Why your AI roadmap feels disconnected from execution

You know you need to be doing more with AI. You've known for over a year. The honest answer for why you haven't is that nobody was ever handed the job of making it happen.

You're not the exception. RSM's 2025 Middle Market AI Survey found the same split holds across the board: a roadmap in place almost everywhere, real integration almost nowhere. The gap isn't the plan. It's everything that was supposed to happen to it after the room where it was written emptied out.

When researchers behind MIT Project NANDA's The GenAI Divide: State of AI in Business 2025 set out to measure how much real structural change AI has actually caused across industries (not adoption, not pilots, but genuine change to how a business operates), they found that despite $30-40 billion in enterprise investment, 95% of organisations are getting zero measurable return. One mid-market manufacturing COO summed up what they heard from most people they interviewed: "The hype on LinkedIn says everything has changed, but in our operations, nothing fundamental has shifted. We're processing some contracts faster, but that's all that has changed." There was a workshop, a roadmap, and a leadership sign-off. And on the ground, next to nothing moved.

Up Strategy Lab has written before about how AI exposes the gap between strategy and execution at the organisational level. Our foundational thesis is simple: organisations that succeed with any kind of digital transformation exercise, including AI, share a common trait: they do not start with tools; they start with making the organisation legible to itself.

This is what that gap looks like inside any roadmap. A roadmap describes a future state. It rarely says who owns which decision on a Tuesday, what a specific team does differently as a result, or how anyone would know if it worked. When that ownership is missing, AI initiatives don't collapse dramatically. They just quietly stop being anyone's job.

It is tempting to read this as a technology failure, but that same MIT report puts it directly: "The dominant barrier to crossing the GenAI Divide is not integration or budget, it is organizational design." Their research, based on interviews with 52 organisations and a review of over 300 public AI deployments, found that companies which keep initiatives centralised, run by a strategy team or steering committee with no one closer to the work actually holding it, are the ones that stall, while those that decentralise implementation authority to the people who own the work, keeping them accountable for outcomes, are the ones that succeed.

Why AI strategy isn't the problem

MIT isn't alone in finding this. McKinsey's most recent State of AI survey, fielded across nearly 2,000 organisations in mid-2025, found that 88% of companies now use AI regularly in at least one business function, yet only a third have moved past piloting to actually scaling it, and just 6% qualify as "AI high performers," attributing more than 5% of enterprise EBIT to their AI use. What separates that 6%? McKinsey's own research identifies workflow redesign and visible senior-leader ownership as two of the strongest factors behind that gap: high performers are nearly three times more likely to have fundamentally redesigned their workflows, and three times more likely to say senior leaders demonstrate real ownership of their AI initiatives.

Boston Consulting Group's 2025 AI Radar survey of over 1,800 executives puts a number on the same imbalance: across the factors that actually separate AI winners from the rest, roughly 70% of the effort sits in people, process and culture, 20% in data and technology, and just 10% in the algorithms themselves.

MIT and McKinsey aren't measuring the same thing. MIT is measuring who holds authority once a project is live. McKinsey is measuring whether the underlying workflow actually gets rebuilt. BCG is measuring where effort gets spent overall. Three different lenses, landing on the same uncomfortable answer: none of it is really about the technology.

None of this is actually new, and it isn't unique to AI. Back in 2018, long before generative AI gave anyone a reason to call a workshop, Up Strategy Lab co-founder Noel Braganza made a version of the same argument about an earlier wave of digital transformation: "Don't end up with great presentations and concepts that will die in a powerpoint somewhere that no one will read."

Up Strategy Lab made the same case again, more recently, in a piece on digital transformation, where we cited a separate McKinsey survey of 700 executives that found the average company has close to two CxO-level "digital leader" titles, sometimes six or more, and a third of respondents couldn't say which one actually owned the work. It's a different flavour of the same problem MIT found at the workflow level: ownership blurs at every altitude when no one is forced to be specific about it. AI just gave the syndrome a new name.

Why the next shiny AI pilot won't fix it either

There's a temptation, once a roadmap stalls, to chase the next wave: Agentic AI, autonomous workflows, whatever the next conference keynote is selling. Gartner's research suggests caution here too: it predicts that more than 40% of agentic AI projects will be cancelled before the end of 2027, driven by escalating costs, unclear business value, and inadequate risk controls. Anushree Verma, Senior Director Analyst at Gartner, describes most current agentic AI projects as "early stage experiments or proof of concepts that are mostly driven by hype and are often misapplied." Misapplied is the operative word: without someone already accountable for a specific workflow, there's no way to judge whether a new tool actually fits a problem worth solving. New technology does not fix a missing ownership structure. It just gives the same disconnect a newer name.

What closing the AI execution gap actually looks like

AI roadmaps aren't pointless: the mistake is treating the roadmap itself as the deliverable, rather than the starting point for reassigning ownership and rebuilding a handful of specific workflows around it. That's what organisational legibility actually means in practice: not a strategy document, but an organisation that can say, plainly, who owns what.

Doing that well isn't a planning exercise. It's an operational one, and it takes a team that has actually taken an AI-powered product into production and lived with the consequences of keeping it working, not one that has only advised on how ownership should look on paper. That's true of MuchSkills, the AI-powered platform Up Strategy Lab built and still owns. Building it meant finding the real bottleneck first, in that case, organisations that couldn't see what their own people were capable of, and then making sure the fix actually got used, not just shipped.

When this discipline is applied to AI, it becomes an audit that identifies two or three specific points where AI creates real operational leverage, followed by a working prototype embedded in one real process within weeks, not a document handed over at the end of an engagement.

The approach also happens to suit mid-market businesses better than it suits large enterprises. MIT's research found the top-performing mid-market companies moved from pilot to full implementation in 90 days on average, where large enterprises, despite piloting more and staffing AI initiatives more heavily, took nine months or longer. Our read: fewer layers between a decision and the person who has to live with it. Being smaller isn't just faster. It's fewer places for ownership to get lost.

Most AI roadmaps fail for the same mundane reason: organisations stop treating them as active work the moment the workshop ends, and nobody was ever assigned to keep it moving.

If your roadmap is stalled in exactly this spot, this is the audit we'd run: Explore AI Transformation.

Frequently asked questions

Why does my AI roadmap feel disconnected from execution? 

Most commonly because ownership was never assigned at the workflow level. A roadmap describes a future state, but unless a specific person or team owns translating that into a changed day-to-day process, nothing changes operationally, no matter how sound the strategy is.

What causes AI initiatives to stall after the strategy phase? 

Research from MIT's Project NANDA points to organisational design rather than technology as the primary cause: initiatives that stay centralised with a strategy function, with no one closer to the work actually holding it, tend to stall, while those that hand ownership to the frontline managers closest to the work tend to scale.

How do you know if your AI strategy is actually working versus just producing pilots? 

Look for workflow redesign, not just tool adoption. McKinsey's research found that organisations getting real business value from AI are far more likely to have fundamentally redesigned specific workflows around it, rather than simply layering AI tools on top of unchanged processes.

Is it better to build AI tools internally or buy from a vendor? 

The real distinction isn't shrink-wrapped software versus in-house engineering: it's who stays accountable for the result after launch. External partnerships that co-develop something built around a company's own workflows consistently outperform teams building and maintaining everything internally and alone, the same ownership pattern that separates initiatives that stall from ones that scale. There's a separate failure mode worth naming too: generic, off-the-shelf tools like ChatGPT or Copilot show up everywhere in the same research, but for a different reason. They work well for individual productivity, and employees use them constantly, but they consistently fail at mission-critical, workflow-specific tasks because they don't retain context or adapt to how a particular business actually operates. This is also why the audit-and-prototype approach described above works: it's co-development in miniature, run by a team that has to keep the result working, not a specification handed to a vendor.