Engagement lead
named in your proposal
Owns the figure with you. Writes the strategy, sits in the room where it is defended, and is still there at the end.
Turn AI into
the advantage
only you own.
We rebuild how you quote, so the quote you lost last month prices the one you send tomorrow. We stay until your margin moves.
Decisions worth rebuilding first / what to quote / who to chase this week / what to keep in stock / which claim can be paid without a person reading it / when to reorder / which customer is about to leave /
01 What happenedthe pilot that moved nothing
Almost everyone has bought AI by now.
AI arrives on a monthly invoice, switched on inside software you already pay for. Finance signs it, not engineering.
The demonstration goes well. The pilot ships, and the budget renews. Then the quarter closes and nobody can point to the line it moved.
That is what MIT counted. Not pilots that broke, pilots that ran. Nobody wrote the line down before it started. Nobody owned it after.
Gartner's cancellations come one budget later. A system nobody can show a result for does not survive the review.
two published figures
95%
of the enterprise generative AI pilots MIT studied produced no measurable change in revenue or cost.
MIT NANDA and MIT Media Lab, The GenAI Divide: State of AI in Business 2025,
July 2025. 52 structured interviews, 153 surveys of senior leaders, 300 publicly disclosed
initiatives, gathered January to June 2025.
40%+
of agentic AI projects Gartner expects to be cancelled before the end of 2027, on cost, unclear value and inadequate risk controls.
Gartner press release, 25 June 2025. The same release estimates that of the thousands of vendors now using the word agentic, roughly 130 are real.
02 Why it happenedand why a better model will not fix it
They “do not retain feedback, adapt to context, or improve over time.”
MIT's own explanation for the 95 per cent
It does not remember. Ask it the same question every day for a month and it learns nothing from any of them. On the thirty first day it begins again from nothing.
A better model answers faster. It still begins from nothing. The memory is the part that has to be built, and it is the only part that can be yours.
03 What is yoursand what is only rented
Whichever model wins, you will be renting it.
The frontier model reaches you through the same interface your competitors use, at the same price, on the same day. Replace it whenever something better arrives.
What cannot be replaced is the record underneath it. What was decided, what it was decided on, who overruled it, what came of it. No supplier can rent you that.
Your people keep the judgement calls. The system takes the repetitive ones.
04 How the work runsevery turn, sharper
Four steps. Skip the third and you never get the fourth.
Intent, Execute, Measure, Learn. You name the figure the system is meant to move. The system does the work. The data records what actually changed. The next turn starts from that record rather than from nothing. We call that cycle the loop.
One pass through it is a project. The value is in the repetition. The system that prices your quotations this month holds every quotation you lost last month, and the reason each was lost. So it quotes better than it did. That is where the compounding comes from: the record, not the model.
INTENT the number we agreed to moveset
EXECUTE the system does the workrunning
MEASURE what actually changedrecorded
LEARN the next turn starts sharper+1
05 The frameworkours, and the spine of the work
Three sides decide the outcome. Three layers connect them.
The loop runs inside an organisation, and the organisation decides whether it can. The AI Transformation Triangle is our own framework for reading that. People, processes and platforms are the three sides. Each one can be built in house or brought in on its own, which is why most organisations end up strong on one and thin on the other two.
What decides the outcome is not any single side. It is where two of them meet. Those three meeting points are the layers in the middle, and together they are the operating model. That is the part we are hired to rebuild, and it is where the loop either runs or stalls.
06 Where we come inthree ways to begin
Three ways to join the loop.
The first tells you whether the other two are worth building. Most organisations start there, and a fair number stop there.
01 Transformation
We redesign how the organisation decides.
We set the strategy and rebuild the decisions themselves, beginning where AI changes what a decision costs to make.
what you get
A strategy you can run, tied to the figure it is meant to move.
02 Implementation
We build it and put it in production.
The products, data foundations and processes that let those decisions happen daily, at volume, inside the systems your people already use.
what you get
A working system in production, yours from the first commit.
03 Optimisation
We instrument what it learned.
Not merely what the system did. A lost quotation with a recorded reason teaches it considerably more than a won one, which is why that field is never optional.
what you get
An advantage that widens cycle by cycle, and is hard to copy for a competitor renting the same model.
07 What you are left withsix things, all of them yours
What is actually handed over at the end.
Six things, each yours from the first commit. The last two decide whether the system is still improving a year after we have gone.
01
The decision map
Every repeating decision, what each costs to make today, and which three are worth rebuilding first.
02
The data foundation
The baseline recorded before anything ships, and the plumbing that keeps it accurate.
03
The systems themselves
Running in production, in your environment, under your own keys.
04
The measurement
The named line, the person accountable, and the reading either side. If it cannot be read, it is not finished.
05
The runbooks
How it works, how it fails, and how to change it without us.
06
The handover
Your team, running it. We say when it is finished rather than letting an arrangement drift on.
All six of these outlive us. Which is rather the point of writing them down.
08 What we buildtwelve parts, one engagement
Our services, and what each one is for.
Most builds use three or four of these, not twelve.
Rebuilding how you quote uses four of them: strategy, the data under your prices, the screen your sales team quotes from, and a reading of margin either side. The other eight stay shut.
09 The technologychosen for the job, not for the brochure
The technology we build with.
The model arrives the same way for everyone. The engineering underneath is ours to get right, and it is what keeps a system running in production.
AnthropicOpenAIMistralGeminiLlamaHugging FaceOllamaNVIDIAHermes Agent
Google CloudKubernetesDockerTerraformPostgreSQLSupabaseElevenLabsHiggsfieldSnowflakeDatabricksSparkAirflowPythonPyTorchLangChainFastAPIRedis
10 What it replacesyour arithmetic, not ours
Most of this is already on the budget, annually.
The figures are yours. The arithmetic is worth doing before you speak to anybody, ourselves included.
what it costs at present
what you own instead
The strategy deck
A firm charges once for the thinking and leaves the building to somebody else. The value
leaks away in between.
The same thinking, from the people who then stay and build it.
The pilot that plateaued
Still running, still budgeted, still impressive in a demonstration, and still invisible in the quarter.
A system with a named line and a reading either side, so it earns its budget
or is switched off deliberately.
The platform subscription
Priced per seat, repriced at renewal, and holding your data. Whatever it learned leaves
when it does.
A single build in your own environment. What it learned about you stays when we go.
The extra hire
A salary each year, for the repetitive half of a job.
The repetitive half handled, and your own people on the half that needs
judgement.
11 Who you work withtwo offices, one team on your build
The people who write the strategy build the system.
Where we work from
Principal engineer
named in your proposal
Owns the architecture and writes the code. The same person through discovery, build and handover.
Data lead
named in your proposal
Establishes the baseline before anything ships, and the definitions everyone in the room can agree on.
Delivery lead
named in your proposal
Runs the phases, the exits and the handover, and is measured on your own team not needing us.
You keep the work whatever happens.
Fixed scope, fixed price. You leave with the figure, what moving it is worth,
and the plan. Hand that plan to another firm and it still works.
We will tell you not to build.
Before anything is built we agree the figure it has to move and how it will be
read. If we cannot find one worth the work, we say so.
You own the system, in writing.
Yours from the first commit, in the contract. Every phase has its own scope,
price and exit.
We stay until the figure moves.
Not until launch, and not until handover. Until the figure we agreed has moved
and your own team has changed the system without asking us.
12 Questionsso the page can be forwarded
The questions that come up on the call.
Answered here so this page can be forwarded to whoever else has to sign.
Our data is a mess. Is it too early?
Sometimes, and we will say so. Where the data already exists as a by-product
of work being done, untidy is workable. If somebody must be hired to produce it, that is the
first project.
What happens when the model changes?
You replace it. The model is a dependency, not the system. What remains is
the record of what has been learned. A better model arriving is a good day.
Who owns the code and the data?
You do, from the first commit, in the contract. Code, data, prompts, model
configuration, runbooks and documentation, in your own environment under your keys.
How long before a figure moves?
Where the baseline was already being measured, movement is usually visible
within a quarter of going live. Where it was not, measuring it is the first phase. You get a
range in writing before committing.
What does the first piece of work cost?
A fixed price for a fixed scope, quoted on the call once we know which
decision you want examined. It is the smallest commitment we offer, on purpose.
13 How to begina conversation, first
Half an hour on your figure. No invoice.
Your message goes to a person, who answers within two working days. Then a call. Then, if you want it, a price. Stopping after the call is normal.
First, the call. Bring the figure you are judged on.
No slides needed. Messy spreadsheets are fine. You leave knowing whether a system could move it, what that would cost, and whether we would tell you not to.
Then, if it is worth building, a price before we start.
Fixed scope, fixed price, both on paper before anyone writes code. What we build is yours to keep and run, whether or not you hire us again.