Lease intelligence | Property

Lease Intelligence

Client Real estate client
Engagement Prototype
Focus Clause-level provenance
DOCUMENT CHAIN RECONCILIATION
CLAUSE-LEVEL PROVENANCE
INTERPRETATION UNDER AMBIGUITY
HUMAN CONFIRMATION
VIDEO DURATION
1m 01s

BRIEF

A commercial landlord’s most important dates and figures don’t live in one document. Over the life of a tenancy the original lease gathers side letters, licences, deeds of variation and rent review memoranda, each quietly changing something that came before. The lease register the finance team relies on is a hand-kept summary, updated when someone remembers to.

Arquin was asked to show that a system could read a tenancy’s full document chain the way a property lawyer would: in order, applying each change, and arriving at the position that applies today. It would then check that position against the register and show its working, so that a finance director could act on the result without having to take the system’s word for it.

Lease Position reads every document in date order and keeps a running log of what it is doing. The log names the document and clause behind each step, so the reasoning can be followed as it happens.

CHALLENGE

The obvious build is to point a language model at a lease and ask for the key terms. It will answer quickly and confidently. But the lease is rarely the whole story. A deed of variation can move a break date by a year in a single sentence, and a side letter can cap a service charge only for as long as one tenant remains. Read the lease alone and the answer is wrong, with no sign that it is wrong.

The harder problem was making the output trustworthy enough to act on. The system had to apply documents in the right order and respect their conditions. It had to say when it was interpreting rather than reading, and flag documents that are referred to but missing rather than filling the gap. Every conclusion needed a source a person could check in seconds.

Differences from the register appear first. Each shows the deadline it affects, what is at stake and a step-by-step account of how it was worked out, with the clause that applies and the clause it superseded.

SOLUTION

Arquin designed the system around a clear separation between reading and deciding. A language model does what it does well: reading legal drafting, following cross-references and explaining its reasoning in plain English. The documents themselves remain the only source of truth, and a person remains the only one who can settle a term.

The output is structured, not conversational. Each term carries the document, clause and exact quote it was taken from, along with any earlier clause it replaced. Each difference from the register becomes an exception with a deadline, an amount at stake and its reasoning. Where the drafting doesn’t settle a point, the system gives an interpretation, a confidence and the reason, and leaves the decision to a person.

The result is a view of each tenancy that leads with what needs attention. In the example, a 2023 deed of variation brought the tenant’s break forward from 2028 to 2027, leaving weeks to serve notice while the register still showed the original date. The system found it, showed exactly why, and waited for someone to confirm it.

GOVERNANCE

In a system that informs financial decisions, trust is designed in from the start rather than added at the end. Three principles govern how Lease Position handles what it reads.

Read is not confirmed. Every term is marked as either read from the documents or confirmed by a named person, and the two look different on screen. The system proposes and a person settles. Conflicts can be confirmed or referred to a solicitor, but never resolved automatically.

Provenance by default. Every position links to the clause it came from, quoted exactly, and to the clause it superseded. If anyone asks why the system believes a date or a figure, the answer is already on screen.

Gaps are flagged, not filled. Where a document is referred to but not on file, the system says so and asks for it. Where the drafting is ambiguous, it states its interpretation and its confidence. It doesn’t present a guess as a fact, and adding a new document marks the position for reading again.

WORKFLOW

SCOPE & ROADMAP

This prototype was deliberately focused. It proves the core: reading a full document chain in order, reconciling it with the lease register, surfacing exceptions with their reasoning, and requiring human confirmation before anything is treated as settled.

The roadmap extends in clear steps. Reading documents directly from a document management system or data room would remove manual upload. Writing confirmed positions back to the register, or to the property management system, would close the loop. Portfolio-wide monitoring would turn a periodic audit into a continuous check, with notice deadlines feeding the finance calendar automatically.

The same pattern applies wherever a position is built up from a chain of documents: loan facilities and their amendments, insurance policies and endorsements, supplier contracts and change orders. Each step adds reach. None relaxes the principles above.

APPROACH

Arquin builds in stages, proving each layer before adding the next, so that value and risk are both visible at every point. The technology is chosen to fit the problem, not the reverse. And the hardest questions, such as what the system is allowed to conclude, who confirms it and how its reasoning is shown, are treated as design decisions from the first day, not compliance added at the end.

BACK TO WORK