Every board expects an AI roadmap now. But having one doesn’t mean having a strategy. MIT research found that 95% of enterprise generative AI pilots deliver zero measurable ROI, and the reason isn’t the technology. It’s that the initiatives were organized around what AI can do rather than what the business actually needs it to do.
Here’s a perspective that might be unpopular: there shouldn’t be a separate AI roadmap. If your team is presenting AI as its own initiative with its own milestones, separate from the business goals you’re already tracking, that’s a signal worth paying attention to. AI should show up inside the product strategy, the operating plan, the OKRs. Not a parallel track, but woven into how the company already runs.
The companies generating real returns from AI share a few patterns: they start with a business problem rather than a technology capability, they pick a small number of bets tied to advantages that are already hard to copy, and they treat this as a continuous evolution rather than a one-time project.
The AI Roadmap Everyone Has, and the Question It Doesn’t Answer
Every board expects an AI roadmap now. What’s worth examining is what’s actually on it, and whether having it as a separate document is itself part of the problem.
Most AI roadmaps I’ve seen are organized around AI capabilities: “we’re adding AI-powered recommendations to the platform, building generative content features for customers, using AI for customer support triage, automating sales forecasting.” That reads well in a board deck. The problem is that it doesn’t answer the only question that matters: what business outcome is each of those improving, and by how much?
According to MIT research, 95% of enterprise generative AI pilots deliver zero measurable ROI. That’s not because the technology doesn’t work. It’s because the initiatives were organized around what AI can do rather than what the business needs it to do. There’s a term that’s been floating around for this pattern: AI theater. Activity that looks like AI strategy but isn’t connected to business results.
Here’s a perspective that might be unpopular: there shouldn’t be a separate AI roadmap. If your team is presenting AI as its own initiative with its own milestones and success metrics that don’t connect back to the business goals you’re already tracking, that’s a signal worth paying attention to. AI should show up inside the product strategy, the operating plan, the OKRs and goals the business is already organized around. If it only exists as a separate document, it’s worth asking why.
Sometimes a dedicated AI view is useful early on, as a way to see the full picture and make sure investments are coordinated. That’s fine as a temporary lens. But if it stays permanently separate, it’s going to remain a parallel workstream rather than becoming part of how the company actually runs. That’s what “business first” actually means. Not AI for the sake of AI, but AI in service of the goals you already have.
Start with the Business Problem, Not the Technology
In my experience, the clearest path to real value starts with a business problem and works backward to the technology, not the other way around.
Here’s a test I find useful. Every AI initiative should be expressible in one sentence:
“We will use AI to improve [business metric] by [amount] in [timeframe], by leveraging [specific advantage we already have].”
For example:
-
“We will use AI to raise expansion NRR by 5 points over 12 months by surfacing better cross-sell opportunities from our proprietary usage data.” That 5-point improvement translates directly to top-line growth and a stronger retention story at exit.
-
“We will use AI to cut implementation time in half this year by automating the 80% of workflow configuration that’s predictable across our install base.” Faster implementations mean faster time-to-revenue and lower cost-to-serve.
-
“We will use AI to reduce support resolution time by 40% this quarter by training models on our ten years of ticket history.” Better resolution times protect retention and show up in the P&L.
If you can’t fill in that sentence, it’s not an initiative yet. It’s a hypothesis. And hypotheses belong in low-cost experiments, not in board presentations.
Fewer Bets, Tied to What’s Already Hard to Copy
Before going further, one thing worth noting: AI shows up in two places in most businesses. It shows up in your product, where it creates value for customers. And it shows up in your operations, where it creates efficiency for your team. Both matter, and the strongest moves are often where the two reinforce each other. But when you’re evaluating your AI investments, it helps to be clear which side of the equation each one serves.
According to McKinsey’s 2025 State of AI survey, the roughly 6% of companies qualifying as true AI high performers are three times more likely to use AI for genuine transformation rather than incremental improvement. They’re not running 30 pilots. They’re picking a small number of bets and going deep.
The strongest of those bets are anchored in the advantages that are already hard to replicate, the ones I described in the first article:
Your proprietary data. Ten years of customer behavior, transaction histories, and domain-specific edge cases will beat a technically elegant model with generic data almost every time. Your data makes your AI better, better AI creates more value for customers, and more value deepens the relationship and generates more data. That compounding cycle is the advantage that actually matters.
Your embedded workflows. AI designed into how customers already work with your product is fundamentally different from AI bolted on as a separate tool. And here’s the deeper point: those workflows will themselves evolve because of AI. The companies that are embedded in their customers’ operations are best positioned to lead that evolution, because they understand the workflow better than anyone else. That’s a front-row seat to redesign how the work gets done, alongside the customer.
Your customer relationships and go-to-market. AI that helps you understand what customers need before they ask, that surfaces expansion opportunities, that makes your support faster and more personalized: this is AI strengthening a relationship that took years to build. And it extends to your partner ecosystem and distribution channels. An AI clone has no distribution. It has no partners. It has no warm introductions.
The common thread: the best AI investments don’t create new advantages from scratch. They make existing advantages stronger.
Assistants Are a Starting Point, Not the Destination
One distinction worth paying attention to when you’re evaluating your organization’s AI work: there’s a meaningful difference between adding AI assistants on top of existing processes and fundamentally rethinking how work gets done.
Most organizations start with the assistant approach: co-pilots that help people draft faster, summarize documents, generate first cuts of analysis. Those are real productivity gains, and they’re a reasonable place to start. But if that’s where your AI investments plateau, you’re capturing a fraction of the value.
The real gains come when the team steps back and asks: if we could redesign this workflow from scratch knowing what AI can do today, what would it look like? Not “how do we make the existing process 20% faster” but “what would this look like if it were built around agentic capabilities from the start?”
That’s a harder question. It’s worth asking your leadership team directly: which category do most of our AI investments fall into? Adding tools to existing processes, or rethinking how the work gets done? The answer tells you a lot about whether you’re getting incremental improvement or real change.
And this connects back to something I said in the first article: your people are the ones best positioned to answer that question. They know which workflows are painful, which data is valuable, which customer problems keep coming up. If the redesign conversation is happening only in the technology team, it’s missing the people who understand the business best.
This Is Not a One-Time Project
The pressure to “do an AI transformation” is intense right now, and I understand the instinct to scope it as a project: define the initiatives, allocate the budget, execute, report to the board.
But as I wrote in the first article, AI is not a transformation with a finish line. The technology will keep changing. What’s possible in eighteen months will look different from what’s possible today. A roadmap that makes sense right now might need to evolve significantly before you’re halfway through it.
The right approach is phased rather than big bang: start focused, learn, prove value through delivery, then expand. Some cycles take several months, not years, especially when focused on a few specific use cases. But the key question for leadership isn’t about the specific phasing. It’s whether the organization is treating this as a project with a defined end date, or building an ongoing capability to keep evolving. One of those mindsets leads to a box that gets checked. The other leads to a company that keeps getting better.
I’ve been through this pattern many times, not just with AI. Integrating acquired companies, rebuilding operating models, introducing new ways of working across global teams: the hardest part is never the technology or the framework. It’s getting the organization aligned on which problems are worth solving and which are distractions dressed up as progress. That alignment is a leadership problem, not a technology problem. And it’s ongoing, not something you solve once.
The Honest Version
Nobody has this fully figured out. Not the companies I advise, not the companies I assess in diligence, not me. We’re all navigating a landscape that’s changing faster than anyone’s ability to plan for it.
But the companies that will look back on this period and feel good about their choices will be the ones who stayed focused on business outcomes rather than getting swept up in AI for its own sake. The ones who picked a few things that really mattered rather than spreading thin across dozens of pilots. The ones who asked their own people where AI would actually help. And the ones who built the organizational discipline to keep evolving rather than treating this as a project with a finish line.
The technology is moving fast. Your job isn’t to match its speed. It’s to make sure that when you do move, you’re moving in a direction that makes your business stronger, not just more AI-decorated.
In the next piece, I’ll dig into the people and organizational side of this: why AI transformation is ultimately a people problem, and what I’ve seen work in getting organizations through change like this.
Sources
- MIT Project NANDA. The GenAI Divide: State of AI in Business 2025.
- McKinsey & Company. The State of AI 2025: Agents, Innovation, and Transformation.