The honest map
What it actually takes to get value from AI.
Almost everyone's personal experience of AI is column one. Almost every business case is column three. People reason from the first and are surprised by the third — and the gap between them is not more effort. It is a different kind of work.
| The question a business person asks | 1 · Ask it somethinga simple prompt | 2 · Run a repeatable taskbasic value creation, simple process | 3 · Support real value creationa complex use case in your business |
|---|---|---|---|
| What it looks like | “Draft a reply to this supplier email.” | “Screen every incoming CV against this job spec.” | “Guide 200 suppliers through a development programme — each at a different stage.” |
| What world must it understand? | Nothing beyond your sentence. | The task's boundaries, and what a good answer looks like. | Your operating environment — the parties, the constraints, the decision rights, where each person stands. Rarely written down anywhere. |
| What must it do, and for whom? | Answer you. Once. | One task, one shape, one kind of user. | Different things for different people at different stages — and it must work out which before it acts. |
| What must it know — and whose knowledge is it? | Whatever the model learned from the internet. | A few of your documents, attached. | Your body of knowledge, organised so the right small fraction surfaces at the right moment. Most organisations' knowledge is not in that state yet. |
| What goes in front of it, every single time? | Your sentence. | Your sentence, a template, a couple of documents. | A designed parts list — dozens of distinct pieces, each with an owner, a shelf life, and rules on where it may be stored. |
| What does it remember? | Nothing. | Nothing between runs. | Within the task, and across months. Two different things, both built deliberately. |
| What stops it going wrong? | You do. You are the check. | A human review step. | Limits it must not cross, checks it cannot skip, and a record of why it said what it said. |
| What has to be built and connected? | Nothing. It already exists. | A template, and somewhere to put the output. | Finding the right material, working out where someone stands, capturing evidence, showing its reasoning — a dozen parts, each able to fail on its own. |
| Who signs it off? | You. | You, or a team lead. | Unclear until you decide. Who approves the knowledge? Who accepts its behaviour? Who owns it at 3am? Different people, different departments — and this is where most projects stall. |
| What happens when it is wrong? | You rephrase. | You catch it in review. | You have to find which part misled it. Without a parts list, “the AI was wrong” is untraceable. |
| What happens when your business changes? | Nothing to maintain. | Update the template. | Every piece has a shelf life and an owner — or it quietly rots while still sounding confident. |
| What it takes | Minutes. Anyone. | Days to weeks. A capable team. | Months — and a discipline most organisations have never had to apply to their own knowledge. |
Every one of these is solvable. The failures come from not knowing they were on the list.
These three levels are illustrative bands, not a standard. They exist to make the shape of the jump visible — not to grade anyone.