Hello,

Three field notes went up on Navata Insights in September.

One examined how to evaluate a Veeva or AI advisory firm. One examined AI models that return probabilities designed for software to act upon. One examined what a completed approval chain actually proves.

Across all three, the same question kept surfacing:

What entitles the next consequential decision?

The first article turned the Navata Method into a buyer’s test.

Credentials, certifications and client references establish whether an advisory firm belongs in the conversation. They do not establish whether that firm can investigate what is actually true, turn intent into a defensible architecture, define evidence capable of disproving its own recommendation, and leave the client able to challenge the decision after the firm has gone.

The four tests - Investigate, Architect, Prove and Own, should leave behind four client-owned artefacts: a decision basis, decision record, acceptance basis and ownership record.

The second article moved the same question inside an AI-enabled workflow.

A model may return a perfectly typed choice, score or probability. That removes malformed output and schema failure. It does not establish whether the result is calibrated well enough, whether the operating threshold reflects the consequences of error, or whether software should be permitted to act when that threshold is crossed.

For a decision-returning model, the validated object is therefore broader than the model. It is the Decision Contract connecting State, Decision Definition, Probability, Threshold, Fallback, Authority and Lifecycle.

A probability becomes operationally meaningful only when the organisation has decided what that probability permits software to do.

The third article examined approval itself.

A platform release may carry valid approvals from Digital, process owners, Validation and Quality. Each person may correctly establish their own part while inheriting claims from earlier steps.

The failure appears when a claim required by the final decision was never independently established by anyone. It was merely assumed to be true because every approval completed.

The Entitlement Test traces each claim backwards from the final decision until it reaches independent evidence or exposes an orphaned, circular or overloaded claim.

September’s conclusion is straightforward:

A credential, probability or signature may be necessary. None is sufficient on its own.

The defensible state lies in the traceable chain connecting evidence, authority and the decision that follows.

September’s field notes:

• The Navata Method: What to Test Before You Hire or Switch a Veeva or AI Advisory Firm

• What Changes When an AI Model Returns Decisions Instead of Text?

• The Architecture of an Approval

If you are selecting or reconsidering an advisory firm, begin with the Navata Method buyer test.

If you are designing regulated AI, begin with the Decision Contract.

If your concern is release or acceptance evidence, begin with the Entitlement Test in “The Architecture of an Approval.”

Rohith
Founder, Navata Ltd