AI for Contract Review in Real Estate: Leases, Licences and the Documents Before Signature

A property company rarely loses money because a document went missing. It loses money because a document was signed that says something slightly different from what everyone in the room believed it said. The heads of terms promised a rent free period conditional on nothing, and the engrossed lease made it conditional on the tenant completing works to the landlord's satisfaction.

This is about the review itself: the pass over a document while there is still time to change it. Legal, asset management and company secretarial teams already own the paper. What they lack is time to read every version with the attention they gave the first.

From heads of terms to engrossment

The heads of terms are where the deal is agreed and the lease is where it is written down, and the two drift apart more often than anyone admits. An agent reads them together and returns a point by point reconciliation: every commercial term from the heads, where it landed in the draft, and whether it landed intact, altered or not at all. On a pre-let or an agreement for lease it also extracts the conditionality the finished lease never shows: longstop dates, the works obligation and the standard it is measured against, and whether the tenant can walk away or only claim.

At engrossment the question is not whether the document is reasonable but whether it differs from the last version your side agreed. The agent compares the two clause by clause, lists every addition, deletion and rewording with its consequences elsewhere, and checks what people stop seeing on a fifth read: defined terms never defined, cross references to the wrong schedule, plans referred to but not attached.

What the review note contains

Not a marked up file. A note a fee earner or surveyor can act on: the issue, the clause, the wording verbatim, what it changes against your standard position, and a suggested action. Points are ordered by consequence, and anything unresolved is quoted in full.

Licences for alterations, assignment, underletting and change of use

Consent documents arrive on the other side's paper, under time pressure, from a tenant wanting an answer this week. An agent checks each one against the lease's own consent machinery: whether the works fall inside the definition the lease uses, whether the reinstatement obligation at expiry is preserved or quietly narrowed, and whether the conditions the lease permits have been imposed. On assignment it checks whether an authorised guarantee agreement is required and whether the annexed form matches what the lease specifies. On underletting it checks that the underlease mirrors the head lease on review basis, user and alienation.

Rent review mechanics and the assumptions that drive them

Rent review is decided by drafting rather than by the market. The assumptions and disregards do the work: the hypothetical term and whether it is the residue or a fresh term, the assumed state of repair, whether the tenant's own fitting out is disregarded, and whether the user clause is treated as drafted or as relaxed. An agent extracts the mechanism from each lease and sets it against your standard position, including review dates and pattern, the trigger and whether time is of the essence, and the determination route. On incoming third party paper it runs the same comparison against the last version of that document you executed.

Break options and the conditions that make them fail

Break clauses fail on conditions, not on dates. Vacant possession where the definition is stricter than anyone assumed, material compliance with covenants, payment of all sums including sums not yet demanded, and notice in a particular form on a particular recipient. An agent produces the operative version of the clause: every condition stated as something a named person must do, the consequence of imperfect service, and the point by which each condition has to be satisfied, frequently many months before the break itself. That work happens at review time, while the clause can still be negotiated.

Rent deposits, guarantees and the covenants that follow assignment

Security documents are short, and are read as though that makes them simple. An agent checks the deposit deed for the events permitting drawdown, the conditions for return, and whether the deposit transfers on a sale of the reversion. On guarantees it checks what is actually guaranteed, whether an authorised guarantee agreement is limited to the assignee's own period or drafted more widely, and whether a proposed variation falls outside what the guarantee contemplates. A variation agreed without the guarantor joining in only surfaces as a mistake when the security is needed.

Sale contracts, replies to enquiries and the disclosure bundle

On a transaction the contract is rarely the hard part. The hard part is consistency between the replies to enquiries, the disclosure bundle and the leases themselves. A reply says no side letters exist, and one sits in the bundle under a different property reference. An agent cross checks every reply against the underlying documents and returns contradictions with both sources cited, reads the contract for conditions, apportionments, time bars on warranty claims and any retention, and maps each disclosure against the warranty it qualifies.

Development documents, appointments and collateral warranties

Professional appointments and their collateral warranties are supposed to match. Frequently they do not. An agent compares each warranty against the appointment it hangs off and reports the gaps: standard of care worded differently, net contribution in one and absent in the other, professional indemnity requirements at different levels or periods, assignment rights that will block a future sale or funding, and beneficiaries named in the appointment for whom no warranty was ever produced. It also checks whether the document was executed as a deed where the limitation period depends on it.

Managing agent contracts, insurance and reinstatement

Agency and facilities contracts are your own paper, and are reviewed less carefully than anything a tenant signs. Worth extracting on each: the scope of what the agent is obliged to do, the fee basis and what sits outside it, the service standards and whether failure has any consequence, termination rights on both sides, and what happens to data, records and bank balances on termination, which decides how painful a change of agent will be.

On insurance the question is whether the lease and the policy describe the same thing. An agent compares insured risks against the cover in place, checks the rent suspension period against a realistic reinstatement period, and reads the provisions on uninsured risk, termination on destruction, and reinstatement that proves impossible.

Service charge provisions read as a liability question

Read as a liability question rather than an accounting one, these provisions answer one thing: what is recoverable from whom, and on what conditions. An agent extracts the recoverable heads, the sweeper wording and how broadly it is drafted, exclusions for initial repair and inherent defect, whether improvements are recoverable or only repair, the apportionment method, any cap and whether it is personal to the original tenant, and the procedural conditions on demanding and certifying. Doing this at review time prevents cost being incurred in good faith under a clause that does not reach it.

Side letters and concessions that never reach the title

Side letters are the most dangerous documents a property company holds, because they modify the deal and live nowhere structured: a cap, a personal right to break, a relaxed user restriction, a rent free extension agreed in an exchange of letters. An agent reads them alongside the lease they vary and produces a single statement of the position in force: whether each concession is personal to the original tenant, whether it survives assignment or a sale of the reversion, and whether it has already expired. On a disposal, that statement is what stops a buyer's solicitor finding something your own team had forgotten.

Portfolio questions nobody could afford to ask

Reviewing at machine speed makes questions askable that were never worth asking. Which leases permit the change of use a prospective tenant wants. Which allow termination on destruction. Where do we hold guarantees that a proposed variation would put at risk. Each answer comes back with the clause quoted and the source document identified, so an asset manager can check the ones that matter rather than trusting a count. Extraction is a draft finding rather than a fact: a sample is verified by a person, disagreements are measured, and the error rate is reported honestly.

Where AI fails here

Legal advice is not on offer. Whether a condition has been satisfied, how an ambiguity would be construed, whether a variation released a guarantor: those are questions for your solicitors, and an agent answering them produces something you should not act on.

Nothing that commits the business is issued by an agent. Break notices, review triggers, grants of consent and anything executed as a deed are drafted, cited to the clause and put in front of the person authorised to issue them. That person issues it. This is a deliberate refusal, not a technical limit: service has legal effect and a wrongly served notice can lose a right permanently.

Commercial decisions stay with named people: whether to accept a wider user clause to win a covenant you want, or to trade a cap for a longer term.

Two practical failures are worth naming. Scanned documents, title plans and older leases in poor condition extract less reliably than clean digital text, so accuracy on those is measured separately rather than averaged into a flattering number. And a review built against a superseded document is worse than none, so the governing version is confirmed before the work starts and rechecked whenever a deed of variation or supplemental agreement lands.

The first review worth handing over

Start with the document type you receive most often and vary least, usually licences for alterations or assignment. Write down the positions your team already applies without articulating them. Run the agent in parallel with your existing reviewer and compare the outputs point by point. Count what the agent found that the human missed, what the human found that the agent missed, and what the agent asserted that was not in the document at all. That third number decides whether this is usable.

When the disagreement rate is low enough and the failure modes are understood, the reviewer moves to the exceptions and to the documents that commit the business, where their judgement was always worth more. That is the work Adopt AI Partners does: a narrow deployment on your own documents, an accuracy figure measured on your own contested paper, and a handover once your team owns the system.