Extracting Square Footage and Premises Description From a Lease

Ask a lease how big the space is and it answers twice, with different numbers, in different sections, and both answers are right. Extracting “square footage” as one field means picking one of them at random, and every rate you compute afterwards inherits the coin flip.

Two numbers, both correct

Usable area is what the tenant occupies: the inside of the suite. Rentable area is usable area plus an allocated share of the building’s common areas — lobbies, corridors, shared restrooms, mechanical rooms. Rent is quoted and charged on rentable area; the space you can put desks in is usable area.

The measurement method is not folklore. BOMA International publishes the standard most North American office leases reference, ANSI/BOMA Z65.1 for office buildings, and it has been revised more than once, so the lease normally names both the standard and the revision year. When it does, that is an extractable field, and it is the field that tells you whether two buildings’ numbers are comparable at all. BOMA International publishes the floor measurement standards.

So the minimum field set is usable_sf, rentable_sf, load_factor, measurement_standard, standard_revision and, for each of the numbers, the section it came from. Any schema with a single square_feet column is lossy in a way no confidence score will reveal, because the model was not uncertain — it just answered a question with two answers.

The load factor has two definitions

The relationship between the two numbers is published under several names — load factor, add-on factor, common area factor, core factor, gross-up factor — and it is expressed in two arithmetically different ways. Take a suite with 10,000 usable and 11,800 rentable square feet:

Usable    10,000 SF
Rentable  11,800 SF

Form A   rentable / usable        = 1.1800   ("a 1.18 factor")
         stated as a percentage   = 18.00%
Form B   1 - (usable / rentable)  = 0.1525   ("15.25% loss factor")

Same suite. 18.00% and 15.25% are the same fact.

A lease saying “the Rentable Area reflects a load factor of 15.25%” and one saying “a load factor of 18%” can describe an identical space. Applying Form A’s arithmetic to a Form B number gives 10,000 × 1.1525 = 11,525 rentable square feet, understating the space by 275 square feet — a 2.3% error that will never look obviously wrong.

The defence is to never derive from the factor when both areas are stated. Extract all three, recompute the factor both ways, and see which one matches the stated figure; that identifies which definition the lease is using and validates the two areas at the same time. Where only one area and a factor are given, store which form the arithmetic implies and flag the record if it is not determinable.

What conflating them does to a rate

The damage shows up in the per-square-foot rate, which is the number everyone actually compares across a portfolio:

Annual base rent          354,000.00
Rentable area             11,800 SF
Usable area               10,000 SF

Correct  354,000 / 11,800 = 30.00 per RSF
Wrong    354,000 / 10,000 = 35.40 per RSF   (+18%)

A portfolio in which some leases were extracted one way and some the other has an 18% spread that is entirely artificial, and it moves in the same direction as the real signal — buildings with generous common areas have higher load factors — so it does not look like noise. It looks like a market insight.

Two cheap checks catch it, both of them cross-field validation rules rather than anything the model is asked to judge. First, assert that rentable_sf >= usable_sf whenever both are present; a violation means the two were swapped. Second, assert that the load factor falls in a plausible band, flagging anything below about 1.05 or above about 1.35 for review — not because those are impossible, but because they are unusual enough to be worth a look, and a factor of exactly 1.0 almost always means the same number was extracted into both fields.

Stipulated area beats measurement

The instinct on finding two areas that do not reconcile is to remeasure, and in most leases that instinct is wrong, because the lease says so explicitly:

"The parties stipulate and agree that the Rentable Area
 of the Premises is 11,800 square feet and the Usable
 Area is 10,000 square feet, and such figures shall be
 binding notwithstanding any subsequent measurement."

A stipulation clause makes the numbers contractual facts rather than measurements. Rent is computed from the stipulated figure even if a surveyor would disagree. So area_stipulated is a boolean worth having, and where it is true, an area that fails a plausibility check is still the operative number and the check should be suppressed rather than escalated — a case worth building into review routing rather than letting it fill a queue.

The opposite clause exists too: a remeasurement right, letting one party remeasure on some trigger and adjust rent accordingly, sometimes with a cap on how much the area may change. That is a conditional future change to a field you have extracted, and it belongs on the record as a flag so that the number is not treated as permanent.

The premises description is not the number

The area figure answers what rent is charged on. The premises description answers what the tenant leased, and it is a different field with different failure modes.

• It usually incorporates a floor plan by reference. “Suite 400 on the fourth floor, as more particularly shown on Exhibit A” means the operative description is a drawing. The extraction should record the exhibit reference and whether that exhibit is actually present in the file; a lease whose Exhibit A is missing from the scan is a common and checkable defect, and handling a missing required field covers what to emit for it.
• “Approximately” is load-bearing. An area described as approximate and not stipulated behaves differently from a stipulated one, so capture the qualifier rather than discarding it during number parsing.
• Appurtenant rights are separate. Parking spaces, storage areas, roof rights and signage are frequently granted in the same section and are not part of the premises area. A storage room with its own square footage and its own rate, extracted into the premises field, corrupts both numbers.
• Multiple suites. Where the premises span two suites or two floors, there may be an area per suite and a total, and the total may not equal the sum if a common-area allocation was applied once rather than per suite. Extract the components and the total, and record the discrepancy instead of resolving it.

Almost every check on this page is arithmetic on numbers the lease itself supplies, which is what makes area extraction unusually verifiable compared with clause extraction. Getting the schema right — components, totals, factor, standard and stipulation flag — is most of the work, and the rent arithmetic that depends on it is worked through in extracting rent escalation schedules.