Service 01 Strategy

AI Strategy & Consultancy

A roadmap you can defend to your board, to your engineers, and to your own scepticism. Including the part where we tell you which ideas to drop.

Start a conversation

The problem with most AI roadmaps

They are ranked by excitement. The ideas at the top are the ones that demo well in a meeting, not the ones with the shortest path between your data and a number that matters. Six months later the pilot works beautifully on the sample and falls apart on Tuesday's real traffic.

A useful strategy starts from constraints, not from possibilities. What data actually exists, in what state, owned by whom. What your team can realistically maintain once we leave. Where a ten percent improvement would be worth something, and where it would be invisible.

What we actually do

  1. Map the ground truth

    Interviews with the people doing the work, plus a hard look at the data as it is rather than as the schema claims. This is usually where the first surprise turns up.

  2. Build the candidate list

    Every plausible use case, written in the same format so they can honestly be compared: expected effect, required data, integration cost, and what could make it fail.

  3. Score against effort and payoff

    Each candidate is placed on the same two axes, with our uncertainty shown rather than hidden. A confident guess and a wild one should not look identical on a slide.

  4. Kill most of them

    The valuable half of this work is subtractive. You leave with a short list and a written reason for every idea that didn't make it, so nobody relitigates it in three months.

  5. Sequence what's left

    An ordered plan with the dependencies made explicit, a proposed first experiment, and the success criteria agreed in advance.

the shortest report we can honestly write

In practice

Arbor Square · London · 2026

Arbor Square is a London-based independent research consultancy specialising in alternative assets. They have a proprietary method of interviewing investors, collecting and analysing information. On a single project, the heavy analysis took weeks.

They wanted to know whether AI could help. The material is strictly confidential, and the interviews had to stay human. We agreed: those conversations work because they are between people. The same exchange does not happen with a machine.

We mapped the rest of the method. The bottleneck was the analysis that follows the interviews. The first thing that map produced was a tool for that heavy lifting, designed around confidentiality from the start. The step that used to take two weeks now takes twenty minutes. A person still does the rest of the analysis.

Not sure it's worth a project yet?

That's the normal starting point. One conversation is usually enough to tell whether there's something here.

Start a conversation See how we work