Skip to main content
K4M2 AI

Service

AI strategy consulting

A strategy is a set of decisions, not a set of intentions. This engagement produces the decisions: which use cases are worth the capital, what each is expected to return, what to build and what to buy, which model and vendor, what has to be true about your data first, in what order, and what governance the result requires.

The business problem

Most organisations do not lack AI ambition. They lack a ranking. Work proceeds on whichever use case had the most convincing demonstration, and because no threshold was agreed, nothing can be concluded either way.

The result is a portfolio of activity that cannot be defended or cancelled. McKinsey's State of AI survey finds most organisations still experimenting rather than scaling, and the UK government's AI adoption research found only about half of AI-using firms felt ready to scale.

Strategy work is what converts activity into a sequence with reasons attached. We wrote the long version of this argument in most companies don't have an AI strategy, they have activity.

When this is useful

The recurring situations that bring people to this engagement.

A budget without a plan

Funding approved for AI and no agreed basis for allocating it.

Competing internal proposals

Several departments with candidate projects and no common measure to rank them.

A vendor decision pending

A platform proposal on the table and no independent view of the alternative.

Pilots that will not conclude

Work in flight that cannot be scaled or stopped because success was never defined.

A board asking for a roadmap

A requirement to show sequence, cost and risk rather than enthusiasm.

Regulatory exposure

Use cases that may fall under the EU AI Act, and no view yet of which.

What we will do

Eight pieces of work. They are sequential, and each one narrows the next.

  • Select the use cases. Identify the repeated decisions in the business, size each by volume, current cost and error rate, and discard the ones where a correct answer cannot be defined.
  • Estimate business impact. Establish a baseline from data you already hold, then express the expected effect in a unit that appears in a management account.
  • Decide build versus buy. Layer by layer rather than as one verdict, using the reasoning in build versus buy for enterprise AI.
  • Select model and vendor. An evaluation method first, then a recommendation, so the choice can be revisited when a better model ships.
  • Test data and infrastructure readiness. Where the inputs live, who owns them, what state they are actually in, and what it takes to reach them.
  • Sequence the implementations. Ranked by expected value net of the work required, with the dependencies made explicit.
  • Define governance requirements. Mapped to the NIST AI Risk Management Framework and, where relevant, the EU AI Act.
  • Write the roadmap. Short, with thresholds, owners and stopping conditions rather than a multi-year Gantt chart.

What you will receive

Five documents, all of them yours, none of them longer than they need to be.

  • A ranked use case shortlist with volume, baseline cost, expected effect and the main uncertainty for each.
  • A business case for the recommended first build, including the pessimistic scenario.
  • A build-versus-buy position by layer, with the reasoning stated.
  • A model and vendor evaluation method, plus the current recommendation.
  • A sequenced roadmap with thresholds, named owners, stopping conditions and a governance annex.
  • An explicit not-now list: what we considered, rejected, and why.

Expected business outcomes

What the strategy is meant to make possible.

Capital allocated on evidence

Spend directed at the workflow with the best expected return rather than the best demonstration.

Fewer stalled pilots

Work that can be concluded, because a threshold and a stopping condition exist before it starts.

A defensible vendor position

A choice you can explain to a board and revisit without a rebuild.

Shorter time to first value

The smallest useful build identified up front, rather than discovered after two quarters.

Governance ahead of exposure

Requirements known before the system is live, which is materially cheaper than after.

A reusable ranking

A method for assessing next quarter's ideas, not just this quarter's.

How the engagement runs

Four to eight weeks, with a decision point you can stop at.

01

Framing

Half a day with whoever owns the numbers. What is the business trying to change, and what is off the table.

02

Evidence gathering

Interviews with the people doing the work today, plus a direct read on samples of the actual data rather than the data model.

03

Sizing

Baselines and impact estimates built from your own systems, with the counterfactual subtracted.

04

Options and decisions

Build versus buy by layer, model and vendor method, governance requirements, and the readiness gaps that change the sequence.

05

Roadmap

A ranked sequence with thresholds, owners and stopping conditions, plus the not-now list.

06

Readout and handover

One session with the decision-makers. The documents are written to be used without us in the room.

Frequently asked questions

Is this different from a readiness assessment?

The assessment is fixed-scope and produces a shortlist. Strategy consulting is broader: it covers sequencing, build versus buy, vendor choice and governance across a portfolio rather than a single first build.

How long does it take?

Four to eight weeks for most organisations. Longer than that and the strategy work becomes the project it was supposed to scope.

Do you produce a roadmap document?

Yes, but a short one. A ranked sequence with thresholds and owners is more useful than a multi-year plan that assumes model capability stays still.

Will you tell us which vendor to use?

We will give you an evaluation method and a recommendation, with the reasoning. We hold no reseller agreements, so the recommendation carries no margin for us.

What if our data is not ready?

That is a normal finding and it changes the sequence rather than ending it. Part of the work is establishing what state the data is actually in.

Who needs to be involved?

Whoever owns the numbers you want to move, someone who can speak for data access and security, and the people who do the work today.

Relevant insights

Most companies don't have an AI strategy. They have activity. →Build versus buy for enterprise AI →How to calculate ROI for an AI initiative →What an AI readiness assessment should include →How to choose a model for production →

Start a conversation.

If you have a budget and a list of candidates, this is the engagement that turns them into a ranked sequence with reasons. Bring the list.

Start a conversationSee discovery