TL;DR: Enterprise AI conversations are dominated by talk of models, GPUs, and prompt engineering. But the reason most organisations stall on Copilot and other AI-enabled tools isn't technology. It is decades of unmanaged content, missing retention rules, and no clear ownership over what should exist inside the tenant in the first place. Records management, the unfashionable cousin of information governance, has quietly become the most important discipline in enterprise AI.
Everyone is asking the wrong question
Walk into any AI readiness conversation right now, and the discussion follows a predictable arc. Which models are approved. Whether the data lakehouse is ready. What the identity model looks like. Where the tokens are processed. All valid questions. None of them the actual constraint.
The actual constraint is that the content the AI will read is a mess. Twenty years of unclassified documents. Retention schedules that exist on paper but not in the tenant. SharePoint sites that outlived the projects they served by a decade. Personal drives full of duplicate financial models. Team channels used as informal filing cabinets. AI does not fail on this content because the AI is bad. It fails because the content should not exist in that form, in that location, with that access profile, in 2026.
Why does everyone treat AI readiness as a technology problem?
Because technology problems are easier to buy.
A licence line for Copilot is a clean procurement action. A Purview add-on is a clean procurement action. An identity uplift is a clean procurement action. A records management programme is none of those things. It requires cross-functional accountability and an unglamorous multi-year effort to fix decisions that were never really made.
So the industry has quietly redefined AI readiness as the set of things you can put on a purchase order. The deeper question, about what content should exist at all, gets deferred.
What does records management actually control that matters for AI?
Four things, all of them foundational.
What exists. Retention and disposition rules determine whether content is still in the tenant at all. Without them, AI reads through the last decade of decisions the organisation forgot to unmake.
Where it lives. Records classification determines whether sensitive content sits in structured, controlled repositories or scattered across personal storage. AI does not respect the distinction unless classification makes it explicit.
Who owns it. Records management assigns accountability for content. Without an owner, no one is authorised to decide whether AI should see it, summarise it, or expose it.
How long it stays. Retention periods define the natural lifecycle. Content beyond its retention window becomes training material for AI that no longer reflects how the organisation operates.
What happens when records management is broken and AI is turned on?
The AI does exactly what it was built to do. It reads the content available to the user, synthesises it, and delivers it in confident natural language.
The user gets an answer built from a retired policy, a superseded pricing model, a draft that was never finalised, and a document owned by someone who left three years ago. The answer is fluent, plausible, and wrong. The organisation absorbs the risk, not the AI vendor.
This is not a Copilot problem. It is a records problem that Copilot happens to expose.
What should organisations actually do about it?
Reframe the conversation, then sequence the work.
Reposition records management as an AI enabler, not a compliance function. This is a leadership decision that changes who owns the budget and who sits at the table.
Baseline the estate. Quantify how much content is beyond its retention window, how much is unclassified, and how much has no active owner.
Introduce lightweight classification that AI systems can actually read. Sensitivity labels, retention labels, and consistent metadata beat perfect taxonomy every time.
Retire content the organisation no longer needs. Disposition is the highest-leverage records activity for AI readiness, and the one most consistently deferred.
Only then, scale AI. Not because AI needs to wait, but because the ground it stands on has to hold.
Organisations that treat records management as an AI prerequisite move faster in the medium term than those who buy their way around it.
Frequently asked questions
- Is records management the same as information governance? Related but not identical. Information governance is the wider discipline covering policies, standards, and controls. Records management is the specific practice of classifying, retaining, and disposing of content according to defined rules. Both matter for AI, but records management is the more direct constraint.
- Can AI tools help clean up records management debt? Partially. AI can accelerate classification and identify candidates for disposition. It cannot replace human decisions about retention periods, ownership, or business value.
- Where should records management sit in the organisation? The strongest models place it under a Chief Data Officer or Head of Information Governance, with a direct line to the CIO and CISO. Sitting it inside legal or compliance alone tends to slow decisions.
- How long does this take? For a mature enterprise tenant, expect eighteen to twenty four months to reach a defensible baseline, with quick wins possible in the first quarter.
