The tools are deployed. Copilot licences are seated, an internal chat model has been running for a year, and three or four departmental pilots have moved from proof-of-concept to whatever comes next. The AI program budget has held steady, or grown. And yet the operating numbers look almost identical to the year before the investment began: cycle times, cost per case, throughput, decision latency. The board paper on AI progress reports adoption metrics because there are no operating metrics to report.
In Brief
- About 5% of enterprises are producing measurable AI value. 60% deployed the tools and are seeing nothing land.
- What separates the tiers is whether the work was redesigned before the AI arrived. Not budget, not model choice, not deployment volume.
- Governance depth (accountability inside a few redesigned processes) outperforms governance volume (the accumulation of estate-wide policy).
- The top 5% run fewer AI initiatives, redesigned more deeply, with a named executive accountable for each operating outcome.
- One test identifies the tier: name the process whose economics are now measurably different, and the executive accountable for it.
This is not a small-sample problem. When the Boston Consulting Group (BCG) examined AI value creation across roughly 1,250 senior executives and AI decision makers in 2025, only about 5% were generating measurable financial value from their AI investment. Around 35% were seeing some benefit, most of it partial and localised. The remaining 60% had deployed the tools and were producing nothing that showed up in the P&L (BCG, 2025). The distribution holds across sectors and across budget bands. Spending more does not move an organisation up the tiers, and deploying more tools does not either.
Only the 5% see benefits
The three-tier distribution silences much of the noise around “AI leaders” into something structurally clearer. Tier 1, the 5%, is generating value that reads on the operating statements. Tier 2, the 35%, has captured local efficiency gains, most of them at the individual-user level, that have not aggregated into anything visible above the team. Tier 3, the 60%, has bought and deployed, and is now waiting for something to happen. What separates Tier 1 from the rest is not the model being used, a specific software suite, or the vendor. What separates Tier 1 organisations is the redesign of their operating models to take advantage of what AI can achieve, not automating human processes or injecting observability software on top of AI.
Table: Three tiers of AI integration depth (adapted from Boston Consulting Group, 2025a)
| Tier 1 | Tier 2 | Tier 3 | |
| Reactive tool | Productivity accelerator | Integrated capability | |
| Distribution | ~60% | ~35% | ~5% |
| AI Role | Human-initiated, single-step interaction. Full human control at every step. | AI accelerates existing workflows. Underlying process design unchanged. Visible but limited. | Agents execute multi-step workflows autonomously within designed governance boundaries — processes redesigned around AI capability. |
| Governance model | None designed. Full human control at every step substitutes for it. | Controls added to the existing operating model — policy, access rules, review gates. The governance is designed; the model it sits in is not. | Governance and operating model designed together. Decision boundaries revised each cycle from what the cycle reveals. |
| Enterprise returns | Individual efficiency gains (faster drafting, summarisation, research) — no enterprise-level capability advantage. | Ceiling equals the existing process's ceiling. | Returns grow with each operating cycle. Governance capability accumulates. |
A concrete example makes the distinction visible. A large financial services firm in one of the Tier 1 cohorts had spent eighteen months rebuilding its commercial credit decisioning process before the AI investment was approved. The redesign shortened the sequence from seven approval stages to three. It moved certain risk categories from committee review to policy-based auto-decisioning. It restructured the credit officer’s role around exception handling rather than case review. And it rewrote the delegation instrument so that outcomes at each stage were traceable to a single accountable executive. AI then went in on top of the new process: first extracting document data, then drafting risk narratives for exception cases, then, later, running the auto-decisioning policy. The economics moved. Cycle time fell by a factor the Chief Financial Officer (CFO) could point to in the results call. But the AI did not produce that outcome. The redesigned work produced it. The AI just made the redesigned work faster.
The Tier 3 version of the same organisation would have layered the same AI tooling on top of the original seven-stage process. Each stage would still exist, each committee would still convene, each hand-off would still happen. AI would draft faster in some places and summarise faster in others, and the operating envelope would remain what it always was, because the envelope is set by the process, not by the speed of any step inside it.
Governance depth beats governance volume
The instinct in most Tier 3 organisations, once the productivity gap is visible, is to invest more in governance. Not more redesign, more policy. A responsible AI framework is drafted. Model risk protocols are added. An AI ethics committee stands up. Procurement gates are strengthened. Usage guidelines are circulated to staff. Each of these documents is defensible on its own; collectively they produce the appearance of serious governance and the reality of governance volume.
Governance depth is a different thing. It is the presence of a small number of clear accountabilities attached to specific redesigned processes: the credit officer’s new role, the exception threshold, the delegation instrument, the outcome measure. A CIO inside a Tier 1 organisation can tell you which three processes have been redesigned, which executive owns the operating outcome of each, and how the AI is contributing to that outcome. A CIO inside a Tier 3 organisation can tell you how many policies have been published, how many staff have completed AI use training, and how many pilots are currently live. Both organisations describe themselves as governing AI seriously. Only one is doing it in a way that produces value.
The pattern that tends to surface across organisations working through this transition is that governance volume grows in inverse proportion to redesign work. When the redesign is not happening, the policy stack thickens as a substitute, because policy is easier to produce, easier to approve, and easier to point to as evidence of executive attention. Redesigning the work is harder. It requires deciding which processes are worth changing, defending those choices against every other candidate, and then absorbing the operational disruption of the change. Most enterprises have a mature muscle for the first activity and a weak one for the second, so the policy stack thickens while the work stays where it was.
What the top 5% did before the AI arrived
The Tier 1 pattern, viewed across the redesign work that preceded the value gains, is unusually consistent. Fewer AI initiatives, not more: typically three to five processes selected for deep redesign, rather than twenty or thirty pilots seeded across the estate. Redesign that changed the operating envelope of the process before AI was layered on, rather than automation of the current envelope. Executive accountability attached to the operating outcome of each redesigned process, not to the AI initiative that supported it. And governance depth measured by the concentration of accountability inside a small number of processes, rather than by the breadth of policy across the whole organisation.
The consequence, over the two years that follow, is a widening gap between Tier 1 and everyone else. Tier 3 organisations continue to publish new policy, license new tools, run new pilots, and produce no operating change. Tier 1 organisations extend the redesign pattern to a second and third wave of processes, each wave producing measurable operating movement. By the third year, the difference in the P&L is a decision. The Tier 3 board is now allocating capital to catch up, on terms less favourable than the ones that would have been available two years earlier.
What this means for senior leaders
The AI decision that matters is not which tools to deploy or how much to spend. It is a matter of which processes are worth redesigning to the point where AI can produce operating benefits beyond what those processes deliver, and which executive is accountable for the operating outcome of each. Everything else in the AI estate is scaffolding around those decisions.
One test to run this week
Choose the AI initiative in your organisation that is presented as most successful. Ask two questions of the executive who owns it. First, which specific process has been redesigned as part of this work, and how do the economics of that process look different now than they did before the AI arrived? Second, who is accountable for the operating outcome of that redesigned process, and where does that accountability sit in the delegation instrument?
If the answer to the first names a redesigned process with different economics, and the answer to the second names a single accountable executive whose delegation reflects the outcome, the program is in Tier 1. If the first answer describes an AI tool that has been deployed and the second names an AI governance committee, the program is in Tier 3. The 5% and the 60% are separated by those two answers.
References
- Boston Consulting Group. (2025). Are you generating value from AI? The widening gap. BCG.
Boston Consulting Group. (2025a). The widening AI value gap: Build for the future 2025. https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap
