There are two halves to this. Where we work — the four surfaces every pound in your firm passes through. And how we build — Institutional Context Architecture, the four layers that make your firm's operating knowledge readable by a machine.
You could take this page to another supplier and brief them from it. We would rather be judged on execution than on holding a secret.
A firm deploying capital is a sequence of decisions, each one gated on information arriving in time. Drawing the sequence is the first thing we do, before AI is discussed at all, because most of what looks like an AI problem turns out to be an information-arrival problem — and you cannot tell which is which until the machine is on paper.
What you have raised, on what terms, and what you have promised to do with it by when.
Raise, JV and LP structure, lender relationships, deployment planning, investment committee papers, quarterly reporting to whoever gave you the money.
Where it usually hurts: the position gets rebuilt by hand every quarter, and idle capital is only visible after the fact.
Everything between a lead arriving and a defensible number with a decision attached to it.
Origination and screening, the buying box, modelling, comparables, refurb estimating, legal and technical diligence, licensing and planning checks, red lines.
Where it usually hurts: capacity is set by one senior person's calendar, and the finding that kills a deal arrives after the legal spend.
From offer accepted to money actually moving, in both directions.
Offer, legals, the lender pack, outstanding conditions, valuation booking, certification, drawdown.
Where it usually hurts: nobody owns the chase, so weeks are lost waiting on a document rather than a decision.
Everything after ownership, up to and including getting the capital back out.
Refurb specification and delivery, contractors and programme, letting, arrears, voids, compliance and licensing, maintenance, reporting, refinance, disposal.
Where it usually hurts: compliance sits in someone's memory, and a programme slipping is only visible at drawdown.
Point a general model at a deal and you get general market advice — the same advice as everyone else who typed the same prompt into the same model on the same morning. The intelligence is a commodity. The context is not.
ICA is the method we use to make that context legible to a machine. Four layers, built in order, because each one depends on the one beneath it. It is derived from DAMA-DMBOK governance practice rather than invented for marketing purposes.
Layers 01, 02 and 04 can all be reconstructed by a competitor with enough time and money. Layer 03 cannot — it is the accumulated judgement of your specific firm in your specific market. It is also the layer that decides whether the other three produce anything useful.
A tool built on layer 04 alone is the middleware trap: fluent answers, no way to check them, and adoption collapsing the first time staff catch it being wrong. Compliance belongs in the layer rather than in the tool, so that it holds when the tool changes.
Both are taken before anything gets built and signed by both parties. A claim about recovered time is only defensible if the starting point was agreed in writing, and the guarantee is only affordable because of it.
This governs how much of the market you can actually examine. It is counted by the system rather than estimated in a meeting, and it is measured end to end — including the waiting, which is where most of it hides.
The gap between what the model assumed and what the asset delivered, broken down line by line. The only measure that improves the buying box rather than merely reporting on it — which is why it compounds and the first one doesn't.
The question every principal asks is what happens when the system misses a break clause or a licensing restriction on a live deal. The answer is that its job is to make sure no document goes unread — not to decide anything.
Every extraction cites its source down to the clause. Nothing is asserted without a reference you can open.
Low-confidence reads are flagged for human review rather than passed on silently. Silence is the failure mode that costs money.
A trail of what was checked, when, and by whom — so a file review reconstructs the decision rather than re-running it.
Nothing is built until the bottleneck has been quantified and the baseline signed. That sequence is the whole method, and it is why we can price the work fixed and refund it when it doesn't find anything.
All four surfaces, from watching the people who do the work rather than from an org chart. Shadow systems included.
In hours and pounds per year, with the arithmetic shown so you can argue with it. Then baseline and countersign.
Only the parts of ICA that bottleneck requires — in order, and no more of the estate than the problem touches.
The same measures taken again against the baseline, ownership passed to a named person, and an adoption check-in booked before we leave.
Thirty minutes on a call, no deck. We go through which of the four surfaces is hurting most and whether a Foundation Session would find anything worth acting on.
Book the Foundation Session — from £2,000