AI architecture and data readiness for production

Production readiness depends on the whole system: information sources, identity and access, integrations, solution behaviour, oversight and support. A model choice or a successful demonstration does not describe all those relationships.

Enterprise architecture places the initiative in the environment that already exists. Which capabilities will it use? Which platforms will be affected? Where do new dependencies arise? The assessment makes those choices visible so that approval includes the operating commitment that comes with them.

Data and context

Are sources available, current and authorised for the intended use? Are meaning, quality, provenance and data ownership understood?

Integrations and boundaries

What does the initiative read or change? How does authentication work? Which systems, interfaces and suppliers become dependencies?

Operations and recovery

How will failures or unsuitable outputs be identified? Who maintains the service and how does the process continue when the solution is unavailable?

Architecture documents, data flows, test findings, operating requirements and consumption estimates help establish what is already known. Each piece of evidence should relate to the intended use: volume, diversity of inputs, access profiles, expected quality and failure conditions.

Consider an illustrative example: a pilot queries a manually reviewed knowledge base for a small group. Expanding to sources that change daily and users with different permissions raises new questions. Updates, access filtering, traceability and support need examination before the pilot result is extrapolated.

Deeper integration can increase usefulness and the impact of a failure. A managed service may simplify part of the operation while creating dependencies on availability, usage terms or portability. Human review may change capacity and response time. These choices need to be assessed in the context of the process.

Usage, maintenance, oversight and change costs should be examined alongside expected value. Without adequate data, an estimate remains a hypothesis. The recommendation may call for an additional test, narrow the scope or make expansion conditional on resolving a dependency.

Within AI Decision Gate, architecture and data review connects to business, governance and risk. A technical finding should explain its consequence: what it prevents, what alternative exists and what evidence would support proceeding. Leadership can then distinguish a desirable improvement from a necessary condition for commitment.

This page describes an area of expertise within independent assessment. It does not promise model development, integration delivery or managed operations. Technical depth, required access and collaboration with architecture, data, security and operations teams are defined around the initiative. Implementation remains with the organisation’s teams and chosen partners.