You're in a board meeting. The portco CEO just walked through the AI roadmap: a forecasting tool, a customer segmentation model, an automated operations layer. The budget request is $800,000. The timeline is 12 months. The deck is polished.
You're the operating partner. You have roughly forty minutes to ask questions that matter.
Here's the one I'd start with: Can you show me where the data for that AI system lives right now?
It's not "do you have a data strategy" or "how would you describe your AI maturity." The specific, operational question is: where does the data live, how current is it, and who owns it?
That answer tells you nearly everything about whether to fund this initiative, or spend the first 90 days on the foundation instead.
Why Most Portco AI Projects Fail Before They Ship
The numbers here are not encouraging. Global AI spending is on track to hit $2.59 trillion in 2026, according to Gartner. Yet only 15% of AI decision-makers report measurable EBITDA lift from their investments, according to Forrester. Gartner projects that over 40% of agentic AI projects will be canceled by end of 2027, not because the models are bad, but because organizations cannot operationalize what they build. IDC, in research conducted with Lenovo, finds that 88% of AI POCs fail to reach production, with insufficient AI-ready data, unclear ROI, and lack of in-house AI expertise named as the primary causes.
The pattern I see most often in PE-backed companies runs like this. A team identifies a high-value use case. They build a prototype in two weeks. The demo is impressive. Leadership greenlights a production rollout. Then reality arrives: the AI system needs data from four source systems that were never designed to talk to each other, real-time feeds that currently run as overnight batch processes, and a governance layer that does not exist. The team spends six months building integration plumbing. The project burns $500,000 in engineering time. Somewhere around Q4, it gets deprioritized.
That is not an AI problem. That is a data infrastructure problem with an AI budget attached to it.
The operating partner who catches this early saves the firm significant capital and twelve months of runway. The one who does not finds out at the next board meeting, when the initiative shows up in the challenges column.
The One Question That Functions as a Go/No-Go Gate
This is not a framework. It is a single diagnostic moment that most operators skip.
When a portco CEO presents an AI initiative, ask to see the data it depends on. Not in a deck. In the actual system.
First: is the data being generated by operational systems already, or does capturing it require new instrumentation? A portco that wants to build AI-powered demand forecasting but currently logs sales data in spreadsheets is not ready to build an AI model. They are ready to build a data consolidation pipeline first.
Second: how often does the data update? The AI systems getting serious attention right now (the ones that automate decisions rather than summarize reports) require data that reflects current reality. A system making inventory or pricing decisions on data that is 18 hours old is not autonomous. It is confidently wrong, on a delay.
Third: who owns the data? If the CEO has to ask three people to answer that question, the data has no clear owner. AI systems break at ownership boundaries. When a field stops getting updated, when a schema changes without notice, when two systems disagree on the same number, the AI system either fails silently or halts loudly. Either way, trust in the system collapses.
If any of those three answers comes back as "we are not sure" or "it is complicated," that is your signal: fix the foundation first.
What a Green Light Looks Like
The portcos that are genuinely positioned to ship AI-powered improvements share a few characteristics. None require a sophisticated analytics team. They require operational discipline.
The data the AI initiative depends on is already being captured systematically. It is not a project to collect it; it is a project to use it. Operational systems generate events, those events write to a database, and someone is responsible for that database.
Data flows between systems without manual intervention. The CRM talks to the ERP. The ERP talks to the financial system. The latency may still be too high for advanced use cases, but the connective tissue exists.
There is a person, or a small team, who can answer the question "is this data correct?" and then actually check. Not a theoretical owner. A person who opens a dashboard or writes a query.
When you see those three things, the AI initiative is building on ground that will hold. You can make a budget decision with confidence that the infrastructure underneath it is real.
What a No-Go Looks Like, and What to Do Instead
Most portcos I work with are not in the green-light position. That is not a failure. It is where middle-market companies typically are. They have been optimizing for revenue and operations, not for data infrastructure. When PE acquires them, there is often a gap between the data sophistication the AI roadmap requires and what actually exists.
The 2026 State of Data Engineering Survey (Practical Data Community, Joe Reis, February 2026), with more than 1,100 practitioner respondents, found that 89% of practitioners report significant data modeling challenges, and only 11% say data modeling is going well. When practitioners named their single biggest organizational bottleneck, legacy systems and technical debt ranked first at 25%, followed by lack of clear leadership direction at 21% and poor requirements at 19%. A separate question in the same survey found that 26% of teams report reactive firefighting consumes a significant share of their time. The combined weight of leadership gaps, unclear requirements, and organizational friction dwarfs any single technical barrier. The number one blocker to scaling AI in most companies is not the tech stack. It is the org structure and ownership around the data.
The right move in a no-go situation is not to cancel the AI initiative. It is to sequence it correctly.
The first 90 days look like this. Map the data the proposed AI system would actually need: not all the data, but the specific tables, fields, and systems the initiative depends on. This is a short exercise measured in weeks, not months. Assess the current state of each input: does it exist, is it machine-readable, how current is it, who owns it. Identify where the gaps cluster. In most portcos, the problem concentrates in one or two places: data that is not being captured at all, data that lives in silos with no integration layer, or data that exists but has quality problems that would corrupt model outputs.
Build the integration plumbing before the AI layer. A company that spends 90 days getting its core operational data into a reliable, structured pipeline, and then builds an AI system on top of it, will outperform one that tries to skip the foundation and ends up rebuilding the AI later. A portco that sequences this correctly is typically six to nine months faster to a working production system than one that does not.
The Portfolio Leverage Argument
One advantage PE operators have that individual companies do not is portfolio-level pattern recognition. Portcos that build the data foundation in the first year of the hold period are significantly easier to exit. Not because they have more AI features. Because their data tells a coherent story.
Acquirers and strategic buyers are running operational analyses that are more rigorous than they were five years ago. A portco with well-structured operational data, clear data lineage, and working analytics pipelines is faster to underwrite. That compresses exit timelines and reduces uncertainty in the buyer's model.
The portco that spent $800,000 on an AI system built on a shaky foundation, and then another $400,000 retrofitting the data layer when it broke, has a different story to tell. And a messier data room.
Before You Fund the Initiative
The go/no-go moment I am describing does not require a lengthy process. It requires asking the right operational question and knowing what answer you are looking for.
If the data is there, structured, owned, and flowing: fund the initiative.
If it is not: spend the first 90 days on the foundation. It is a faster path to the same destination.
If you want a structured way to run that diagnostic before committing the budget, the Blueprint engagement is designed to do exactly that. In a few weeks, you get a clear picture of what is ready and what needs to come first. That is a much cheaper decision point than finding out after committing the full $800,000.