Corporate AI Strategy
Work out what AI is for before you build it
An assessment, a shortlist of use cases worth the money, and a roadmap your board can read. Independent of who ends up building it.
What AI strategy means here
Most companies do not have an AI problem. They have a prioritisation problem wearing an AI costume. We start by finding the work that is genuinely expensive, then ask which parts of it a model can actually take on.
What you get
- An honest read on whether your data is ready, before anyone promises a result
- A shortlist of use cases ranked by value and by effort, not by how good the demo looks
- The governance and policy work procurement and legal will ask for anyway
- A costed roadmap with a first phase small enough to prove or kill in a quarter
- A written recommendation, including the option of doing nothing
Strategy, not the build
This page is about the decision. If you already know what you want built and you are looking for the team to build it, that is a different conversation and it lives on the platforms and AI page.
We are happy to do both, and we are equally happy to hand a roadmap to your in-house team or to an incumbent vendor. Advice you can only act on by hiring the person who gave it to you is not advice. If you want the longer version of how we think about this, we wrote it up in the gap no one is talking about.
Capabilities
Assessment
What you actually have to work with.
- Data inventory: where it lives, who owns it, what condition it is in
- Systems and integration map across ERP, CRM and the spreadsheets nobody admits to
- Post-mortem on pilots that stalled, and why
- Team capability and appetite, honestly recorded
- Security, privacy and residency constraints up front
Direction
The shortlist, ranked and costed.
- Use-case discovery workshops with the people doing the work
- Value and effort scoring, with the assumptions written down
- Build, buy or wait, decided per use case
- Model and platform selection on your constraints
- A phased roadmap with a first slice you can kill cheaply
Governance
The part that gets skipped and then bites.
- Acceptable-use and disclosure policy your staff will actually follow
- Human review and escalation paths for anything customer-facing
- Evaluation harness so quality is measured, not vibed
- Cost controls and monitoring before the bill is a surprise
- A register of what was decided and why, for the next team
Our Approach
01
Frame the problem
Two weeks of interviews across the people who own the work, not just the people who own the budget. We are looking for expensive, repetitive and well-documented.
02
Audit the data
The answer is usually decided here. A model cannot retrieve what nobody maintains, so we go and look rather than take the systems diagram at face value.
03
Score the candidates
Every use case gets a value estimate, an effort estimate and the assumptions behind both, so you can argue with the reasoning rather than the conclusion.
04
Pick the first slice
One use case, scoped so it proves or disproves itself inside a quarter. Small enough to cancel without a post-mortem meeting.
05
Write the governance
Policy, review paths and evaluation, drafted while the detail is fresh rather than retrofitted the week before launch.
06
Hand it over
A roadmap, a costing and a recommendation, in a document that survives without us in the room. Yours to run with whoever you like.
What you get
Everything below is a document or a decision you own outright. There is no part of this engagement that only works if we do the build.
- AI readiness assessment
- Data and systems inventory
- Ranked use-case shortlist with costings
- Build, buy or wait recommendation per use case
- Phased roadmap with kill criteria
- Acceptable-use and disclosure policy
- Evaluation and quality-measurement plan
- Board-ready summary deck
Where would AI actually pay?Let's find out properly.
Start the conversation
Questions, answered.
- How is this different from your AI platforms service?
- This engagement decides what to build and whether to build it at all. The platforms service is the build itself. Plenty of clients do the first and then run the second somewhere else, and that is a legitimate outcome we price for.
- We already tried an AI tool and it went nowhere. Is that a problem?
- It is the most useful thing you can bring us. A pilot that stalled tells us more about your data, your approvals and your appetite than any workshop will, and the post-mortem is usually the fastest part of the assessment.
- Do you insist on a particular model or vendor?
- No. We have shipped on Amazon Bedrock, Azure OpenAI and Google Vertex, and the right answer depends on where your data already lives and what your procurement team will sign. Vendor choice is an output of this work, not an input.
- How long does it take?
- Most assessments run four to six weeks. Longer if the data has to be mapped across several systems first, which is common and is worth knowing before a build is scoped rather than after.
- What if the answer is that we should not do this?
- Then that is the deliverable, in writing, with the reasoning. It is a cheaper outcome than a build that quietly gets switched off in year two, and it has happened.
Keep exploring