The difference between AI that earns its place and AI that performs

Not every problem is an AI problem. The ones that are require a different kind of discipline: one that starts with data before it starts with models.

There are two kinds of AI investment. One solves a problem. The other performs the act of solving a problem. The distinction is not always obvious at the commissioning stage. It becomes obvious eighteen months later, when the system is quietly decommissioned.

The distinguishing question

When an organisation considers an AI investment, the diagnostic question is simple: what decision will this change?

Not what will it automate. Not what will it optimise. What decision, made by a named person in a specific role with accountability for the outcome, will be different as a result of this system?

If the answer is specific and verifiable, the AI investment may be justified. If the answer is general (better visibility, improved efficiency, strategic advantage) then the investment is likely performing innovation rather than achieving it.

Why the data question matters more than the model

AI models require data to learn from. Not just any data. Structured, consistent, labelled, historically complete data. The model is the last thing you need. The data is the first.

Most institutions that commission AI systems have not audited their data. They know they have data: years of transactions, records, communications, operational logs. What they do not always know is whether that data is structured for analysis, whether definitions are consistent across periods, whether the historical record reflects what actually happened or how a previous system happened to capture it.

An AI model trained on inconsistent historical data learns to reproduce inconsistency, not to predict.

Three tests before the model

Before any AI investment proceeds, three conditions should be met.

1.The problem is defined. There is a specific decision, made by a specific person, that the system will improve. The improvement is measurable.
2.The data is audited. The data required to train, validate, and run the model has been examined. Source, structure, consistency, completeness, and historical depth are understood. Gaps are documented. A remediation path exists.
3.Accountability is assigned. Someone is responsible for the output. If the model produces a prediction, someone's role requires them to act on it. If no one's role changes, the model changes nothing.

Three conditions before any AI model

PROBLEM DEFINED

A specific decision, made by a named role, with a measurable improvement if the model performs correctly.

DATA AUDITED

Source, structure, consistency, completeness, and historical depth understood before the model is specified.

ACCOUNTABILITY ASSIGNED

Someone's role requires them to act on the model's output. The recommendation has a human owner.

If these three conditions are not met, the honest recommendation is to address them before building anything.

Performs

The AI is the visible product. Its presence is the deliverable.

Outputs are impressive in demonstrations. Adoption is low in practice.

No named role acts consistently on the recommendation.

The system is evaluated on its sophistication, not its outcomes.

Earns its place

The outcome is the product. The AI is not mentioned in the room.

A specific decision is made faster or more accurately by a named role.

Someone's job depends on acting on the output correctly.

The system is evaluated on the decision it enables.

What earning its place looks like

AI earns its place when it is invisible. When the analyst makes better decisions without knowing they are being supported by a model. When the operations team catches an exception earlier because a system flagged it. When the credit committee has better information than they had before.

Performing AI is visible. It has a name. It has a launch event. It has metrics that measure its own existence rather than its impact. The difference matters because one is worth building and the other is not.

←All insights

If this resonates with a challenge you are navigating, we are happy to talk through it.

Start a conversation