Businesses need a better account of themselves

Giving AI access to what an organisation has written is not the same as helping it understand what the organisation knows.

A new Alteryx study reported by ITPro has put a number against a problem I’ve been thinking about for a while. Seventy-seven per cent of the IT leaders surveyed said business context was critical to producing accurate and relevant AI output, while 53% said their organisation struggled to incorporate it into AI systems and workflows.

The original research announcement defines context as the rules, definitions and operational knowledge that shape how an organisation works. Much of it remains scattered across spreadsheets, documentation, email threads and the expertise of the people closest to the work.

That makes the problem sound largely technical: connect the systems, retrieve the documents and give the model access. Some of it certainly is. But access only solves the part where the information was difficult to find. It does not solve the harder question of whether the organisation has ever worked out what that information means.

An organisation can possess information without possessing an account of itself

A company could connect thousands of documents to an AI system and still struggle with fairly basic questions. Which evidence deserves authority? What have customers told us consistently, and where have they disagreed? Which assumptions are now out of date? What was tried, what changed, and what did the organisation learn from it? What does it believe strongly enough to guide a decision or make a public argument?

The answers rarely sit inside one definitive source. They have to be developed across research, experience, decisions, contradictions and changes of mind. A search system can retrieve the version that was written down. It cannot establish, without further work, whether that version was well supported, superseded three months later or quietly rejected by everybody who actually understood the problem.

This is why I think businesses have an interpretation problem as well as a context problem.

The distinction matters because “give the AI more context” risks becoming another technical answer to an organisational failure. The business may already have the relevant report, transcript, customer quote and decision record. What it lacks is a maintained account of how those pieces relate: which matter, which conflict, which remain uncertain and which should influence the next piece of work.

That is an editorial function, although not one confined to content production. It involves establishing significance, preserving qualifications and developing an argument without pretending every source says the same thing. It sits between stored knowledge and whatever a person or AI system is expected to do with it.

This is the territory where Editorial Intelligence may have a useful role. I would not extend that claim to every form of business context. Tax rules, financial controls and supply-chain thresholds need domain experts, formal governance and deterministic systems. Editorial judgement is not a substitute for any of them.

The more defensible territory is the material that organisations rely on but cannot reduce so neatly: customer evidence, market interpretation, institutional memory, strategic arguments and narrative choices. These are less about whether a piece of information exists than about what the organisation is entitled to conclude from it.

Context is not something you upload once

There is another problem with the language of feeding context into AI. It makes context sound stable.

It isn’t. Research dates, markets change and terminology shifts. A reasonable assumption hardens into company folklore. A customer quote loses its qualifiers as it gets copied from a transcript into a presentation and from the presentation into a brief. A claim that began as a working hypothesis is repeated often enough to acquire the status of fact.

Connecting an AI system to that material does not correct the problem. It can retrieve yesterday’s assumption faster and present it with tomorrow’s confidence.

That is why I described context as capital. Its value can compound because old research, interviews and decisions can contribute to new work rather than disappearing after their first use. But capital also depreciates. Accumulated knowledge becomes an advantage only if the organisation can challenge, qualify and update it. Otherwise memory turns into automated inertia.

This creates a less glamorous requirement than buying a better model. Useful context has to be investigated, interpreted, structured and maintained. Somebody has to notice when the evidence no longer supports the story, preserve the inconvenient interview that did not fit the original argument and distinguish what the organisation knows from what it has merely said many times.

AI can make that maintenance easier. It can compare material at a scale a person could not sustain, surface contradictions and bring an old observation back when it becomes newly relevant. But it cannot remove the need to decide whether the connection is meaningful. The same capability that helps an organisation remember can also help it reproduce an old mistake more consistently.

The organisation should benefit before the model does

The Alteryx findings do not prove Editorial Intelligence. This is vendor-sponsored research among IT leaders, and “business context” is broad enough to contain several problems that have little to do with editorial work. It would be convenient for me if a widely reported enterprise AI problem turned out to validate the discipline I’m building, which is a good reason not to make that leap too quickly.

What the findings do support is the problem territory. Organisations are discovering that the quality of AI output depends on more than the quality of the model or the volume of material connected to it. It depends on whether they have made their own knowledge usable.

A possible role for Editorial Intelligence is to help turn scattered evidence and human judgement into usable context for people and AI. The order matters. The organisation should benefit before the model does. People should be able to find the strongest evidence, understand previous decisions, see where accounts differ and build on earlier thinking. AI then makes that maintained knowledge easier to retrieve, compare and apply.

The aim is not to document everything or construct an immaculate corporate second brain. That would create another system people have to feed and eventually learn to ignore. The test is narrower: does what the organisation has preserved improve the next piece of thinking?

If it does, the business has more than a collection of documents. It has a better account of what it knows, what it does not know, what it has learned and what it may need to reconsider.

If it cannot produce that account for itself, giving the archive to an AI system is unlikely to produce one on its behalf.