Context

The organisation had a large and mature data estate built over many years to support reporting, operations, and regulatory needs. This data landscape included legacy platforms, bespoke integrations, and varying levels of documentation and stewardship. As demand for AI grew, teams increasingly wanted to reuse this data for modelling and automation. There was broad agreement that the data was valuable, but far less consensus on how much of it needed to change before AI could be used effectively.

‍

The Challenge

The gap between aspiration and practicality was significant. Rebuilding data foundations to be “AI first” would take years and disrupt critical processes. At the same time, treating existing data as good enough led to repeated friction: unclear meaning, inconsistent access, and heavy manual effort to prepare data for each AI initiative. Centralising all remediation risked overwhelming data teams, while pushing responsibility to individual AI efforts created duplication and fragile, one off solutions. The organisation needed a way to move forward without committing to an unrealistic reset.

‍

The Decision

The organisation chose to treat AI readiness as an incremental operating discipline rather than a one time transformation. Instead of rebuilding platforms or attempting to standardise everything upfront, leadership made deliberate choices about which data needed to be made AI ready, under what conditions, and for what purpose. They defined a threshold for readiness that was sufficient for reuse and accountability, even if it fell short of theoretical perfection. The alternative-either rebuilding the entire data estate or allowing every AI team to resolve readiness independently-was explicitly rejected.

‍

What Changed

Data conversations became more grounded in intent and consequence. Teams became clearer about which data was suitable for AI use, which required additional stewardship, and which was not worth enabling in the near term. Data remediation efforts were prioritised based on reuse and longevity rather than local demand alone. Some AI initiatives slowed as data constraints were surfaced earlier, but fewer stalled mid delivery due to latent issues. Readiness became visible and cumulative rather than implicit and repeatedly rediscovered.

‍

Why This Matters

Many organisations delay AI adoption by assuming data must be perfect before it can be useful, or they rush ahead and absorb hidden complexity downstream. Treating AI readiness as an incremental, decision led process allows enterprises to unlock existing data while respecting legacy constraints. Progress comes not from rebuilding everything, but from being explicit about what “ready enough” means and who is accountable for maintaining it.