Most CEOs today will tell you they are serious about AI. They have approved budgets, launched pilots, stood up AI task forces, and explained to their boards how these activities will keep the business from falling behind. What most of them are actually doing, though, is managing an AI portfolio—a collection of experiments, proofs of concept, and incremental productivity tools—rather than leading an AI transformation. These are not the same thing, and the gap between them is widening fast.

There’s plenty of frustration. Bain’s most recent CEO survey finds that roughly 80% of CEOs are unhappy with the pace of their AI transformation programs. But the data tells a sharper story underneath: Around 85% of companies are not executing well. The pace frustration is real, but it is largely a symptom of how the programs are being run, not a reflection of how fast the technology allows them to move.

The emerging leaders are not simply moving faster on the same path; they are operating on a different logic, one that’s rooted in structural divergence. How? By building proprietary intelligence that puts them on a different curve entirely. They have made deliberate, board-level choices about where AI will change the economics of their business. They are rebuilding workflows from the ground up rather than layering AI onto existing processes. They are investing in data, an agentic software capability, and organizational learning as strategic assets. And they are doing all of this with a horizon that extends over years, not the next quarter.

A company builds proprietary intelligence when it combines three things no competitor can copy:

• Unique, proprietary data: the accumulated record of its customers, operations, and outcomes
• Encoded workflows: the institutional knowledge of how it actually operates, human-led and built into agents that act on that knowledge at scale
• Learning architecture: the human and AI feedback loops that make every deployment smarter than the last, compounding into a capability advantage that grows faster than any competitor can counter by writing a larger check

Proprietary data sharpens the agents, agents sharpen the people, the people redesign the work and encode the new workflows, and the new work generates better data.

Ramp, the comprehensive financial operations platform for businesses, shows what proprietary intelligence looks like on the ground. After reaching 99% adoption of AI tools across the company, the team realized most employees had plateaued—not because the models were inadequate, but because there was no shared infrastructure to connect tools, propagate workflows, or carry context across sessions. The team built Glass, an internal AI productivity layer where one person's breakthrough rapidly becomes the company's baseline and the system accumulates persistent memory across the organization. As the company’s leaders put it, internal productivity is a moat, and an organization does not hand its moat to a vendor.

This is one critical difference in the current strategic landscape. Prior technology waves rewarded patient adopters. Companies that waited several years to move to the cloud, to digitize, or to modernize their data platforms were able to catch up. The gaps built by early movers were not insurmountable with a bigger check, a better vendor, or a capable CIO.

AI is not behaving that way. Agentic AI is an entirely different class of software from the systems and tools that enterprises have operated before. Agents plan multi-step tasks, take action on systems through application programming interfaces (APIs), maintain state across long-running interactions, and operate with delegated authority on behalf of the business. The advantage compounds from Day 1 through reinforcing mechanisms that did not exist in the same way in prior technology waves.

Seven decisions to create proprietary intelligence

Seven decisions separate the companies creating proprietary intelligence and winning with AI.

1. Posture. Multi-year capital committed to a strategic position the CEO has determined and is narrating personally to the organization, not an in-year ROI test applied to every pilot.
2. Domain focus. Three to five concentrated bets where AI changes the economics of the business, not dozens of pilots spread across functions, none big enough to matter.
3. Data. Proprietary data and the semantic layer funded as the foundation of competitive advantage and built ahead of the agents, not retrofitted after the program stalls.
4. Technology architecture. Orchestration layer built and operated in-house, tied to proprietary data and workflows, not ceded to a single vendor’s platform.
5. Operating model. Workflows and workforce redesigned in parallel, not layered onto the way work is done today, with active CEO sponsorship enabling changes to job definitions, organization silo boundaries, and incentives that bottom-up programs work around.
6. Learning system. Architecture designed so each deployment makes the next one smarter, faster, and cheaper, with feedback loops, shared memory, evaluations, and social visibility built in from Day 1, not bolted on after launch.
7. Governance. A second governance motion designed to change the business, running parallel to the governance that runs the business, with one leader at the top of the organization personally accountable for AI risk.

Building the necessary capabilities for this kind of transformation may be harder than some CEOs realize. Two decades of offshoring and software-as-a-service (SaaS)–first thinking have left most large enterprises without the internal software development muscle that agentic AI demands. And you cannot buy your way to a proprietary enterprise orchestration layer. It must be built, and the capability to build it has atrophied across most of the Fortune 500.

But leading companies are doing just that. Bradesco's first agentic deployment illustrates what this looks like in practice. The bank's initial design relied on a few large, complex agents. Beta testing surfaced that the design was too slow, too costly, and unable to scale safely. What mattered was not that the failure happened, but that the team surfaced it early, was honest and clear-eyed about it, and rebuilt the architecture rather than force a brittle solution into production. The reset cost roughly five months and became the platform on which everything that followed could be built. The bank is now running customer-facing AI inside several core banking moments at a scale of 22 million customers. None of that could have been attempted without the muscle and the trust built through the earlier, lower-stakes work.

What leaders are doing differently

Operating on a different logic also requires different behaviors from the CEOs that are pulling ahead. Three traits show up consistently across every leader we have studied, and they are the traits that make the seven decisions hang together rather than fall apart.

The first is personal commitment. Not budget allocation, not a sponsored initiative, but the personal commitment from the CEO that this transformation is a strategic priority, visible in how they spend their time, what they hold their team accountable for, and the narrative and trust they build with the organization in their own voice about how the business and its people will thrive. The leaders are the storytellers in chief, not the sponsors of someone else’s story. Nothing demonstrates that commitment and curiosity like CEOs who are using AI to help run and change the business, forcing their executive teams to run to catch up.

The second is ruthless concentration. Leaders are not running more initiatives than their competitors; they are running fewer initiatives that are better resourced and more consequential in the two or three domains where AI changes the competitive economics of their business.

The third is investing for learning. The leaders treat every architectural and governance decision as a choice about whether their advantage will accumulate or evaporate. They invest in the unglamorous foundations—the semantic layer, the enterprise orchestration platform, the shared memory, the evaluations function—because they understand that the tenth agent should be better and cheaper than the third, and that only happens if the architecture is designed for it from the start.

Denis Machuel, CEO of Adecco, the global recruitment services company, has personally led an AI transformation as his first responsibility, investing his time in learning about the technology while visibly signaling to the organization that AI is core rather than peripheral. His central conviction is that AI has to happen with people, not to people. He replaced fear with role-specific training, codesigned the agentic recruitment process with recruiters so agents took high-volume tasks while humans kept judgment-based work, and framed it as a growth story rather than headcount cuts, building the trust and excitement to support the company’s ambition of having agentic AI power 50% of its revenue by the end of 2026.

The window is open

It is easy to say “yes” to the seven decisions above. Whether you have actually made them shows up elsewhere—in calendars, in budgets, in who sits on which committee, and in what gets reported to the board. Ask yourself:

• When you describe your AI program to the board, do you reach for the number of pilots underway, or for the learning on the most value critical workflow you have torn up?
• How many hours last week did you personally spend inside the tools your teams are using?
• Can your CFO point to the AI investment furthest from in-year ROI and explain why you are protecting it?
• When the last agent underperformed in beta, did the meeting end with a status update or with what was learned?
• If you asked your CFO, your head of sales, and your data team today what a "customer" means in your business, would they give the same answer?
• If your largest AI vendor doubled its prices tomorrow, how long would it take you to switch, and what would you lose in the process?
• If a competitor watched your program for 90 days, would they conclude you have placed real bets or that you are hedging?

The seven decisions are not best practices. They are the deliberate, often uncomfortable choices that separate the companies building proprietary intelligence from the ones generating activity. None of them can be delegated. All of them sit with the CEO. Are you answering them deliberately, or by default?

Go deeper

Turn Artificial Intelligence into Proprietary Intelligence

A CEO's guide to seven decisions that define AI transformation leaders.