Your company has been running a pilot for eight months. Nobody can tell you when it becomes a real project. Nobody can tell you what would make it succeed or fail. Nobody can tell you who decided to start it, or who has the authority to end it.
This is how organizations avoid decisions while looking busy.
The pilot as decision avoidance
There is a specific moment in the life of every portfolio company where a difficult decision needs to be made. Invest in data infrastructure. Replace a legacy system. Commit to a new go-to-market motion. Fund the integration of an add-on acquisition properly.
The decision has a cost. It requires budget. It requires someone to own the outcome. It requires the organization to say “we are doing this” and be accountable for whether it works.
The pilot replaces all of that with something much more comfortable. “Let us run a proof of concept first.”
Nobody can argue with a pilot. It sounds prudent. It sounds data-driven. It sounds like the responsible thing to do before committing resources.
But in practice, the pilot does not replace the decision. It replaces the urgency to make one.
How pilots become permanent
The pattern plays out the same way every time.
A portfolio company identifies a need. The management team agrees the need is real. The operating partner agrees the initiative belongs in the value creation plan.
Instead of funding the initiative, someone suggests a proof of concept. Narrower scope. Smaller budget. Lower risk. “Let us validate the approach before we commit.”
The POC launches. A small team starts working. The quarterly update to the board says “promising results.” The operating partner notes the progress and moves on.
Six months in, the POC has not failed. It has not succeeded either. The results are ambiguous enough to justify continuing. The team asks for more time. The board agrees. “Extend through Q3.”
Nine months. Twelve months. Fourteen months. The status updates never change. “Promising. Need more time. Close to proving value.”
Nobody kills it because nobody owns the criteria for success or failure. The POC was launched without a decision framework. Without a defined question it was supposed to answer. Without a deadline after which the organization would commit or walk away.
The pilot did not replace the decision. It replaced the urgency to make one.
The math on a five-year hold
Fourteen months of proof of concept on a five-year hold is 23% of the runway spent not deciding. On a seven-year hold, it is still 17%.
And the cost is not just the time. It is the compounding benefit that was forfeited.
Companies that achieve operational improvements in the first 100 days post-acquisition tend to sustain those gains throughout the holding period. The investment made in year one compounds through every subsequent year. A 2% margin improvement in year one becomes four years of compounded benefit on a five-year hold.
A 14-month pilot that delays the same improvement until year two and a half cuts the compounding window in half. The margin improvement is the same. The total return is not.
One mid-market platform company I studied delayed value creation initiatives by six months because of fragmented data across add-on acquisitions. The cost was $8 million in forgone EBITDA. That was not a pilot. It was an unplanned delay. But the economic effect is identical. Time spent not executing is time the return is not compounding.
The expensive part is not the POC that fails. A POC that fails in 60 days and produces a clear answer is a good investment. The expensive part is the POC that “almost works” for fourteen months while the organization avoids the decision it was supposed to inform.
The AI proof of concept trap
The most common version of this trap in 2025 and 2026 is the AI proof of concept.
80% of PE/VC firms deployed AI by late 2024. Over 50% of mid-market portfolio companies have active AI initiatives. Global AI spending is forecast to exceed $2 trillion in 2026.
The pressure to “do AI” is enormous. And the easiest way to respond to that pressure is a proof of concept.
The pattern is specific and predictable. The AI vendor arrives with a demo. The demo runs on clean sample data. It works beautifully. The management team is impressed. The board is excited. The POC is funded.
Then the POC moves from sample data to the company’s actual data. The actual data has duplicates. Inconsistent hierarchies. Revenue figures that do not match finance. Customer records in the CRM that do not match the billing system. Product codes that were changed three times without updating historical records.
The model works perfectly. The outputs are useless. Not because the AI is bad. Because the data underneath it was never ready.
The company now has two choices. Fund the data infrastructure that would make the AI work at scale, which is the decision they were avoiding in the first place. Or extend the POC, narrow the scope further, find another clean subset of data, and produce another round of “promising results.”
Most choose the second option. The POC continues. The board deck reports “we have AI.” What they have is a demo that runs on a spreadsheet.
“AI proof of concept” has become the most expensive way to avoid a data strategy decision.
What a real pilot looks like
A pilot that works is defined by one thing. A specific question with a deadline.
“Can this tool reconcile our CRM to finance within 2% accuracy, using our actual data, in 30 days?” That is a pilot.
“Let us explore the potential for AI to improve our demand forecasting.” That is a stall.
The difference is the decision framework. The first version has a binary outcome. Either the tool hits 2% accuracy or it does not. Either way, the organization knows what to do next. Fund the rollout or try a different approach.
The second version has no exit criteria. There is always more to explore. There are always more promising results to investigate. The pilot extends because there is no definition of done.
A real pilot has five components.
A single question. Not a set of objectives. One question, stated clearly enough that anyone in the organization can repeat it.
A deadline. 30 days. 60 days. 90 at the absolute maximum. If the question cannot be answered in 90 days, it is not a pilot question. It is a project hiding behind pilot language.
A decision owner. One person who has the authority and the budget to act on the result. If the pilot succeeds, this person funds the rollout. If it fails, this person redirects the resources. If nobody has that authority, the pilot is theater.
Success criteria defined before launch. What does “yes” look like? What does “no” look like? Write it down. Get agreement. If the criteria are not written down before the pilot starts, they will be negotiated after it finishes, which means the pilot will never finish.
Actual data. Not clean sample data. Not a curated subset. The company’s real, messy, inconsistent operational data. If the pilot only works on clean data, you have not validated the approach. You have validated the demo.
The operating partner’s role
Operating partners can stop the permanent pilot pattern by asking one question at every quarterly review.
“What is the specific question this pilot is trying to answer?”
If the management team cannot say it in one sentence, it is not a pilot. If they cannot name the decision owner, it is not a pilot. If they cannot tell you what “done” looks like, it is not a pilot.
Do not ask for more status updates. Do not ask for a detailed timeline. Do not ask for a revised business case. Ask for the question. If the question does not exist, the pilot is a stall.
Then set the deadline. “You have 60 days to answer this question. On day 61, we decide.”
This is not micromanagement. It is discipline. The management team will often be relieved. They know the pilot is drifting. They know it is not producing real answers. What they need is organizational permission to stop exploring and start deciding.
Stop extending. Decide.
The portfolio companies that are pulling ahead in this environment are not the ones running the most sophisticated pilots. They are the ones making decisions.
Fund the data infrastructure or do not. Replace the legacy system or do not. Integrate the add-on properly or accept the consequences at exit. Commit to the AI initiative with proper data foundations or acknowledge that the data is not ready and focus there first.
Every month spent in a pilot that avoids these decisions is a month of compounding benefit forfeited. On a five-year hold, that is expensive. On a seven-year hold, with the exit market tightening and the quality premium compressing, it is the difference between a premium exit and one that sits in the backlog.
Stop extending. Decide.