The Five Clauses That Break AI Lease Abstraction

Every vendor in this category publishes an accuracy number. We are not going to repeat any of them, because none of them are independently verified and, more importantly, because a single accuracy percentage is the wrong shape for this problem.

Lease abstraction does not fail evenly. It fails in specific places, and those places correlate almost perfectly with where the money is.

The easy fields are genuinely solved. Commencement date, expiration, square footage, base rent, renewal options, notice addresses. These are short, structured, and appear in predictable places. If a tool is getting those wrong, it is a bad tool.

The five below are different. They are where retail leases carry economic weight, and they share a property that makes them hostile to extraction: the answer is not in one place, and often is not stated at all.

1. Co-tenancy

The single hardest clause in a retail lease and the one that most rewards getting right.

A co-tenancy provision lets a tenant exercise remedies if the center fails to meet stated occupancy conditions. It comes in two structurally different forms. An opening co-tenancy says a tenant need not open, or need not pay full rent, until named stores or a threshold of stores are open. An operating co-tenancy governs what happens if that condition fails later, during the term.

Now consider what a machine has to resolve to answer "is this tenant currently entitled to reduced rent."

The condition may be tied to named key tenants, or to a percentage of total occupancy, or to a square footage floor, or to a combination. Named tenants raise the question of whether a successor or replacement of comparable quality satisfies the clause. Occupancy thresholds raise the question of whether you are measuring gross leasable area, occupied area, or store count, and whether temporary tenants count. The remedy is usually a ladder: substitute rent for a period, then a further reduction, then a termination right that may require notice within a window.

The status of a co-tenancy clause is a function of the clause, the current rent roll, and a calendar. Extraction can pull the language. It cannot tell you whether you are in breach today, and a tool that presents the clause as a tidy field is flattening something that is not flat.

2. Exclusive use

The clause is short. The obligation it creates is portfolio-wide.

An exclusive says the landlord will not lease other space in the center to a competing use. The trouble is that "competing use" is defined by prose, not by a code, and the definitions vary: some are drawn to a named product category, some to a percentage of a prospective tenant's floor area devoted to that category, some carve out existing tenants, some run to successors and some do not.

The extraction problem is that answering "can I sign this deal" requires checking a prospective tenant's use against every exclusive in the center simultaneously. That is a cross-document question. Almost all abstraction tools are built to produce a per-lease record, which is the wrong unit for the question that actually matters.

3. Percentage rent

Percentage rent is owed on sales above a breakpoint. The mechanics are arithmetic, and the arithmetic is not the problem.

The problem is the definitions around it. Gross sales gets defined in the lease and the definition is negotiated: returns, employee discounts, gift card redemptions, online orders fulfilled from the store, curbside, third-party delivery, and sales taxes are all included or excluded by specific language. Two tenants in the same center commonly have different definitions.

The breakpoint may be natural, derived by dividing base rent by the percentage rate, or artificial, a negotiated number that does not follow from base rent. A tool that assumes a natural breakpoint where the lease specifies an artificial one produces a number that looks reasonable and is wrong.

4. CAM caps, exclusions, and gross-ups

Covered at length in our CAM piece, and it belongs on this list too.

The recurring extraction failures are the cap structure, which may be cumulative or non-cumulative and compounding or not, applied only to controllable expenses; the exclusion language, where "capital expenditures," "expenditures of a capital nature under GAAP," and an enumerated list are three different obligations that read similarly; and the gross-up provision, which is frequently a single sentence with large consequences and is easy to miss entirely.

5. What the lease does not say

The most consequential extraction failure has no field.

Abstraction produces a record of what is present. It does not flag what is absent, and absence in a retail lease is often the point. No audit right. No cap on controllable expenses. No relocation protection. No exclusive at all. A silent lease produces a clean-looking abstract with no warnings on it, which is precisely backwards from how a human reviewer would react.

What to do about it

The practical answer is not to avoid these tools. It is to use them for the right half of the job and to know which half that is.

• Use extraction for the structured fields and for finding where the hard clauses live. "Locate and quote every co-tenancy provision across 40 leases" is a task these tools do well and a human does slowly.
• Treat the five clauses above as review items, not data fields. Extract the language verbatim with a page citation, then have a person read it.
• Ask for the source text, not just the value. Any tool that gives you a populated field without the language it came from cannot be checked, and unverifiable output is worse than no output because it gets trusted.
• Test on your hardest lease, not a clean one. The vendor demo lease is always well drafted. Yours are not, especially the ones you inherited in an acquisition.
• Watch the silences. Build your review checklist around the clauses that should be there, so an absent audit right shows up as a question rather than as nothing at all.

The summary

Lease abstraction is real and worth using. The category's weakness is that the fields it is most confident about are the ones that matter least, and the provisions that decide whether a tenant can stop paying full rent are the ones it is least equipped to resolve.

Buy it for the reading. Keep the deciding.