Only 4% of 767 U.S. operations and supply chain leaders report success on all four conditions PwC says predict AI value capture — AI fully embedded enterprise-wide, autonomous agents scaling without significant barriers, a horizontal operating structure in place, and technology investments delivering as promised. 89% say their tech investments have not delivered. 87% name data quality as the binding constraint. And only 41% are running the horizontal operating model that 94% say they need (PwC, April 23, 2026). The 96% are not failing because their AI is worse. They are failing because they are sequencing what the 4% are co-developing.
For a 200-FTE Head of Operations finalising the rest of the 2026 AI roadmap, that is the operational inversion that should reshape this quarter's plan. Run AI deployment, horizontal restructuring, data hygiene, and accountability redesign as one AI operating model program — or join the 96% that PwC's own data show cannot extract value from any of them in isolation.
The 4% Are Not Better at AI. They Are Better at Sequencing.
PwC's 2026 Digital Trends in Operations Survey, released April 23 and covering 767 U.S. operations and supply chain executives, defines the "4%" not by AI maturity in isolation but by simultaneous achievement across four workstreams that most mid-market ops functions treat as separate quarters of work (DC Velocity, April 2026). Each condition on its own is unremarkable. The variance is in whether they run in parallel.
Look at the gap between aspiration and execution. 94% of respondents say a horizontal, collaborative operating model is required to capture AI value at scale. Only 41% have one. That 53-point gap is not an awareness problem — every CHRO and COO in the survey knows the model needs to change. It is a sequencing problem. The 96% are saying: we will fix the operating model after the AI pilots show value. The 4% are saying: the AI pilots will not show value until the operating model is fixed.
The data backs the second posture. 89% admit current technology investments have not delivered expected returns. 87% cite poor data quality as the value blocker (PwC, April 23, 2026). These two numbers are not independent. Tech investments do not deliver because the data they feed on is poor. The data is poor because the operating model has no single accountable owner for it. The operating model has no owner because the company is "still piloting AI." The dependency closes the loop, and the only escape is to act on all four problems in the same quarter.
Why the Sequential Playbook Fails Mid-Market Operations Specifically
The sequential pattern — "first deploy AI tools, then redesign the operating model, then clean up data, then assign accountability" — is the default in most mid-market operations functions because it matches how budget cycles, hiring plans, and vendor procurement actually run. Each is a separate line item, owned by a different VP, scored on a different metric. That is exactly the structure that produces the 96% outcome.
McKinsey's State of Organizations 2026 report frames the same finding from the operating-model angle: "Achieving the productivity gains of AI requires challenging and redesigning the operating model of individuals and teams, rewiring end to end, and building capabilities at the same time" (McKinsey, 2026). The operative phrase is "at the same time." The mid-market temptation is to read that as "eventually." McKinsey's analysis is explicit that eventually does not work — the rewire is what generates the productivity, and the AI deployment is what generates the rewire's payoff. Decouple them and neither produces the return.
There is a second reason the sequential playbook fails worse in mid-market than in enterprise. A 200-FTE operations function has roughly 8–12 quarters of runway before its AI investment thesis either compounds or gets cut. Spending two of those quarters deploying tools, two more redesigning the operating model, two more on data, and two more on accountability burns the runway before any one workstream starts reinforcing the others. The compounding never begins. The board sees four cost lines and one flat ROI line. The program gets descoped. The 4% pattern is not about scale; it is about cycle time, and the mid-market has less of it to waste.
What "Parallel" Actually Looks Like in a 200-FTE Operations Function
Parallel does not mean four equal-priority programs running independently — that produces chaos at mid-market scale. It means one program with four workstreams under a shared OKR set, sequenced inside the quarter rather than across quarters.
Shared OKR — not shared status report
The 4% pattern at mid-market scale typically anchors on one quarterly OKR that explicitly requires all four workstreams to move: a measurable improvement in a single high-judgement decision throughput metric (contracts cleared per week, exceptions resolved without escalation, qualified deals routed correctly), where the metric only moves if AI is deployed against the workflow, the workflow is owned horizontally, the data feeding it is clean, and a named person is accountable for the outcome. The OKR is the forcing function. Without it, the four workstreams default back to four separate roadmaps in four separate VP one-on-ones.
One sponsor, one weekly forum, one backlog
The 41% with horizontal operating models share a structural feature most mid-market ops functions do not: one sponsor for the AI program who owns all four workstreams, one weekly forum where the four leads sit at the same table, and one backlog where AI requests, operating-model changes, data fixes, and accountability decisions are prioritised against each other. Splitting the backlog across IT, HR, data, and operations is how the sequencing slips back in.
Data hygiene as a daily operation, not a project
The 87% data-quality finding tells you something specific about how the 4% handle data. They are not running multi-quarter data-cleanup projects in parallel to AI deployment. They are folding data hygiene into the daily operating cadence of the team using the AI — every agent escalation that fails on data quality generates a same-day fix request that the operator owns. The data improves at the speed of agent usage. Treat data as a separate workstream and it lags AI by four quarters; fold it into agent ops and it leads by two weeks.
The Counter-Argument and Why PwC's Data Closes It
The natural counter from a finance-disciplined mid-market COO is: parallel execution costs more upfront, and the 96% are sequencing precisely because they cannot afford the parallel bet. That logic is internally consistent and produces the wrong answer. PwC's data is unusually direct on the math: 89% report that the current sequential approach has already produced technology investments that have not delivered, and 87% report it has produced a data layer that blocks value capture (PwC, April 23, 2026). The sequential bet is not cheaper. It is just spread across more quarters, which makes the failure less visible quarter-to-quarter and impossible to reverse at the end.
The parallel bet at mid-market scale is also smaller in absolute terms than the McKinsey-style enterprise rewire that the same report describes for $10B+ firms. A 200-FTE operations function can run a credible parallel-four program with one additional senior operator (the program sponsor), one weekly forum, and a redirected — not increased — data and engineering budget. The marginal investment is one role and one meeting. The marginal return, per PwC's distribution, is the difference between the 4% outcome and the 96% outcome on the same AI deployment.
What the PwC Data Does Not Say
Two boundaries are worth naming, because the "4%" framing has been used in both directions and the survey supports neither extreme.
The PwC data does not say AI is not working in operations — it is working, in the 4% cohort that hit the parallel conditions. It also does not say mid-market operations cannot replicate the 4% pattern — the survey's sample is U.S. operations and supply chain leaders across the spectrum, and the parallel-four pattern is structural, not scale-dependent. What the data does say is narrower and more useful: the sequential AI roadmap, executed as four serial quarters of work, has a 96% historical failure rate at the cohort where it has been most thoroughly studied. That is not a forecast. That is a backward-looking measurement of what already happened.
The second boundary: "parallel" is not the same as "simultaneous procurement." The 4% pattern is about a single program with shared accountability, not about buying everything at once. Mid-market ops functions that try to procure the AI platform, the operating-model consulting, the data tools, and the accountability framework in the same quarter typically produce the 96% outcome twice as fast. The parallel logic applies to execution accountability, not to vendor purchasing.
The Decision for This Quarter
For a Head of Operations finalising this quarter's AI roadmap between now and end of Q2 2026, the operational implication compresses to one sentence:
No AI deployment workstream gets approved this quarter unless it is paired with the horizontal operating-model change it requires, the data hygiene cadence it depends on, and the named accountability for the outcome it is supposed to produce — and all four sit under one OKR with one sponsor.
If a proposal cannot describe all four in the same document, it is a sequential bet wearing parallel branding, and PwC's distribution says it will produce the 96% outcome. If a proposal can describe all four, it is a candidate for the small minority of AI operating model investments that actually compound. The triage cost is one meeting per proposal. The downside cost of not triaging, at the spend trajectories the survey describes, is most of the AI budget over the next four quarters.
The 4% number is not aspirational. It is what already happened across 767 operations leaders who had the same options as the reader of this article and chose one structure over another. The question now is which structure the next quarter's plan is built on.
