How is this different from a consultancy’s AI strategy report?

The unit of analysis is a work task, not a market. We interview the people doing the work, score scenarios on frequency, duration and cost of error, and name the ones not worth doing. The report is meant to end an internal argument, not to open one.

Do you then build what you recommend?

We can, through the agent delivery engagement, and clients often ask us to. The diagnosis is priced and delivered separately so the recommendation is not an order form.

What if the honest answer is "do not build anything"?

Then that is the report. Several scenarios in every inventory are better solved by a rule, an existing system or a process change, and we write that down.

AI advisory and technology selection

Do the arithmetic before writing code: inventory the use cases, decide which are worth automating, choose models and tooling, and put cost and risk boundaries in writing.

Most money lost on AI projects is not lost in development — it is lost building the wrong use case. We start with a diagnosis: rank candidate scenarios by frequency × time per instance × cost of error, cut the ones that rules or existing software already solve, and only then discuss which model, build or buy, how much data, and how long before anything shows. The output is a written report you can take into a board meeting, not a concept deck.

Who it is for

Leadership has committed budget to AI but the organisation disagrees on what to build first
You are holding several vendor proposals and need a third party who does not sell development to compare them
A first proof of concept stalled and you need to know where it stalled

Not a fit:

Anyone who wants an industry report. The diagnosis depends on your real process data and interviews.

How we run it

Inventory the scenarios. Interview five to eight frontline roles and record where their time actually goes.
Rank and cut. Score against a quantifiable basis and state explicitly which scenarios we recommend against, and why.
Select the technology. Model, deployment (cloud or private), tool chain, data preparation effort — each with the trade-off written out.
Cost and risk. One-off investment, monthly run cost, compliance and data risk, and the cost of exit if it fails.
Roadmap. What to build first, what the acceptance criteria are, and the conditions under which to stop.

What you get

Scenario inventory and priority ranking, including rejected scenarios with reasons
Selection report: model, deployment and tool chain compared and recommended
Cost model and risk register
Phased roadmap with acceptance criteria per phase

Related cases

Eli Lilly

AI FoundationScope delivered

Novartis

Digital human marketingScope delivered

Questions we get asked

How is this different from a consultancy’s AI strategy report?: The unit of analysis is a work task, not a market. We interview the people doing the work, score scenarios on frequency, duration and cost of error, and name the ones not worth doing. The report is meant to end an internal argument, not to open one.
Do you then build what you recommend?: We can, through the agent delivery engagement, and clients often ask us to. The diagnosis is priced and delivered separately so the recommendation is not an order form.
What if the honest answer is "do not build anything"?: Then that is the report. Several scenarios in every inventory are better solved by a rule, an existing system or a process change, and we write that down.