Every retail and CPG board I sit with has approved some version of the same thing: a data readiness program. Multi-year. Enterprise-wide. A single governed foundation that every future agent will consume.
It sounds responsible. It is a trap.
As companies push their AI pilots to scale, data is emerging as a constraint. Leaders are prioritizing data readiness, connecting structured and unstructured data into a governed, reusable foundation. McKinsey is right about the diagnosis. The constraint is real. The models are not the bottleneck. The data underneath is.
Where I break with the reflex is the treatment. The enterprise-wide foundation program is the single most reliable way to defer value past the patience of the people funding it.
The numbers are an argument against the big program, not for it
Look at what the readiness industry keeps quoting. Gartner projects that through 2026, organizations will abandon 60% of AI projects that lack AI-ready data foundations. And a Cloudera and Harvard Business Review Analytic Services survey of 1,574 enterprise IT leaders, published in March 2026, found that only 7% say their organization’s data is completely ready for AI adoption.
Everyone cites these to justify a larger foundation project. Read them again. A 60 percent abandonment rate and a 7 percent completion rate are not evidence that you need a bigger program. They are evidence that the big-program model has a catastrophic finish rate.
Enterprise-wide readiness is a boil-the-ocean project wearing a governance costume. It has three quarters of runway before a CFO asks what shipped. The honest answer is usually a data catalog and a lineage tool. No margin. No agent in production. Just a bill.
Meanwhile the sponsor rotates. The mandate softens. The program becomes a line item nobody will kill and nobody can defend.
The operators shipping agents scope readiness to one workload
Here is what the teams actually putting agents into production do differently. They do not try to make the estate ready. They make one workload ready.
Closing the data-readiness gap does not require pausing all AI development for a multi-year infrastructure overhaul. The most successful technology leaders run data remediation and AI deployment in parallel, focusing on specific, high-ROI workflows.
Pick replenishment. Or invoice matching. Or the rework loops in your contact center. Assess the specific data sources required for a single use case against strict quality, governance, and structural metrics. Fix missing schemas, resolve duplicates, and establish clear access controls for that isolated dataset.
The failure mode is the opposite instinct. Attempting to cleanse the entire organizational data estate simultaneously leads to project paralysis.
Workload-scoped readiness inverts the funding logic. Instead of asking the business to bankroll a foundation on faith, each workload carries its own business case. The data work is scoped to what that agent touches. The agent ships. Margin shows up. That margin funds the next workload, and the readiness backbone grows out of live usage instead of a slide.
This is also why the discipline matters more than the enthusiasm. A disciplined prioritization framework evaluates impact. It also evaluates feasibility and data readiness, in addition to reuse potential. This prevents wasted energy and ensures AI is deployed where it can reshape performance. Readiness is one input into a prioritization decision, not a prerequisite gate in front of the whole portfolio.
The honest tradeoff, and who owns it
I will not pretend workload-scoped readiness is free of cost. It is not.
You will duplicate effort. Two workloads will remediate overlapping data and you will pay for some of it twice. You will not end up with a pristine, uniform enterprise estate. You will accumulate a patchwork that a purist will hate. Governance has to be threaded through each workload rather than declared once from the center, which demands more coordination, not less.
That is the trade. A messier map in exchange for agents in production and a foundation that earns its keep as it grows.
Someone has to own that tradeoff explicitly, and it is not the data team. It is the C-suite. The CFO decides whether the enterprise funds a foundation on faith or funds workloads that self-liquidate. The Chief Data Officer decides how much duplication is tolerable before a shared layer gets promoted from a workload up to the center.
Now the Cost of Doing Nothing. The CODN on the big program is not neutral delay. It compounds. 47% of survey respondents said they’re using or assessing agentic AI, with 20% saying AI agents are already active in their organizations and another 21% reporting agents are coming within the next year. While you perfect the estate, competitors are learning from live agents in the market. That learning gap does not wait for your foundation to finish.
Stop funding readiness as a destination. Fund it as the byproduct of workloads that pay for themselves. The enterprise foundation you actually want is the one you assemble from workloads already in production, not the one you promise a board you will finish in three years.