AI governance is the fastest, most sustainable path to AI business value and innovation.
When Your AI Makes a Decision That Affects a Customer or a Dollar, Who Is Accountable?
If your leadership team cannot answer that in one sentence, your AI governance and operating model are already behind the build, with the gap and risk compounding.
Yet companies continue to pour money into AI, even as only 39% report any enterprise-level profit impact.[1] We find the issues are not the technology but the way the business decides what to build, who is accountable, and where the guardrails sit. When the business is making these decisions, we see results like pilots scaling to positive ROIs in weeks (see the case below).
Repeatedly, these key questions are not addressed until pilots have stalled or an incident has forced the issue. Bolted on at the end, such governance feels like a brake to innovation. Designed from the start, governance is an accelerator: a clear, efficient process that manages risk while pushing AI innovation toward measurable business benefit. Well-designed governance accelerates AI efforts faster because it removes the common blockers: relitigation of decisions, rescoping projects that fail risk review in late stages, and rebuilding trust after an incident. Guardrails set in advance are cheaper than rework. And if you are underway it is never too late for good governance.
The Real Constraints Are Organizational and Fixable
McKinsey’s March 2026[2] findings on “AI trust” and other reports find governance among the least-mature capabilities in most organizations, with only about a third reaching higher maturity in strategy and governance even as deployment accelerates. Most companies are stuck in pilots, with only 39% reporting enterprise-level profit impact.[1] In our own work we see late, afterthought governance as a leading cause of stalled and shut-down AI efforts. The disciplines to address governance for AI are not new, they just need to be applied.
2 Predictable Failures When Governance Comes Late
When governance is applied after build, two failures are predictable:
- Priorities multiply and focus dims. Without a clear, shared way to make choices, every idea becomes a “priority,” investment scatters across pilots, and nothing reaches the scale where value appears. Sometimes this looks like a list of use cases with no ranking, where the ones that advance are simply the ones with the loudest sponsor.
- No one owns the outcome. When an AI system recommends or acts, when it interacts with a customer or a unit of revenue, no one can answer who is accountable. Add unmanaged “shadow AI” and rising regulatory exposure (the EU AI Act among others) and the result is not just wasted spend and a stressed technology function but eroded trust slowing every future initiative.
Governance and Operating Design Are Related and Additive
If these conditions are familiar, you may have an ineffective governance model and operating design for AI. They are distinct functions and confusing them is why many AI programs stall.
Governance is decision rights, guardrails, and risk management. It defines who decides, what is allowed, and how risk is controlled. These are one-time decisions with periodic refresh as AI efforts expand with success.
Operating design is how the AI work is implemented and operated day-to-day. This involves designing clear swim lanes of activities, defined roles, and a steady cadence for measurement, reporting, and refinement.
A bolt-on typically produces a policy document with no operating model behind it. Real governance is designed as both, together, before the first model goes live. Clients who take this approach spend their energy and capital building competitive advantage, not negotiating who gets to decide.
But Will This Slow Us Down?
If governance sets the boundaries before work starts, then your teams move inside a known envelope without having to ask permission. The result is not more review but fewer escalated decisions. The difference with our methodology is where the decisions get made, up-front, not after the fact.
Designed-in governance is faster because it removes typical activities that slow AI programs: relitigating decision rights, rescoping projects that fail a late review, and rebuilding trust after an incident. Guardrails set in advance are cheaper than rework, every time.
Bent Flyvbjerg, the Oxford professor whose project database covers more than 16,000 large projects, found that just 0.5% of projects were delivered on budget, on time, and with the benefits promised.[3] The principle he draws from that evidence is think slow, act fast: The projects that succeed invest disproportionately in planning before execution begins, then move quickly inside a settled plan. The ones that fail, reverse the order.
His data also reports IT projects (the category AI initiatives now belong to) average a 73% cost overrun, and 18% of them that exceed budget by more than half overrun by 447% on average. A separate analysis of 1,471 IT projects found the same shape. The average cost overrun was a survivable 27 percent, but one in six projects turned into what the authors call a black swan: a 200% cost overrun and a schedule almost 70% longer than planned. Their conclusion is that by focusing on averages rather than outliers, most managers miss the real risk in IT entirely.[4] Deliberate design followed by fast execution is not a stylistic preference. It is the pattern that measurably performs better by category leaders.
A Blueprint for Leaders
The organizations that get this right tend to organize their AI program around three swim lanes of activity, each with clear ownership:
- Governance and Oversight: Includes decision rights, risk thresholds, and escalation paths; owned by the oversight committee.
- Innovation and Demand Management: Project intake, value scoring, and portfolio sequencing; owned by the business.
- Execution and Change Management: Build, deploy, adopt, and measure; owned jointly by technology and function leads.
These lanes of activity run as one program with one plan to ensure a holistic, enterprise approach to an effort that should be end-to-end. Here are four first moves to make it concrete:
- Stand up an AI oversight committee with clear decision rights and plain-language risk guardrails, so accountability is settled before deployment.
- Create a structured intake and value-based prioritization process that scores every idea on value to the business and the customer. This mechanism solves the “everything is a priority” problem.
- Adopt a stage-gated lifecycle process of ideate, prioritize, test, and deploy, so initiatives advance within governance rather than around it.
- Publish responsible-AI guidelines and build an adoption network that moves teams from fear to confident use, so the guardrails are lived, not shelved.
How Ankura Delivered Value in Less Than 8 Weeks
A large U.S. franchise network with more than 2,300 locations saw AI’s potential but had no structured way to govern, prioritize, and adopt it. Ankura designed and implemented an AI operating model that aligned client leaders from all key functions into a single, agile oversight committee. With that committee, we designed and implemented the value-based intake, stage-gated lifecycle, and the change program for proposed AI candidate projects.
Results in Less Than 8 Weeks:
- An AI-enabled IVR and chatbot selected, implemented, and live in the call centers serving the client’s largest franchisee.
- AI initiatives brought under managed inventory, reducing enterprise risk.
- Leadership aligned on priorities and investment, with future spend concentrated on the top five highest-value use cases.
- Return on the technology investment turned from negative to positive.
What To Do in the Next 30 Days
Here are three steps toward improved AI ROI:
- Name a single accountable owner for every AI initiative already in flight.
- Inventory the shadow AI your teams are already using.
- Draft one page of decision rights and circulate it to your leadership team.
That is enough to start, and the rest will follow.
The Question to Ask This Quarter
Return to the diagnostic. When your AI makes a decision that affects a customer or a dollar, who is accountable? If your leaders cannot answer in a sentence, the next step is not a new committee or a longer policy. It is to map your decision rights, accountable owners, priorities, and swim lanes before the next pilot, because designing governance is faster and cheaper than retrofitting it later.
The winners in enterprise AI will not be the companies that governed it last. They will be the ones that designed for it first. If that is a conversation your leadership team is ready to have, we would welcome it.
Sources
[1] McKinsey & Company, “The state of AI in 2025” (November 2025).
https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
[2] McKinsey & Company, “State of AI trust in 2026: Shifting to the agentic era” (March 2026). https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/state-of-ai-trust-in-2026-shifting-to-thehttps://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/state-of-ai-trust-in-2026-shifting-to-the-agentic-eraagentic-era
[3] Bent Flyvbjerg and Dan Gardner, How Big Things Get Done (Penguin Random House, 2023). Base rates and category overrun figures from the book’s appendix, drawn from the authors’ database of more than 16,000 projects.
[4] Bent Flyvbjerg and Alexander Budzier, “Why Your IT Project May Be Riskier Than You Think,” Harvard Business Review, vol. 89, no. 9 (September 2011), pp. 23-25. Analysis of 1,471 IT projects.
© Copyright 2026. The views expressed herein are those of the author(s) and not necessarily the views of Ankura Consulting Group, LLC, its management, its subsidiaries, its affiliates, or its other professionals. Ankura is not a law firm and cannot provide legal advice.
