The Break Clause Buried on Page 43 of a Lease

Notice under clause 34.2

The notice is one page. It arrives by recorded delivery at the fund's registered office, with a copy emailed to the managing agent: the tenant of unit NOT-07, a distribution warehouse off the M1 near Nottingham, gives notice under clause 34.2 of its lease to determine the lease on 24 June 2027.

Owen, the property manager at the managing agent, forwards it to Imogen with one line: "Didn't know they had a break. Did you?"

Imogen is the asset manager for the fund's English industrial units. She opens the seller's lease schedule, the spreadsheet that came with the portfolio eight months ago, and finds row 61: tenant, rent, expiry in 2032, rent reviews in 2022 and 2027. In the column headed Break, one word: none.

Then she opens the lease, a 58-page scan in the data room. The break is on page 43, clause 34.2, under the heading "Tenant's option to determine": the tenant may end the lease on the tenth anniversary of the Term Commencement Date, on not less than six months' written notice, if it gives vacant possession and owes no principal rent. The word "break" does not appear anywhere in the lease. Whoever built the schedule probably searched the scan for it.

Row 61 says none

This fund does not exist. I made it up, with its people and its warehouse, to show a problem that does: about 140 commercial leases in England and Germany, bought eight months ago, delivered as a data room of scanned PDFs and a seller's schedule that disagrees with the leases more often than anyone would like.

The notice itself is not the disaster. The tenant has a right to break, and Harriet, the fund's real estate solicitor, will check whether it was served properly and whether the conditions are met on the day. What is lost is the year before it. The break date is also the date of the second rent review, and the business plan the fund bought on assumed this rent until 2032. Had anyone known about clause 34.2 at completion, Imogen would have been talking to the tenant long before its window closed, about a longer term in exchange for a rent-free period, while there was still something to trade.

Rhys, who runs asset management, asks the question that matters more: "If row 61 is wrong, which other rows are?" Nobody can answer, because the schedule does not say where any of its values came from. Checking it means reading 140 bundles again: leases, deeds of variation, licences, rent review memoranda, side letters, and for the German offices the Nachträge. By hand that is months of careful work, and the step most likely to go wrong, applying the amendments in the right order, is also the hardest to check afterwards.

Two bundles, the messiest first

On our first video call I ask Rhys for two lease bundles, not the whole data room: the one for NOT-07, and the worst one he has. He picks an office floor in Düsseldorf whose lease has four Nachträge, one of them sent by fax.

The first thing I look for in the schedule is a source, and no row has one. That sets the rule the rest of the build follows: no value enters the abstract without the document, page and clause it came from. A schedule without sources is only as good as the last person who checked it, and nobody knows who that was.

What surprises me in the NOT-07 bundle is not clause 34.2. It is the rent. The lease provides for upward-only open market reviews on the fifth and tenth anniversaries, and there is no memorandum for the 2022 review anywhere in the bundle. Either the review was never triggered and the rent has not moved since the lease began, or the memorandum was never scanned. Row 61 gives a rent figure and does not say which.

The limits come on the same call, before anyone asks. The workflow will not judge whether the tenant's notice is valid; that is Harriet's job. It will not write to a tenant or negotiate. And the model will never compute a date. It reads a clause and cites it, code does the arithmetic, and a person confirms both.

A chain before a single field

The first piece I build extracts nothing. The rent, the break and the notice period that apply today can sit in three different documents, so each bundle is loaded into a SharePoint library per property and unit, every file is hashed so a duplicate is read once, and a vision model classifies each document: lease, deed of variation, licence, side letter, rent review memorandum, Nachtrag, each with its date, its parties and the clauses it says it amends. Code orders the chain and flags gaps, such as a Nachtrag that a later one refers to and the bundle does not contain.

Only then does the model fill the abstract, one schema per jurisdiction, each field with its value, the quoted clause, the document, the page and the clause number. A value without a citation is refused. Code applies later documents over earlier ones, clause by clause, and computes every deadline from the lease's own notice and service clauses, with holiday calendars per country and German state. Where a period ends on a weekend, it takes the earlier working day and says so. The reading side is covered in the document extraction guide, and the schemas and date rules in the blueprint for lease abstraction.

Here is row 61 again, next to what the abstract records for NOT-07:

The seller's schedule, row 61

• Expiry: 23 June 2032
• Break: none
• Rent: GBP 396,000 a year
• Rent reviews: 2022, 2027

The abstract, same unit

• Expiry: 23 June 2032, fifteen years from the Term Commencement Date of 24 June 2017. Lease, cl. 1.1, p. 3. Asset manager confirms.
• Tenant break: 24 June 2027, the tenth anniversary of the Term Commencement Date, on not less than six months' written notice, if there is vacant possession and no arrears of principal rent, both conditions quoted word for word. Lease, cl. 34.2, p. 43. Notice window closes 24 December 2026. Lawyer confirms.
• Rent: GBP 396,000 a year from the Rent Commencement Date, 24 December 2017. Lease, cl. 1.1, p. 4. Asset manager confirms.
• Rent reviews: upward-only, open market, on the fifth and tenth anniversaries of the Term Commencement Date. Lease, Schedule 4, p. 51. 2022 review: no memorandum in the bundle, outstanding or missing, request sent to the managing agent.
• Reminder: to Imogen, twelve months before the tenant's notice window closes, by fund rule, so the conversation starts while the tenant is still deciding.

Twenty bundles, abstracted twice

Nothing from the portfolio reaches Yardi on my say-so. The pilot is 20 mixed bundles from both countries: the asset managers and the fund's lawyers abstract them by hand while the workflow does the same, and the results are compared field by field. The rule for moving on is that every key date matches or the difference has an explanation, and every miss becomes a test case.

Most differences turn out to be errors in the seller's schedule. One is an error in my code. In a retail lease in Sheffield, a 2016 deed of variation renumbers clause 9, and a 2022 deed amends "clause 9.1", which by then meant something else. My first version applied every amendment to the lease as first signed, so the abstract changed the wrong clause and cited it perfectly. From then on, each document is applied to the lease as it stood when that document was signed, and two documents that both claim to change the same term go to a lawyer instead of letting the later date win by default.

The Düsseldorf bundle adds a rule of its own. In Germany a fixed term longer than a year only holds if the lease meets the statutory form, so the abstract records how each Nachtrag was signed and exchanged, and the one that arrived by fax goes to the fund's German lawyer with that fact on top. Pages with low OCR confidence are marked, and a handwritten change to a term always goes to a person, because it is exactly the kind of change a reader skims past.

Harriet confirms the breaks

Who looks at what depends on the kind of term, not on how sure the model sounds. Standard terms with a clean citation and a value in the expected range go to Imogen in a batch. Breaks, options, anything conditional, an uncertain anchor date and unusual clauses go to Harriet. Illegible pages and gaps in the chain become a request to the seller or to Owen. The reviewer sees each value, its computed date and the page image of the clause side by side. Confirming is a click, and changing a value takes a reason.

Break conditions are kept word for word because English courts apply them strictly: since Marks and Spencer v BNP Paribas in 2015, a tenant who pays a full quarter's rent to satisfy one cannot expect the overlap back. Whether the NOT-07 tenant meets its conditions on 24 June is Harriet's question, like everything else that needs legal judgment. The workflow's job is to put the clause, the date and the page in front of her early.

Confirmed terms go into Yardi Voyager keyed by lease and term, so a rerun updates instead of duplicating, and each value carries its citation as a note: the next person to question a date opens the clause, not the bundle. Dates reach the named owners' Outlook calendars at once, marked unverified until confirmed. A reminder for a wrong date costs a minute to dismiss. A missing one costs what NOT-07 cost.

The next break, a year early

Imogen's calendar now holds every tenant break in the portfolio, each with a reminder a year before the tenant's window closes and the clause one click away. The conversation about a longer term happens while a tenant is still deciding, not after its solicitor has posted the notice. The 2022 review at NOT-07 is on Owen's list, with the clause and the missing memorandum named.

When the lender asks for the schedule of breaks and expiries, it comes out of confirmed abstracts, and every row has a document, a page and a clause behind it. A new deed of variation reopens only the terms it touches. And when Rhys asks which rows are wrong, the answer is a review queue with a count, not a feeling.

When a law firm running Kira is the better answer

Lease abstraction is a mature product category. Kira, now part of Litera, and Luminance extract lease terms well and are widely used by law firms for due diligence, Prophia is built for commercial real estate teams, and MRI has its own lease abstraction offering. For a one-off review at acquisition, a law firm running one of these tools, or an abstraction service, is often the fastest route, and I would tell you so.

A build makes sense when the abstract has to be yours: your fund's template, deadlines with your own notice buffers, results written into your Yardi or MRI setup and calendars, and amendments that keep arriving after completion. A hybrid works too, with a law firm's due diligence output as the starting point and the workflow adding citations, computing deadlines and keeping abstracts current. For a portfolio like this one it usually fits the multi-step tier of AI workflow automation; more countries and worse scans move it up. The full blueprint has the flow, the schema and the rollout stages, and new lettings can be checked before signature with contract review against a playbook.

Frequently asked questions