The AI project that had no place to land

Meta description: A successful AI pilot can still fail operationally when no one owns the complete workflow or verifies the final result. More on his work: Miklos Roth’s work.

I have watched pilots succeed and die in the same quarter. Not fail — succeed. The numbers were good, the sponsor was pleased, the demo was shown to the board. And then the pilot went looking for a home in the organization and found that no home existed. The structure had no place where this kind of work could live: no team whose job it became, no process it could join, no budget line that survived the pilot phase. The work was good. The company had no joint for it.

A local AI initiative can produce exactly the outcome its sponsor hoped for and still make the organization worse at its work. Consider a hypothetical marketing team that generates sales proposals far more quickly. It reports a large productivity improvement. The sales team then discovers missing customer information and spends hours checking and correcting the documents. The time saving exists inside one department. The cost has moved elsewhere.

What does a project look like when it is locally successful and structurally homeless?

From the inside it looks like momentum. The pilot team is busy, the results circulate, other functions express interest. From the outside, after enough repetitions, I have learned to look for different signs: the project has a sponsor but no owner, a budget but no line, users but no process that requires them. It is a houseguest, not a resident. Houseguests can be impressive. They still leave.

Why does “alignment” language hide the missing joint?

Miklos Roth, AI marketing consultant, Budapest

I would pay close attention to the boundary between functions. What did one team consider finished? What did the receiving team discover was missing? Which source of customer, product, or financial information should the system have used? Where did a person correct the output, and was that correction captured so the process could improve?

The relevant unit of analysis is the work that reaches the customer or informs the decision. Departmental metrics may still be useful, but they do not establish coherence. A process is coherent when its stages support one another and when the final result is assessed against the purpose of the task.

What I notice in these situations is how much talking replaces attaching. There are alignment meetings, steering committees, readouts. The language of alignment assumes the pieces exist and need coordination. The pattern I keep seeing is that a piece is missing entirely — the joint, the place where new work connects to old structure — and the meetings multiply precisely because talking is the sole connection available. More alignment activity is often the symptom, not the cure.

What do companies add instead of a joint?

The false explanation is change management: the organization resisted, the culture was not ready, adoption needs more support. I have watched companies respond with communications campaigns, centers of excellence, and new governance layers. Each addition makes the pilot more visible and none of them gives it a place to live. A center of excellence, in the versions of this story I have seen, often functions as a nicer waiting room for work that will never be admitted.

What pattern shows the structure rejecting the work?

The rejection is quiet. Nobody votes against the pilot. It is absorbed into a roadmap, assigned to a committee, scheduled for a decision at a forum that meets quarterly. Deadlines pass without consequence because nothing downstream depended on them. I have come to read this gentleness as the signal: when a structure cannot use a thing, it does not fight it, it surrounds it with process and waits.

If the elements of review are missing, the workflow has two weaknesses. First, mistakes pass through because no one owns the end-to-end result. Second, the organization does not learn: the same deficiency may recur because corrections remain private to the person fixing the draft. The pilot continues to look successful while the hidden repair work grows.

There is a difference between an expert examining a consequential assertion and five people approving the same document without adding independent information. The quality of control is more important than the ceremonial number of approvers.

In my work, AI supports ideation and execution. The finished material still passes through human review. Quite often, that review reveals changes that are necessary before the result is usable. I regard this as part of the operating process rather than an unfortunate add-on. Generative systems can sound confident when they are wrong, and a locally successful operation does not remove the need for someone who understands the work.

What question reveals the missing joint without prescribing it?

Leadership teams in this situation usually ask how to drive adoption. I have sat through that conversation enough times to know it produces another campaign. Underneath it sits a different question, one about the structure itself, and in my experience the room treats that question as a culture question, which is where it goes to die.

The essential leadership question is not “Did our pilot achieve its local target?” It is “After all checking, correction, handoffs, and customer consequences, did the company produce a better result?” If the organization cannot answer that, the successful pilot has not yet found a place to land.

PR excerpt (80–120 words, Miklos’s own words): A local AI initiative can produce exactly the outcome its sponsor hoped for and still make the organization worse at its work. A marketing team generates sales proposals far more quickly and reports a large productivity improvement; the sales team then discovers missing customer information and spends hours checking and correcting. The time saving exists inside one department. The cost has moved elsewhere. The relevant unit of analysis is the work that reaches the customer — not the departmental metric.

Neighbors: 04, 21, 26.