio.

Common evaluation methods, vendor knowledge, security principles, and implementation lessons can reduce duplicated effort. Similar companies may also benefit from shared patterns and reusable capabilities.

But portfolio standardization has limits.

Companies differ in their customers, processes, systems, data, leadership, and operating maturity. A use case that works well in one company may have little value in another.

The most useful portfolio approach combines shared guardrails and accumulated learning with company-specific accountability.

The portfolio can make experimentation safer and faster. The individual company still needs to own the business result.

AI will continue to generate pressure for companies to act quickly. Waiting for complete certainty is not realistic, but neither is treating activity as progress.

Portfolio companies do not need an enterprise-scale AI program to begin. They need a disciplined way to select opportunities, test value, manage risk, and expand only what proves useful.

The standard should be practical:

• Does it solve a meaningful operating problem?
• Can the value be measured?
• Can the company support it?
• Are the risks proportionate to the benefit?
• Is there enough evidence to justify the next investment?

AI creates value when it becomes part of how the business performs, not when it remains an impressive demonstration.

The goal is not more pilots. It is better outcomes.