Private intelligence is often commissioned as a feature: connect the files, add a model, let the principal ask questions. The interface arrives quickly. Dependable intelligence does not.
A product has a defined user, job, source boundary, quality standard, owner, service level, feedback loop and retirement plan. Without those, the system becomes an impressive route into an unmanaged archive.
The intelligence contract
Define whose decision the product supports and what it must never decide. Specify the authoritative sources, freshness requirement, jurisdictions and confidence states. Make clear whether the output is research, advice, evidence or an instruction.
The operating loop
- Acquire sources lawfully and assign ownership.
- Prepare and classify information without destroying provenance.
- Retrieve evidence appropriate to the user’s authority.
- Generate or assemble an answer with visible limits.
- Review consequential output with accountable expertise.
- Capture correction and improve the source or rule that failed.
A model upgrade may improve prose while leaving this loop broken. Product performance should measure decision usefulness, evidence coverage, correction time, leakage and the cost of manual rescue.
Ownership cannot sit with technology alone
Technology owns reliability and architecture. Subject experts own meaning. Legal and privacy functions own boundaries. The business or principal’s office owns the job and accepts the residual risk. One product owner must reconcile them and decide when the system should stop.
The product review
Every quarter, sample real questions and trace answer to source and action. Identify questions users repeatedly ask outside the system; they reveal missing trust or capability. Remove sources with unclear authority. Retire functions whose value does not justify their evidence and privacy burden.
A private intelligence product earns trust by being useful, bounded and correctable. A feature merely makes the underlying uncertainty easier to query.