Access is no longer the hardest part

Modern AI tools are widely available. OECD data also shows firm-level adoption continuing to increase, although adoption remains uneven by company size and sector.

That does not mean implementation is solved.

A business can have access to a capable model and still fail to produce repeatable value.

The difficult part is everything around the model

The process

If the underlying workflow is unclear, AI usually accelerates confusion rather than fixing it.

The data

Useful output depends on relevant, accurate and appropriately governed information.

The integration

A model response in a browser tab is not the same as a working business process. Value appears when output reaches the right person or system at the right time.

The exceptions

Real operations contain missing information, unusual cases and contradictory inputs. A production workflow needs a way to handle those cases.

The accountability

Someone still needs to own what happens when the output is wrong.

Why demonstrations are easier than operations

A demonstration normally has:

  • a clean input
  • a clear prompt
  • a cooperative example
  • no edge cases
  • no integration problem
  • no customer or compliance consequence

Daily business work has all of those complications.

That is why an impressive demo should not automatically be treated as evidence that a process is ready for autonomous operation.

Where businesses should start

Useful early candidates often have four characteristics:

  • repetitive
  • high enough volume to matter
  • relatively structured inputs
  • low to moderate consequence if an error occurs

Examples can include internal summarisation, first-draft generation, document classification, routing and knowledge retrieval.

Higher-risk use cases should introduce stronger controls and human review.

The economic test

AI should not be implemented because the tool is fashionable.

Ask:

  • What problem is being solved?
  • How much time or cost does the current process consume?
  • What error rate is acceptable?
  • What review remains necessary?
  • What software and integration cost is introduced?
  • Who owns the workflow after launch?
  • How will value be measured?

If those questions cannot be answered, the organisation probably has an experiment rather than an operating solution.

Sources used

  • OECD, The Adoption of Artificial Intelligence in Firms, 2025: https://www.oecd.org/en/publications/the-adoption-of-artificial-intelligence-in-firms_f9ef33c3-en.html
  • OECD, AI use by individuals surges as firm adoption expands, 2026: https://www.oecd.org/en/about/news/announcements/2026/01/ai-use-by-individuals-surges-across-the-oecd-as-adoption-by-firms-continues-to-expand.html
  • International Labour Organization, The Impact of GenAI on Jobs, Productivity and Work Organization, 2026: https://www.ilo.org/publications/impact-genai-jobs-productivity-and-work-organization-review-empirical