Case 01 / Real estate

Two days to twenty-five minutes, with a human signature on every contract.

A document-intelligence system that assembles, checks, and routes property contracts, so deals stop dying in the two-day wait and the brokerage closes more of the pipeline it already has.

The delivered figures on this page are real and shown with the client's permission, with their name withheld at their request. The second-order growth figures are illustrative estimates of the value unlocked.

FIG. 01The engagement, at a glance
01The situation

A fast-growing brokerage was closing deals at the speed of its slowest clause.

The client is a Dubai property brokerage and developer handling several hundred lease and sale agreements a month across residential and commercial portfolios. Every deal moved quickly until it reached the contract. There, it stopped.

Agreements were assembled by hand from a sprawl of Word templates, each subtly different, each maintained by a different person. A junior team drafted, a small legal function checked, and corrections travelled back and forth over email. A straightforward lease took the better part of two days to leave the building, and a busy week left a queue of agreed-but-unissued deals sitting in someone's inbox.

The cost was not only time. Every manual assembly was a chance to paste last quarter's rent clause, miss a mandated disclosure, or send a version legal had never seen. The people doing it were capable, expensive, and spending their days on copy and paste.

02The diagnosis

The bottleneck was not drafting speed. It was that nobody could trust a draft without reading all of it.

We ran our Four-Layer Diagnostic before proposing anything to build. The presenting symptom was that contracts were slow. The value driver underneath it was deal cycle time, the single number leadership actually watched. The binding constraint was not typing speed. It was verification: because any clause could be wrong, a senior person had to re-read the whole document, every time.

That reframed the build. Automating the drafting alone would have produced faster drafts that still needed the same full human review, moving the queue rather than clearing it. The leverage was in making a draft trustworthy by construction, so the human step became a signature on a known-good document instead of a line-by-line hunt for errors.

It also told us where to stop. Negotiated commercial terms and bespoke clauses were left firmly with people. The system earns its keep on the four-fifths of every contract that is standard, repeated, and rule-governed.

FIG. 02Symptom over root
03What we built

A document-intelligence pipeline that assembles a contract and proves it is correct before a person ever sees it.

The system ingests the deal inputs, structured CRM fields and unstructured paperwork alike, including bilingual Arabic and English documents, and resolves them into a single clean record. It assembles the agreement from a versioned, legally-owned clause library rather than a copied file, so every contract is built from the current approved language by construction.

Each assembled clause is then scored against a deviation model. Anything that departs from the approved template, a non-standard figure, a missing mandatory disclosure, an unusual term, is flagged and explained. Deterministic checks run the regulatory requirements as code, not as a checklist someone remembers. Only then does the draft reach a person, arriving with its risks already surfaced.

Nothing issues without a named human approval. On sign-off the system writes to the record, updates the CRM, and files the version. Every step, every figure, every approval is logged. This is the governed line CLRT builds under every engagement: the work, the agent, a human approval, the system of record, and an audit trail behind all of it.

FIG. 03The governed line
04The controls

The point was never a faster black box. It was a faster process leadership could still stand behind.

A person approves every contract before it issues; the agent never sends. The approver sees a draft with its deviations already flagged, so the review is a judgement on exceptions, not a proofread of boilerplate.

Every figure in a contract is placed by rule from the deal record, never generated by a language model. The model assembles and explains; it does not invent a number. Rejected drafts route back with the reason attached, and a complete audit trail means any issued contract can be reconstructed and defended after the fact.

05The results

The queue disappeared, and the expensive people stopped doing clerical work.

Contract turnaround fell from roughly two days to about twenty-five minutes of assembly plus a human review measured in minutes. The backlog of agreed-but-unissued deals cleared and did not come back.

The equivalent of 2.4 full-time people was freed from manual drafting and redeployed onto client-facing work the business actually wanted them on. After the running cost of the system, the admin saving settled at about AED 24,000 a month, and once the deals the faster desk saved are counted, the build paid for itself inside a fortnight.

The quality numbers moved with the time. Every issued contract now passes the same compliance checks by default, and not one has left the building without an approval on record.

FIG. 04Where the monthly saving comes from
FIG. 05Better as it got busier
06The leverage

The admin saving was the smallest number on this page.

Freeing 2.4 people and AED 24,000 a month is the part that is easy to count. It is not the part that moved the business. When a contract took two days, deals cooled in the wait: buyers had second thoughts, financing lapsed, a faster competitor got there first. A share of agreed deals quietly died between the handshake and the signature.

At twenty-five minutes, that gap closes while the intent is still hot. Contract-stage fall-through fell from about 12% to about 4%, which on this brokerage's volume is roughly sixteen more agreed deals completing every month. That is an estimated AED 240,000 a month in commission that would otherwise have walked, about ten times the admin saving, and earned without sourcing a single extra lead.

That is the leverage: not a cheaper contract desk, but a business that closes more of what it already has, and can grow without growing its back office.

FIG. 06From a faster desk to a bigger business
07What we did not automate

We told them where to stop.

Negotiation stays human. Bespoke commercial terms, anything genuinely novel, and the relationship itself were deliberately kept off the system. An agent that tried to negotiate would have been a liability dressed as a feature. The build is aimed precisely at the repetitive, rule-governed core of the work, which is exactly where an agent earns its keep and exactly where it is safe to let one run.

08How long it took

Week 0

Four-Layer Diagnostic. The contract desk is chosen as the first workflow to build.

Weeks 1 to 3

The clause library is formalised; drafting and the deviation model are built on real deal data.

Weeks 4 to 6

Compliance checks, the approval gate, and the audit trail go live in production behind a human sign-off.

Start here

Where is your process losing days it cannot see?

Every business has one workflow quietly costing it more than the rest. Our free diagnostic finds it and puts a number on it.