AI Strategy Engagement
What a four to six week AI strategy engagement produces: a ranked list of where AI pays, frozen baselines, and a written refusal list.
| How it is bought | Fixed fee |
| Duration | 3 weeks |
| Written for | Leadership teams that need a defensible plan and a sequence, not a list of ideas |
An AI strategy that is a document rather than a sequence of shipped systems is a strategy nobody can execute or measure.
What you receive
Named artifacts, not activities. If one of these is not delivered, the engagement is not complete.
- A sequenced roadmap: what ships first, what it depends on, and what is deliberately deferred
- The target architecture, with the trade-off each choice commits you to
- A governance position mapped to NIST AI RMF and, where relevant, the EU AI Act dates that apply to you
- KPI definitions for readiness, operating and inference readiness, with the baseline for each
- A build-versus-buy call per capability, with the reasoning written down
How it runs
- Week 1: current state, constraints, and the decisions already made that cannot be unmade.
- Week 2: options and trade-offs, tested against your actual estate rather than the general case.
- Week 3: the roadmap, the governance position and the readout.
What this deliberately does not include
Published with the same prominence as the deliverables. A scope with no stated exclusions is a scope that will be argued about later, and the argument always happens at the worst possible moment.
- Implementation. This produces the plan and the sequence.
- A procurement exercise or vendor selection.
- Anything that requires access we cannot get inside three weeks.
How you know it is finished
Somebody who was not in the room can read the roadmap and understand what ships first and why.
What you need ready
Your current architecture documentation, the constraints that are genuinely fixed, and access to the people who would deliver it.
Every engagement starts with a written scope confirmation before any invoice is raised. Nothing here is a click-to-buy: we confirm what you need, agree the scope in writing, then invoice. If the scoping conversation shows that a smaller engagement, or none at all, is the right answer, we say so before anything is signed.
Request this engagement Talk it through first See the free stack advisor
Most AI strategy work produces a deck. This produces a ranked list of processes, a measured baseline for each, and a written record of what we recommended you not automate.
The distinction matters because the common failure is not choosing badly. It is never choosing at all: a year of pilots, an adoption dashboard, and nobody able to say whether anything got cheaper.
The question the engagement answers#
Not "should we use AI". That question has no answer, invites opinion, and is settled by whoever is most senior in the room.
Instead: which twenty processes consume the most human hours, and which of those can be automated or augmented? That has an answer, the answer is a list, and a list can be wrong and therefore corrected.
What happens#
| Week | Work |
|---|---|
| 1 | Process inventory. Hours measured from system logs and observation, not from asking. People underestimate routine work and overestimate exceptional work, consistently |
| 2 | Ranking by annual hours, then scoring each candidate on volume, structure, reversibility and tolerance for error |
| 3 | Baselines for the top candidates, defined so they can be measured again afterwards |
| 4 | Sequencing, costing, and the refusal list |
| 5 to 6 | Optional: a costed implementation plan for the first two changes |
What you receive#
- The ranked inventory, with hours attached, so the argument about priorities is settled by data
- A scored shortlist, with the constraining axis named for each candidate
- Frozen baselines for the top candidates, which is what makes any later benefit claim checkable
- A written refusal list: what we recommend you do not automate, and the condition that would change that
- A sequenced plan with costs and dependencies
The scoring rule we will not bend#
A process scoring low on reversibility or tolerance for error does not go to full automation, whatever its volume. It may go to augmentation, where a person still decides.
That rule is the difference between a transformation programme and an incident, and we would rather lose the argument now than be right about it later.
What we will not do#
- Recommend automating something because it is technically possible. The test is reversibility and accountability, not capability.
- Produce a benefit forecast without a measured baseline. A forecast with no baseline cannot be checked, which is usually why it is offered.
- Claim a saving that is really a reallocation. If four hours of work becomes twenty minutes of work plus three hours of supervision, nothing was saved.
How we work, in public#
The operating model we use internally is published in full: the KPIs, the procedures, the templates and what we deliberately do not let an agent decide.
You can read exactly how we would run this before speaking to anyone. That is deliberate: it is easier to judge a method you can inspect than a capability you are asked to take on trust.
Getting started#
The inventory is the part with the most value and the least glamour, and it is where we would begin. Get in touch with which function is most expensive to run, and we will tell you whether this is worth doing before you commit to it.
What else is coming for AI Strategy
The engagement Ready
What it produces and how it runs.