Naming an accountable owner for a clinical AI deployment makes that deployment scale roughly six months faster. Black Book Research measured it in its 2026 Health System and Hospital AI Governance Benchmark. Programs with governance dashboards and defined ownership structures reached early return on investment in approximately seven and a half months. Programs without them took approximately thirteen and a half. Separately, pilots with a named owner and a tested kill switch were twice as likely to scale system wide within one year. This is the first external measurement, with an effect size attached, of a proposition GPe Research has argued on structural grounds since Working Paper No. 01. The Named Owner Principle holds that every AI deployment requires two named persons in the audit trail, a Governance Owner and a Decision Owner, and that the absence of either opens The Accountability Gap™. The benchmark does not validate the two seat architecture. It measures one of the two seats. What it establishes is narrower and, for the institutions deciding where governance sits in a budget, more consequential: the Decision Owner is not only a control. It is a speed variable.
1. What was measured
The benchmark was released on November 11, 2025, drawing on survey data collected from hospital and health system executives across North America, Europe, Asia-Pacific and the Middle East. Results were validated at the 95 percent confidence level with a margin of plus or minus 3.7 percentage points.
Four findings bear directly on deployment economics.
Time to early return. Programs with governance dashboards and defined ownership structures reached early return in approximately seven and a half months. Programs without reached it in approximately thirteen and a half.
Probability of scaling. Pilots with a named owner and a tested kill switch were twice as likely to scale system wide within one year.
Governance council effect. Health systems operating an AI Governance Council were twice as likely to achieve return within twelve months.
Contract exposure. Sixty percent of contracts lacked a material change re-validation clause, and eighty percent of executives reported that vendor AI claims were difficult to verify without formal governance.
Doug Brown, president of Black Book Research, characterised the moment in the accompanying release by saying that healthcare is entering the accountability era of AI.
A separate set of figures reported in December 2025, drawn from 182 hospital leaders surveyed between October 15 and November 8, 2025, ranked the barriers to auditing a deployment. Limited documentation from vendors came first at 41 percent. Incomplete tracking of data inputs and model versions came second at 37 percent. Unclear ownership between information technology, quality and safety, and compliance came third at 33 percent.
These two sets are treated separately throughout this paper. See the methods note in section 7.
Governance was budgeted as friction. The return column measured it as speed.2
Pilots with a named owner and a tested kill switch were twice as likely to scale system wide within one year.
of hospital leaders reported unclear ownership between information technology, quality and safety, and compliance as an audit barrier.
Two named seats. One junction.
1 Black Book Research, 2026 Health System and Hospital AI Governance Benchmark. 2 GPe Research interpretation, not a benchmark finding.
2. Why the finding is not obvious
Governance enters the institutional budget as overhead. It is scoped alongside legal review, security attestation and compliance documentation, and it is understood as the cost of being permitted to deploy rather than as a contributor to the deployment succeeding. The internal argument that follows is familiar to anyone who has sat through it. Governance is necessary, governance is expensive, and governance is the reason the timeline slipped.
The benchmark inverts the causal direction. In the surveyed population, the programs carrying more governance structure did not deploy more slowly. They realised value in a little over half the time.
Two readings are available and they are not equivalent.
The weaker reading is selection. Institutions capable of defining ownership are also institutions with better project discipline, clearer executive sponsorship and more mature information technology functions, and those attributes drive the speed rather than the ownership itself. The benchmark is observational and cannot exclude this.
The stronger reading is mechanism, and it is the one this paper argues. The absence of a named owner does not slow a deployment during the build. It slows it at the moment the deployment has to expand, because expansion is a decision and an unnamed decision has no one to make it.
3. The mechanism: where the six months actually sit
A clinical AI deployment that clears its pilot has already cleared the hard technical gates. The model has been validated. The integration is stable. Monitoring is running. Uptime commitments are being met. Nothing in the engineering record explains a further six months.
What remains after the pilot is a decision, and it has a particular shape.
Someone has to determine that the tool moves from the unit that piloted it to the units that did not. That determination carries clinical risk, workflow disruption and budget. It is a decision in the full sense of the word, and it is the one decision in the deployment lifecycle that most governance documentation does not assign.
Deployment records routinely name four roles. Who maintains the model. Who operates the platform. Who manages the integration. Who receives the monitoring dashboard. Each of those is an operational responsibility, and each is correctly assigned.
None of them is the person who decides the tool expands.
When that seat is unnamed, the expansion question does not disappear. It routes to whatever body is available to receive it, which in most health systems is a committee with a quarterly cadence. A question that arrives one week after the committee meets waits eleven weeks for an answer that could have been given in an afternoon. A second unit that requests the tool needs a clinical sponsor to carry the request, and when no sponsor was assigned at the pilot, the request waits for one to volunteer.
Neither of those delays appears in the project plan. Both of them appear in the return.
This is the same structural observation Working Paper No. 01 made about the clinical handoff, arriving from the opposite direction. A real handoff transfers a named owner along with the information. The pilot to scale transition transfers a validated tool and an expectation, and leaves the receiving seat empty. In the clinical case the consequence is an unowned safety signal. In the deployment case the consequence is eleven weeks of nothing.
4. The three measured conditions, read against the framework
Black Book measured three structures. Read together, they describe a seat rather than a process.
A named owner. This corresponds to the Decision Owner in The Accountability Gap™ framework: the person who carries the call the system produces, per deployment rather than per portfolio. The benchmark measured it in the context of scaling. The framework specifies it in the context of the clinical decision. Both require the same thing, which is that the name attaches to a single deployment and not to a category of deployments.
A tested kill switch. This is the operational test of whether the named owner holds authority or only holds a title. A named owner without the power to suspend is a signature on a charter. The benchmark measured the named owner and the kill switch together, and the pairing is not incidental. The institution is not buying a name. It is buying a person who can stop the thing.
A governance dashboard. This is visibility of the seat rather than visibility of the system. Dashboards in most institutions report model performance, drift and uptime. The benchmark’s association is with ownership being defined and visible, which is a different object. It is the difference between knowing how the model is behaving and knowing who answers for what it produces.
The Governance Owner, the second seat in the framework, is not measured anywhere in the benchmark. That seat charters what a system may decide before the build begins, and no instrument in the surveyed set appears to have asked about it. The framework’s claim that two seats are required therefore remains unmeasured externally. What has now been measured is that one of them pays.
5. The same seat, priced twice
The audit barrier figures and the return figures describe one absence from two positions.
At 33 percent, unclear ownership between information technology, quality and safety, and compliance ranked third among barriers to producing an auditable account of a deployment. Each of those three functions owns something real. Information technology owns integration and uptime. Quality and safety owns review after an event. Compliance owns policy and attestation. All three answers are correct within their own scope, and none of them is the clinical decision.
The institution therefore holds the same vacancy in two ledgers. In the risk ledger it appears as an audit that cannot be completed. In the financial ledger it appears as six months of a deployment that has been paid for and is not yet producing.
Only the first of these has historically been used to argue for governance investment, and it is the weaker argument, because it describes a contingent future cost. The second describes a cost the institution is paying now, on every deployment currently sitting between pilot and scale.
6. What this changes
For a chief medical information officer building a governance case, the argument no longer has to be made on risk alone. The benchmark supplies a return argument, and the return argument survives contact with a finance committee in a way that a risk argument often does not.
For a chief financial officer, the relevant question about any stalled clinical AI deployment is no longer whether governance slowed it down. It is whether the expansion decision had a name attached to it, and if not, how long the deployment has been waiting for one.
For a board, the finding reframes an oversight question as a performance question. Asking which deployments have named owners is no longer only a compliance enquiry. It is an enquiry into where value is currently trapped.
The operational recommendation that follows from the benchmark is narrower than most governance recommendations and therefore cheaper to act on. Assign the accountable clinical owner at the pilot rather than at scale. Give that person the authority to suspend. Make the assignment visible in the reporting the executive team already reads.
None of that requires new infrastructure. It requires a name assigned earlier than institutions currently assign it.
7. Methods, attribution and limitations
Source separation. The return and scaling findings in section 1 derive from the 2026 Health System and Hospital AI Governance Benchmark released November 11, 2025, whose stated methodology describes respondents across four regions validated at 95 percent confidence with a margin of plus or minus 3.7 percentage points. The audit barrier figures derive from reporting of a survey of 182 hospital leaders conducted October 15 to November 8, 2025. These sample descriptions do not reconcile to a single instrument. This paper does not treat them as one survey and no finding here combines them.
Observational design. All findings discussed are associations reported from executive survey data. None establishes causation. The mechanism proposed in section 3 is an argument from structure and from the shape of the deployment lifecycle. It is not a measured causal pathway and should not be cited as one.
Self interest disclosure. GPe Research publishes the Named Owner Principle and The Accountability Gap™ framework and has a direct interest in evidence that supports them. The benchmark was produced independently of GPe Research and was not commissioned, funded or reviewed by it. Readers should weigh section 4 accordingly, since it is interpretation rather than finding.
What is not claimed. The benchmark does not measure the Governance Owner seat, does not measure the documented handoff, and does not address clinical AI specifically as distinct from health system AI generally. Its validation of the framework is partial by construction.
Framework glossary
The Accountability Gap™ (TAG™). The structural failure point where an AI system produces an output and no named person holds the decision that follows from it.
Named Owner Principle. Every AI deployment requires two named persons in the audit trail, a Governance Owner and a Decision Owner, with neither substituting for the other.
Governance Owner. The person who charters what a system may decide, assigned before the build begins, accountable for the decision boundary.
Decision Owner. The person accountable for the call the system produces, assigned per deployment, carrying safety signals, escalation and suspension.
The Handoff. The documented junction at which authorisation stops being the same act as approval, recorded at the moment the decision is made rather than reconstructed afterwards.
Early return. As used in the benchmark, the point at which a deployment begins producing measurable value, reported here in months from deployment.
References
-
Black Book Research. 2026 Health System and Hospital AI Governance Resource Guide: Global Benchmark and Regulatory Readiness Report. Released November 11, 2025. https://blackbookmarketresearch.com/governing-hospitals-ai-2026-board-to-bedside-accountability-guide
-
Advisory Board. Reporting on Black Book Research survey of 182 hospital leaders, fielded October 15 to November 8, 2025. December 16, 2025. https://www.advisory.com/daily-briefing/2025/12/16/ai-governance
-
Johnson M. The handoff that isn’t: how clinical AI escapes accountability. GPe Research Publications, Working Paper No. 01. June 6, 2026. https://publications.gperesearch.com/papers/the-handoff-that-isnt
-
Johnson M. The Accountability Gap™: two seats, one handoff, any regulated sector. GPe Research Publications, Framework Brief No. 02. July 26, 2026. https://publications.gperesearch.com/papers/the-accountability-gap
-
Johnson M. Overridden: what a quarter century of ignored alerts establishes about accountability. GPe Research Publications, Position Paper No. 03. September 8, 2026. https://publications.gperesearch.com/papers/overridden
Frequently asked questions
- Does this mean governance makes deployments faster?
- It means that in the surveyed population, programs with defined ownership and governance dashboards reached early return in roughly half the time. The design is observational, so the finding is an association rather than proof of cause.
- Is a named owner the same as an executive sponsor?
- No. An executive sponsor advocates for a deployment. A Decision Owner is accountable for the calls the deployment produces and holds the authority to suspend it. The benchmark measured the named owner together with a tested kill switch, which distinguishes the two.
- Can a committee be the named owner?
- Not usefully. A committee can hold a governance function, but it cannot hold a decision that has to be made between meetings. Where the expansion decision routes to a quarterly committee, the committee's cadence becomes the deployment's timeline.
- Our deployments each have an owner on paper. Does this apply to us?
- The test is whether the owner is assigned to the individual deployment or to a portfolio, and whether that person can suspend the deployment without escalating. One approval covering nine tools does not create nine accountable deployments.
- What is the difference between this and general AI governance?
- General AI governance addresses model risk, data handling and vendor management. Clinical AI governance addresses the decision that reaches a patient. The two are frequently conflated, and the conflation is why the decision seat goes unassigned while the technical seats are all filled.
- Where should the named owner appear?
- In the deployment charter at the pilot, in the reporting the executive team already reads, and in the record created at the point of care. The third of these is where audits are lost.
- What does this cost to implement?
- It is an assignment rather than a build. The constraint is organisational willingness to attach a person's name to a clinical AI decision before the deployment succeeds, not budget.
Funding
None declared.
Conflicts of interest
Mo Johnson, MD MBA is the founder of GPe Research. The Accountability Gap™ (TAG™), the Named Owner Principle, and the frameworks named in this paper are works of GPe Research. The author has a commercial interest in the adoption of the frameworks described in this paper.
How to cite
@techreport{johnson2026unscaled,
author = {Johnson, Mo},
title = {Unscaled: What the First Measurement of the Decision Owner Establishes About Deployment Speed},
institution = {GPe Research Publications},
type = {Position Paper},
number = {№04},
year = {2026},
month = {9},
url = {https://publications.gperesearch.com/papers/unscaled}
}
Version history
Version 2 — September 17, 2026
Read time corrected from 13 minutes to 9. The computation counted this paper's native figure scoped stylesheet as prose, adding 744 words that a reader never sees. No prose, figure or citation was changed.
Version 1 — September 15, 2026
Original publication.