Plan
AI strategy and roadmap
An AI strategy your board can approve and your organisation can execute: where AI changes how you compete, what you will build, buy and refuse, what it costs over three years, and what you will have to be able to do differently. Sequenced into quarters, with owners and money attached.
Duration
5–7 weeks
Built on
ISO/IEC 42001 · ISO/IEC 38500 · NIST AI RMF
Indicative price
€8,500–19,000 per engagement
Who this is for
CEO or owner
Has to say something about AI to a board, a bank or a client, and wants it to still be true in two years.
CIO or CTO
Is being handed AI on top of a delivery plan that is already full.
CFO
Is asked to fund something with no baseline and no end date.
Board or audit committee
Wants to see a plan rather than a pilot.
A strategy is what you will not do
Most AI strategies are a list of ambitions and a vendor's diagram. They read well and they decide nothing, which is why the organisation carries on doing whatever the most enthusiastic department was already doing. A strategy is useful in proportion to what it rules out: the use cases you will not fund this year, the data you will not move, the work you will not automate because the human judgment in it is the product you sell.
The second failure is planning AI as a technology programme when the constraint is almost never the technology. Models are a commodity and getting cheaper. What is scarce is clean data, stable process, people willing to work differently, and somebody accountable for a system after the project closes. A roadmap that sequences models rather than those four things gets re-planned within a year.
Where AI changes competitive position it does so in one of three places, and they are worth separating because they are funded and measured differently. It changes what you sell — an offer that was not possible before. It changes how you deliver — the same offer at a materially different cost or speed. Or it changes what you know — decisions made on evidence you did not previously have. Most organisations write a strategy for the first and quietly execute only the second.
Sovereignty is a planning constraint, not an afterthought. Where a workload may run, which providers are acceptable for which classes of data, and what happens if a provider changes its terms — those answers shape the roadmap, and in regulated sectors they remove options before cost does. We settle them at planning time, because discovering them at build time is expensive.
The output is a strategy short enough to be read and a roadmap detailed enough to be funded: quarters, owners, money, dependencies, and the decisions you are deferring recorded as deferred rather than forgotten.
How we do it
- 01
Ambition and constraint
1 week
What the business is trying to achieve over three years, and the constraints that are real — regulatory, contractual, data, capital and capacity. Ambition without the constraints is a wish list.
- 02
Current position
1 week
What already exists, including the pilots, the unsanctioned usage and the vendor commitments nobody wrote down. Where a readiness or value assessment has been run, we start from it rather than repeat it.
- 03
Strategic choices
1 week
Where AI changes what you sell, how you deliver or what you know. What you will build, what you will buy, and what you will deliberately not do — each with the argument recorded, so it can be revisited on evidence rather than on mood.
- 04
Operating and sovereignty constraints
3–5 days
Data classes against permitted locations and providers. What may leave the organisation, what may not, and the fallback if a provider changes terms.
- 05
Roadmap and investment case
1–1.5 weeks
Quarters, owners, dependencies, and a three-year cost envelope that separates build from run. Benefits expressed as things that will be measurable, with the baseline named.
- 06
Board readout and approval pack
3 days
The version that gets approved: short, specific, trade-offs visible, and the decisions needing a board signature isolated from the ones that do not.
Named artefacts
What you receive
- AI strategy document — the choices, and the argument for each
- Explicit not-doing list, with the reasoning recorded
- Three-year roadmap by quarter, with owners and dependencies
- Investment case separating build cost from run cost
- Benefit baselines — what will be measured, and what it is today
- Data classification and permitted-location map for AI workloads
- Build, buy or partner position per capability
- Capability plan — what the organisation must be able to do differently, and by when
- Programme risk and dependency register
- Board approval pack
What we need from you
- Executive time, properly. Three to four working sessions with people who can commit budget — not delegates.
- The commercial plan the AI strategy is meant to serve. An AI strategy that does not reference the business plan is a technology plan.
- Honesty about capacity. Most roadmaps fail on the delivery team's existing load rather than on the AI.
- Whatever has already been committed — vendor agreements, running pilots, promises made to clients.
- Someone with the authority to write the not-doing list and defend it afterwards.
What changes
- 01A strategy that says what you will not do, which is what makes the first quarter deliverable.
- 02Investment sized over three years, with run cost visible rather than discovered.
- 03Benefits tied to baselines that actually exist, so the programme can be judged.
- 04Data and sovereignty constraints settled before they become build-time surprises.
- 05A plan your board can approve in one sitting.
What it costs
€8,500–19,000 per engagement
All prices exclude VAT.
Questions
Do we need the assessments first?
Not formally, but the strategy is only as good as its inputs. Where no readiness or value assessment has been run, the first fortnight does a compressed version of both and the strategy carries more assumptions. If you have recent internal work, we will use it rather than redo it.
Will you tell us to slow down?
Sometimes, and it is usually about sequence rather than ambition — the ambition survives, the first quarter changes. The most common pattern is an organisation planning six use cases when one unresolved data problem gates four of them.
How long before it needs revisiting?
Review quarterly against the roadmap, and rewrite the strategy when an assumption breaks rather than on a calendar. In this market the assumptions that break first are usually about cost and about what the models can do unaided.
Can you deliver what you plan?
Yes, and we will say so plainly rather than write a roadmap only we can execute. Where a workstream is better done by your own team or another supplier, the roadmap says so. A plan written to generate follow-on work is easy to spot and expensive to own.
What does it cost?
€8,500–19,000 per engagement. The range reflects the size of the organisation and how many functions are in scope.
Leave with your top three risks documented
Thirty minutes with a senior practitioner. No slideware, no sales engineer.