The AI Lens: Designed Process, Governed Data

Why capable pilots change nothing

A pilot proves that a capability exists. It says nothing about whether the enterprise can absorb it. Absorption requires a defined process to insert the capability into, a measure that will register the improvement, and an owner accountable for the outcome rather than the technology.

Where those are missing, the pilot produces an impressive demonstration and a quiet return to the previous way of working. The organization concludes that the technology is immature. Usually the technology was the only mature element in the exercise.

AI applied to an undesigned process automates a decision nobody can explain.

Governed data is a prerequisite, not a workstream

Models inherit the discipline of the records they read. Where the customer master has three versions of the same account and the item master carries obsolete records nobody may delete, the output is confidently wrong — and confidently wrong is considerably more expensive than obviously wrong.

Governance here means something concrete: a named owner per data object, a rule for creation, a rule for retirement, and a measure of record quality that someone reviews. Four things, rarely all present.

Name the measure before the use case

The most useful discipline in an AI portfolio is to require, for every candidate use case, the measure it is expected to move, its current value, and who owns it. Candidates that cannot supply all three are research, not initiatives, and should be funded and governed as research.

This single filter typically removes half of an AI backlog and doubles the credibility of what remains — which is the outcome a board is looking for when it asks what the investment has produced.

Exhibit — The AI candidate filter

One row per candidate use case. A blank in any of the first four columns means the candidate is not ready to fund as an initiative.

The filter is not a governance formality. It is the difference between an AI portfolio and a collection of demonstrations.

Before the next pilot is funded

• The process the capability will sit inside is defined as performed.
• The measure it is expected to move is already being measured.
• Each data object it reads has a named owner and a retirement rule.
• An accountable business owner exists, distinct from the technology sponsor.
• The organization has capacity to adopt the change, not only to trial it.

AI rewards enterprises that have already done the unglamorous work. That is inconvenient, and it is the entire finding.

Data and AI amplify the operating system beneath them — including its gaps.