Claude for Rent Roll Reconciliation: A Worked Example

Use Claude to organize the rent roll, billing detail and operating statement into a review table, then verify each difference against the source. Align dates and accounting definitions before comparing totals. Recalculate accepted figures in the workbook and retain unresolved items instead of asking the model to force a tie-out.

Separate extraction from reconciliation

Extraction asks whether the assistant copied a value correctly. Reconciliation asks whether values describe the same thing and, when they differ, whether the difference has an explanation. A correctly extracted annual total can still be the wrong comparison for a current monthly schedule.

Our rent-roll and T-12 extraction guide covers the first handoff. This example begins after the files are readable and focuses on what the analyst should accept, question or leave unresolved.

Anthropic's file-upload documentation describes supported file types and handling. Check the capabilities available in your account. Use permitted documents, retain the originals, and confirm that the assistant can actually inspect the supplied tables. An upload succeeding does not demonstrate that every row was read accurately.

Define the period and the meaning of rent

Write down the rent-roll as-of date, the billing month, the collection cutoff and the operating-statement period. Ask whether the rent column is contractual base rent, billed rent, market rent, effective rent or collected cash. Keep reimbursements and other income separate unless the source explicitly combines them.

Fannie Mae's Multifamily Analysis of Operations definitions distinguish income and expense categories used in its lending analysis. That is useful background for consistent labels; a specific lender's requirements should not be silently copied into every acquisition model.

For this exercise, contractual monthly rent is the lease schedule before concessions. Billed rent is that month's charge after the stated concession. Receipts are the cash collected by the cutoff. A receivable is not automatically bad debt.

Worked example: the current month

All units and amounts are synthetic. This is a training exercise, not a client property or a measured Claude benchmark. The pack contains CSV inputs and an answer key.

The monthly bridge is $4,950 contract rent - $200 concession = $4,750 billed; $4,750 billed - $150 unpaid = $4,600 received. The $150 remains an unpaid balance in this packet. No write-off is supplied.

The practice rent roll deliberately contains a second A102 row marked as an import duplicate. If both rows are counted, the contract total becomes $6,700. Flag it and verify the duplicate marker before excluding it. In a real file, repeated unit IDs may reflect roommates, transfers or charge detail, so deduplication requires more than spotting the same unit number twice.

Download the rent-roll reconciliation practice pack. Open the inputs first and keep the answer key aside until the review is complete.

Compare the current schedule with the T-12

The synthetic T-12 records $55,200 of rental receipts for September 2025 through August 2026. Annualizing the September 2026 contract schedule produces $59,400. The $4,200 difference is a comparison between historical collected cash and a current contractual run rate, not a proven collection shortfall.

The packet does not include monthly historical rent rolls or the full tenant ledger. A sound output asks for those records before allocating the $4,200 between prior vacancy, lease changes, concessions and payment timing. An answer that invents that allocation fails the exercise.

Keep this comparison outside the historical NOI calculation. The multifamily underwriting walkthrough shows how to preserve the seller's history and build a separate buyer-adjusted case.

Give Claude a bounded review instruction

Use the instruction below with the practice inputs. It is a starting prompt, not a promise about the model's output:

Reconcile the supplied rent roll, September billing and receipts, and historical T-12. First state the period and basis of each file. Preserve original values and cite file names and row IDs. Identify duplicates and explain any proposed exclusion. Produce separate tables for extraction issues, explained differences and unresolved differences. Do not invent lease history, write-offs or missing monthly records. Show arithmetic for recomputation and list the documents needed to resolve each open item.

Ask for a table with source value, proposed interpretation, evidence location, difference, status, reviewer and next question. Keep a supported extraction separate from an approved underwriting assumption.

If the assistant references the answer key before the test, discard that run as an evaluation. Record the model, date and supplied inputs so later runs face the same task.

Check the output in your workbook

Recompute the accepted totals with formulas. Inspect the duplicate exclusion, the concession and the unpaid balance individually. Compare source row counts with imported row counts before relying on a grand total.

A reviewer should also sample apparently correct rows. A process that only checks flagged items can miss the assistant's confident errors. Record corrections, unanswered questions and time to an accepted result if you want to evaluate the workflow.

Use a change exercise: replace the A103 receipt with the full billed amount and rerun the review. September receipts should increase by $150 and the unpaid balance should fall to zero. The historical T-12 should remain unchanged. This checks whether a correction affects the appropriate period.

Carry the accepted figures forward

The final handoff needs approved figures, source references, open items and the workbook version. It should explain what changed and who accepted it. Keep the original import and the correction history so a later reviewer can reconstruct the result.

The underwriting model-builder pack provides a structure for that model handoff. If your team needs this review built around its existing workbook, discuss the underwriting workflow with a permitted example of the document mismatch.

Frequently asked questions

Should a rent roll multiplied by twelve equal the T-12?

Not necessarily. A rent roll is a point-in-time schedule; a T-12 describes a historical period. Lease changes, occupancy, concessions, collections and accounting treatment can create legitimate differences. Reconcile the basis and period before labeling a mismatch an error.

Can Claude decide which rent figure is correct?

Claude can identify conflicting figures and point to supplied sources. The analyst still needs to verify the period, lease terms and accounting basis, then approve the figure that belongs in the model. An unexplained difference should stay open.

Is an unpaid balance the same as bad debt?

No. An unpaid balance can remain a receivable. Do not treat it as an approved write-off or deduct it again from reported net collections without reviewing the ledger and the accounting treatment.

Does this example show Claude accuracy?

No. It is a synthetic practice packet with an answer key. It provides a repeatable test for your own run, but it is not a measured benchmark of any model or a claim about performance on client documents.