The honest answer is: a little strategy, then tools, then more strategy. Companies that write a 40-page AI strategy before anyone has used the tools end up with a document based on guesses. Companies that hand out tools with no direction end up with scattered experiments and data risks. The order that works is a one-page direction, a hands-on phase, and then a proper strategy written from what you learned.
Why "strategy first" often fails
A strategy written before hands-on experience tends to be generic. It lists use cases copied from vendor brochures, assumes data that does not exist, and sets targets nobody can test. Leaders sign it off, then discover in month three that the priority use case is impractical. The document was not wrong in principle; it was written blind.
Why "tools first" often fails
The opposite mistake is just as common. A company buys licences for everyone, sends a launch email, and waits. A few enthusiasts get great results, most people try it twice and stop, and someone pastes a client contract into a free tool. Without a direction, tools create activity, not change.
The sequence that works
Phase 1: A one-page direction (week 1 to 2)
Before any rollout, agree four things on a single page:
- Why you are doing this (time back for skilled staff, faster customer response, better decisions).
- Where you will start (one or two teams, named).
- Rules for safe use (what data never goes into AI tools, which tools are approved).
- Who owns it (one accountable manager).
That is a strategy in the sense that matters: it gives direction without pretending to know everything.
Phase 2: Hands-on use (week 3 to 12)
Now the tools. Train a pilot group on real tasks, track time and quality, and collect what works and what does not. This is where you learn which use cases are real in your organisation, what your data can support, and where people struggle.
Phase 3: The real strategy (month 3 onwards)
With evidence in hand, write the fuller strategy: which processes to redesign, which tools to standardise on, what to integrate, how to train the wider organisation, how to measure progress. It will be shorter and far more credible than a strategy written on day one, because every line is grounded in something you tested.
A quick test: which do you need right now?
- If people are already using AI informally with no rules, you need Phase 1 urgently, mainly the safe-use rules.
- If you have rules but little use, you need Phase 2: training and a pilot.
- If you have scattered pilots that never scale, you need Phase 3: a strategy that chooses and standardises.
The leader's role in each phase
The person leading this does not need to be technical, but they do need to have used the tools themselves. It is very hard to set sensible direction, judge a pilot or write a credible strategy about something you have never tried. That is why our Leading AI Innovation & Organisational Transformation course starts with hands-on use before moving to rollout and planning.
If you want to see how strategy and tools relate conceptually, our earlier piece on the difference between AI tools and AI strategy covers the definitions; this article is about the order of work.
Frequently asked questions
Can a small business skip the strategy entirely?
Not the one-page version. Even a five-person firm needs agreed rules on what data goes into AI tools and someone responsible for checking results.
How long should the hands-on phase last before writing a full strategy?
Around two to three months is usually enough to see patterns across a few teams. Longer than six months without a strategy tends to produce scattered, unscalable experiments.
Who should write the AI strategy?
The manager who owns the rollout, with input from the teams that ran pilots, IT and whoever handles data protection. Consultants can help, but someone inside must own it.