Executive AI strategy
A session with your board or leadership that ends in a decision you can defend: what AI should do for your company first, on which stack, what it costs to run, and who is responsible.
This is for you if
- The board has asked for the AI strategy, and what you have is licences and a few experiments.
- Every vendor has shown you a feature, and nobody has shown you where AI pays in your company.
- You want to decide once, defend it, and not revisit it every time a new model is announced.
Three roads, and which fits which job
Most companies end up on all three. The mistake is to take one road for every job.
| The licences you have | The features in your systems | Knowledge workers built for you | |
|---|---|---|---|
| What it is | Copilot, Gemini or Claude, set up with permissions, connectors and guardrails | The AI your ERP, CRM and accounting system ship with | Agents built for your own jobs, on your own data, in your own systems |
| Fits when | People write, summarise and search their own documents | The whole job happens inside one system, and the vendor built the feature for it | The job crosses systems, follows your rules, and must be checked |
| Who runs it | Each person uses it as an assistant | The system runs it, the way the vendor set it up | The operator, who instructs, inspects and signs off |
| What you own | Nothing beyond the licence | Nothing beyond the licence, for the built-in features | The repository, the agent and the builder agent; their tools and data stay in your cloud |
| What it relies on | The documents and mail each person can open, and the sources you connect | The data in that one system | The data foundation, your dictionary and your access controls |
What you leave with
- The result
- A decision you can defend. We write it down in your words.
- A list
- It covers what a strategy must decide: where AI pays, what it relies on, what the law expects, and what changes for your people.
- A test
- We test which of the four basics an idea improves: customers, processes, decisions or innovation. If it improves none of them, skip it.
- An honest answer
- If the right move for your company is to wait, we say so.
What we do, what you do
| We | You |
|---|---|
| Bring the list of what a strategy must decide. | Bring the people who decide. |
| Work through it with your systems, your numbers and your jobs as the examples. | Say what the hours and the errors cost today. |
| Name the three roads for each job: the licences you have, the features in your systems, knowledge workers built for you. | Decide, and own the decision. |
Worth knowing
The first question boards ask is the wrong one. "Which AI is best?" will not help you, because the model you pick this year is not the one you run next year. A strategy is everything that stays true when the model changes.
Frank answers
- Is this a strategy document?
- No. It is a session that ends in a written decision. A strategy that no running job follows is only a slide.
- Do we need to have chosen a model or a vendor first?
- No, and you should not choose one first. Models change, and the session decides what stays true when they change.
- What if the honest answer is that we should wait?
- Then we say so. Some companies should put their data in order first and nothing else; some should buy nothing this year.
- Who should be in the room?
- The people who decide: the board or the leadership, the person who owns the numbers, and whoever controls access to your systems.