Five Questions That Tell You If Your AI Initiative Will Stall

AI initiatives that stall don’t fail suddenly. They slow down gradually — losing momentum across weeks and months until the energy behind them quietly dissipates and the project joins the list of things the organization is still planning to do.

The stall is almost always visible in advance. Not in the technology. In the answers to five questions that most organizations never ask before deployment begins.

CEOs I work with who have deployed AI successfully share one pattern: they asked these questions early — before the project had accumulated too much momentum to stop and reconsider. The answers told them what needed to be resolved before development continued. Some answers sent them back to the drawing board. Most sent them forward with significantly more clarity than they had before.

These five questions are not a framework. They’re a diagnostic. Run them against your highest-priority AI initiative before the end of this week.

Question 1: Can you name the one person who owns this initiative — the decision to deploy and the outcome after?

This is the single most predictive question in the diagnostic. Not because ownership is the most technically complex governance challenge — it isn’t — but because the absence of a named owner predicts almost every other governance failure that follows.

When nobody owns the outcome, production criteria stay undefined because nobody has the authority to define them. Data quality gaps persist because nobody has the accountability to resolve them. Deployment decisions stall in circular stakeholder alignment because nobody has the authority to call it.

The answer to this question is either a name or it isn’t. A committee is not an answer. A shared accountability structure is not an answer. A name — one person, with authority and accountability — is the answer.

If you can’t name that person today, that’s your first governance gap. And it’s resolvable before this week is out.

For more on how decision rights structure determines deployment outcomes, read the Decision Authority Alignment post on the RSA blog.

Question 2: Do you have data that’s probably good enough for this specific use case?

Not perfect. Probably good enough — for this one deployment, at this stage of the initiative.

The data question that stalls mid-market AI deployments is almost never “do we have enough data.” It’s “is the data we have consistent enough, complete enough, and accurately defined enough for this AI to produce reliable outputs.”

A McKinsey analysis of AI deployment failures found data quality issues present in more than 60% of underperforming AI initiatives. In most cases the data existed. The quality — consistency of definitions, completeness of records, accuracy of historical entries — was the problem.

The right data question before deployment is scoped and specific: for the inputs this AI requires, is the data adequate for production use? That assessment takes days, not months. And it almost always surfaces something worth knowing before development goes further.

If you’re not sure whether your data is adequate for this use case, that uncertainty is the answer. It means the assessment hasn’t happened yet.

Question 3: Does everyone involved agree on what “ready for production” means?

This is the question that most directly predicts whether your AI initiative will deploy on time or drift indefinitely.

Production criteria — the specific, measurable conditions that must be satisfied before the AI goes live — are the governance element most consistently missing from mid-market AI deployments. Not because organizations don’t know they need them. Because nobody sat in a room before development started and agreed on what done looks like.

Without defined production criteria, every stakeholder applies their own standard. The result is a deployment that’s always almost ready. Legal wants one more review. IT wants one more test. The business owner wants one more pilot. The AI sits in deployment limbo not because it isn’t ready — but because nobody agreed on what ready means.

Ten to fifteen specific criteria — security cleared, data quality confirmed, compliance verified, business value demonstrated, team trained — agreed in writing before development begins, owned by the named deployment authority, change this dynamic entirely.

Gartner’s AI governance research identifies undefined production criteria as one of the top three causes of stalled AI deployment in mid-market organizations. The fix costs an afternoon of focused conversation and one page of documentation.

Question 4: Have the people who will use this AI been part of the conversation?

This question sits at the intersection of governance and human readiness — and it’s the one most consistently skipped in the rush to deployment.

The AI governance frameworks that stall in production almost always share one characteristic: the people who were supposed to use the AI weren’t involved in the decisions that determined how it would work. They received a briefing. They attended a training. They were informed of a deployment that had already been decided.

Mercer’s 2026 Global Talent Trends research found that 62% of employees feel leaders underestimate the human impact of AI decisions. The organizations where that gap produces resistance, workarounds, and shadow AI almost always skipped this question.

Involving people doesn’t mean governance by committee. It means the people who will use the AI have had a genuine conversation about what it does, how it affects their work, and where their judgment still leads. That conversation takes ninety minutes. The resistance it prevents can take months to overcome after the fact.

For more on how this conversation connects to AI governance outcomes, read The AI Conversation Your Leadership Team Hasn’t Had Yet on the RSA blog.

Question 5: Can you explain this AI’s decisions in plain language to someone who didn’t build it?

Not technically. In plain language — to a skeptical employee, to a regulator conducting a review, to a board member asking how the AI made a particular recommendation.

This question matters for two reasons that most deployment checklists separate but that are actually the same reason.

The first is compliance. Depending on your industry and the nature of the AI’s decisions, explainability requirements exist in regulation and are expanding. An AI that makes consequential decisions without a plain-language explanation of its logic creates compliance exposure that surfaces at the worst possible time — after deployment, when the cost of addressing it is highest.

The second is organizational trust. The people using an AI they can’t explain are the people who stop trusting it — and stop using it — when the first unexpected output appears. Explainability isn’t just a regulatory requirement. It’s the foundation of the human trust that makes AI adoption sustainable.

If you can’t explain this AI’s decisions in plain language today, that’s not a reason to stop development. It’s a reason to build the explanation before deployment rather than after the first incident.

What the Answers Tell You

Five questions. Five binary answers — either you have them or you don’t.

A named owner. Adequate data. Defined production criteria. People involved. Plain-language explainability.

If you can answer yes to all five for your highest-priority AI initiative, you’re ready to deploy. Not enterprise-wide. Not permanently. For this initiative, right now.

If you can’t answer yes to some of them, those gaps are your governance priorities. They’re specific, addressable, and bounded. They tell you exactly what needs to happen before deployment — not in general, but for this initiative.

The organizations waiting for a comprehensive governance framework to be complete before answering these questions are doing it in the wrong order. These questions build the framework. The framework doesn’t answer the questions.

For a structured assessment of where your organization stands across all seven governance dimensions — including the readiness factors behind each of these five questions — the free CAGF Assessment takes approximately ten minutes and tells you exactly where your gaps are.

The Monday Morning Question