The roadmap that reads well in a board deck and the one that creates value across the hold are two different documents. Which one a company ends up with is decided before a word of either is written.
Every middle-market portfolio company with an AI thesis has produced a roadmap. Far fewer can say what kind of roadmap it is, and that matters, because there are two kinds and only one of them creates value.
A technology roadmap is a sequence of tools, integrations, and infrastructure decisions, such as which platform to pilot, which data to clean first, and which workflows to automate. It is necessary, and it is the kind of roadmap an IT function knows how to produce. A business-value roadmap is a different document. It ties each decision to an outcome the company is held accountable for, whether that is margin expansion of a specific size in a specific function on a specific timeline, customer-acquisition-cost reduction in a named channel, or cycle time on a defined process. Every line maps to something the sponsor would recognize in the financials, and every line has an owner accountable for the outcome rather than only the activity.
Middle-market companies need the second kind. Left to their own processes, they tend to produce the first, handed to them by IT or a vendor, with the value-creation logic implied rather than stated. That gap is the audit problem, and it carries a direct talent consequence. The executive who can build a business-value roadmap in a resource-constrained company is a specific and rare profile, and most AI hires are made without ever testing for it.
Why the assessment, not the roadmap, is where value is won or lost
The roadmap problem actually originates one step earlier, in the readiness assessment. As the first article in this series described, no single executive is clearly accountable for AI in most middle-market companies, and those same companies often have little internal capacity to assess themselves. The CIO is frequently a director-level hire brought in for security and integration, the data-science function is often one analyst running reports for the CFO, and there is no benchmark to test the company’s view of itself against. So the assessment gets outsourced, most often to the AI vendor the company is already in late-stage talks with.
The structural trap is straightforward. A vendor’s assessment of a company’s AI readiness will reliably conclude that the company is ready to buy what the vendor sells. This is not bad faith. It is simply the shape of the relationship. The vendor is the only party in the room with the depth to run a credible assessment, and also the only party with a commercial stake in its conclusion. Without independent internal capacity to push back, the loop closes on itself, and the company builds its roadmap on a read of itself it never tested.
What the best operating partners do differently before the roadmap is built
The operating partner is the natural pressure point here, the same role that should force the ownership question. The ones who get this right tend to share three practices, and these are the most practical takeaways in this series.
- First, bring an independent perspective to the assessment itself. That might be an outside advisor, a peer portfolio company’s CIO acting informally, or a structured workshop designed to surface assumptions. The goal is not to override the company’s view of itself, only to make sure that view has been tested against something other than itself.
- Second, reframe the output in business-value terms. If the roadmap cannot state the EBITDA, customer, or operational outcome it targets in language the sponsor would recognize from any other value-creation conversation, then it is not finished. It is still only a draft.
- Third, treat the roadmap as a living document. The most common failure is a roadmap that was right at month six and wrong by month eighteen, because the business moved and the roadmap did not. A quarterly review against the value-creation thesis is a small discipline with outsized returns.
Bain’s research reinforces the distinction. The firms generating real returns from AI in portfolio companies are the ones applying it to strategic priorities rather than to general infrastructure modernization. In practice, the line between test-and-learn and value-creating is the same line that separates a technology roadmap from a business-value one.
How assessment and roadmap rigor should shift at entry, at scale, and at exit
At entry, the 100-day assessment matters more than the 100-day roadmap. A correct read grounds everything that follows, while a wrong read produces well-intentioned decisions on a foundation that will not hold. At scale, the reassessment matters as much as the original audit, since the team that was right in year one may be wrong in year three, and Bain’s data shows the gap between AI leaders and laggards widening fast. At pre-exit, the roadmap becomes part of the narrative buyers diligence, and they can usually tell within the first hour of management presentations whether it is a credible value-creation story or a list of pilots that never scaled.
Hiring the executive who can actually build the roadmap you need
All of which returns the question to the executive carrying it. The profile is neither a pure technologist nor a pure operator. It is someone who can hold both, work without a large team, and resist the pull toward vendor-led or IT-owned roadmap design. Testing for it is its own kind of assessment problem.
Has this candidate built a roadmap from scratch, or only managed one handed to them? Have they driven outcomes that showed up in the P&L, or overseen pilots that never moved? Have they pushed back on a vendor whose recommendation conflicted with the business, and if so, what did they do?
So the question worth holding is this. If the CEO handed the operating partner an AI roadmap tomorrow morning, would the operating partner know whether it was the right one? When the answer is uncertain, the work begins with the assessment, and often with a conversation about whether the executive who owns it is the one the thesis actually requires. That is a conversation we have with sponsors at Kingsley Gate early and often, well before a brief is written.
References
Bain & Company, Field Notes from the Generative AI Insurgency in Private Equity, in Global Private Equity Report 2025, March 2025. Survey of private investors representing $3.2 trillion in assets under management.
Bain & Company, Global Private Equity Report 2026, February 2026.
English 
Español de México