Cisco's 2025 AI Readiness Index puts the gap between ambition and execution in one number. Across 8,000+ senior business leaders surveyed in 30 markets and 26 industries, 13% of organisations qualify as "Pacesetters" — fully ready to extract value from AI. That share has stayed remarkably stable across three years of the same survey. Pacesetters report gains in profitability, productivity, and innovation at roughly 90% — versus around 60% for everyone else.
The interesting story for technology leaders is not the size of the gap. It's that we now have unusually good research on what closes it. Prosci's 2025 AI adoption study of 1,107 participants — 525 frontline employees, 393 people leaders, 193 executives — singled out specific executive behaviours that differentiated successful adoption from stalled pilots. IBM's May 6, 2025 study of 2,000 CEOs put numbers on the data and architecture problems sitting underneath. Cutter's CEO Insights 2025 is direct about why most enterprises are stuck in pilot purgatory while a smaller group is already in production. MIT Sloan's emerging-agentic-enterprise framing draws the same conclusion from a different angle. Read together, the picture is consistent.
The four behaviours that move the needle
Prosci's framework is unusually concrete for change-management research. Executive impact on AI adoption sorts into four areas — what they call the four V's: Visibility, Vision, Voice, and Value. Across Prosci's data, the difference between stalled pilots and sustained impact consistently comes back to executive behaviour: executives who drive successful adoption show up visibly, communicate with clarity, and anchor AI efforts to business outcomes.
The study also exposed a structural trust gap. Frontline workers reported very low trust in AI (around +0.33 on a –2 to +2 scale), while executives were significantly higher (around +1.09). That delta is what slows adoption — workers don't apply tools they don't trust, and executive enthusiasm doesn't survive the gap.
In plain terms, the four executive behaviours that compound:
Visibility. Executives are visibly using the tools themselves — model interactions, agent outputs, dashboards. Not in a quarterly demo. In their actual working day. The behavioural signal to the rest of the organisation is "this is real, I trust it enough to put my time in it."
Vision. AI is positioned as an enterprise change blueprint, owned by the executive, with a roadmap that ties to core drivers — customer experience, operating margin, competitive resilience — not as a technical backlog owned by IT.
Voice. Executives talk about AI internally with the same regularity they talk about quarterly results. Pilots that succeed have CEOs naming specific outcomes; pilots that stall have CEOs delegating the conversation entirely to a chief AI officer who is then disempowered.
Value. Initiatives are scoped against measurable outcomes (cycle time, defect rate, conversion lift) rather than activity (pilots completed, models trained). When the executive is the one defining the value metric, the initiative inherits accountability.
The teams shipping autonomous AI systems into production — sales agents, support agents, research agents, internal-ops agents — are almost without exception working inside an organisation where these four behaviours are visible. The teams running stalled pilots usually have one or two of them at most.
Data architecture is the bottleneck
The technology bottleneck most CEOs name is wrong. It isn't the model. It's the data layer.
IBM's 2025 CEO study, which surveyed 2,000 CEOs across 33 countries and 24 industries, put the data-layer story in numbers any boardroom can use:
- 68% of CEOs identify integrated enterprise-wide data architecture as critical to unlock generative-AI value.
- 61% are actively adopting AI agents today and preparing for scaled implementation.
- 50% acknowledge their organisation has "disconnected, piecemeal technology" due to a rapid investment pace.
That third number is the killer. Half of enterprises are buying frontier models, agent platforms, and copilots while their underlying systems remain fragmented — which means even when the model is right, the agent is operating on partial context. The investment goes into the visible model layer; the failure modes come from the invisible data layer.
Cloudera and HBR Analytic Services put a complementary stat next to it in March 2026: only 7% of enterprises say their data is completely ready for AI. Whatever framing you prefer — Cisco's 13% Pacesetters, IBM's 50% reporting disconnected technology, Cloudera's 7% data-ready — the picture is the same. Most enterprise AI failure isn't a model failure; it's a substrate failure.
This is the part of enterprise AI that maps most directly to what we ship at Axccelerate. Every operational AI workflow automation build runs into the same rate-limiting step — clean data plumbing has to come first. The CEOs whose AI initiatives compound are the ones who personally backed the integrated-data programme a year or two before they started buying agent platforms. The ones who skipped that step are running pilots that look impressive in a deck and then quietly fail when they meet real workflow.
Pilots are not the strategy
Cutter's CEO Insights 2025 frames the stall directly. Enterprise engineering organisations have hundreds of AI use cases in active development at any given time, but the gap between cases-in-development and cases-in-production is enormous, and most large companies are in what Cutter calls a "proof-of-concept trap." A pilot succeeds in isolation; the executive sponsor takes it as evidence the approach works; the investment splits across more pilots; nothing crosses the production threshold.
The shift Cutter advocates is from pilot orchestration to process redesign — replacing entire workflows, not augmenting individual tasks. The first wave was AI-assisted productivity (copilots in the IDE, chat in the docs app). The second wave was specific automated tasks (summarisation, classification, retrieval). The third wave — where Pacesetters live — is process redesign: taking a multi-step workflow that previously needed seven hand-offs and rebuilding it as one agentic loop with humans only at decision gates.
The startups that outpace incumbents on AI rollout aren't winning on raw model quality. They're winning because they were built around process redesign from day one — no legacy process to retrofit, no decade-old data warehouse to integrate against, no internal politics around which department owns the redesigned workflow. Incumbents that match them are doing it by treating AI not as a feature roll-out but as a workflow rewrite, with the CEO sponsoring the redesign rather than the model selection.
MIT Sloan's emerging-agentic-enterprise research lands in the same place from a different angle. Agentic AI is spreading faster than process redesign can absorb. Without intentional redesign, agents are deployed inside legacy workflows — and the result is a faster version of the same flawed process, not a better outcome.
The CEO as systems architect, not pilot sponsor
Here's the role the research consistently points to. The CEOs whose organisations get AI working at scale aren't the ones with the largest AI budget, the most credentialled chief AI officer, or the most pilots. They're the ones who treat AI as work transformation — investing in people, data, and process redesign alongside the technology, with executive visibility over all four.
Three concrete moves separate the Pacesetters from everyone else over the next four quarters.
Take ownership of the data architecture decision. This is not an IT call. The Pacesetters in Cisco's index are universally the ones whose CEOs personally backed an integrated data programme one to two years before they started buying agent platforms. If you're considering an agent rollout against a fragmented data estate, the agent will fail — not because the model is wrong, but because the agent is reasoning over an incomplete picture of your business.
Pick one workflow and rebuild it end-to-end. Not "add AI to" the workflow. Replace the workflow with one that assumes agents are part of it. Lead generation, support triage, financial close, employee onboarding, customer research — pick one process where you have clean signal and rebuild it from the agent outward. Treat the rest of the AI portfolio as productivity gains around the edges, but pour strategic attention into the redesigned case.
Make AI use part of how you operate, not how you communicate. The visibility behaviour Prosci flags is the cheapest of the four V's to get right and the one most often skipped. CEOs who model active tool use — including in front of their executive team — accelerate cultural adoption far more than CEOs who only talk about AI in earnings calls and town halls.
The 13% Pacesetters didn't get there by accident. They got there because someone at the top of the organisation decided AI was a substrate change rather than a feature line item — and treated data, leadership behaviour, and process redesign as the actual unlock, with the model layer as the cherry on top. The remaining 87% have, on the available evidence, somewhere between two and four quarters to make the same decision before the gap becomes structural.
