How to Use AI for CAM Reconciliation

CAM reconciliation is one of those tasks that nobody enjoys and everybody gets wrong at least once. Every year, commercial landlords true up the estimated CAM charges billed to tenants against the actual operating expenses incurred. The process involves cross-referencing lease-specific CAM provisions — caps, exclusions, base year stops, pro-rata shares — against a general ledger that was almost certainly not organized with reconciliation in mind.

For a 15-tenant office building, this means opening 15 leases, extracting 15 different sets of CAM terms (because no two are ever identical), mapping each tenant's provisions against the actual expense categories in the GL, and calculating the over/under for each tenant. It is tedious, detail-intensive, and the margin for error is high. Overbill a tenant and you have a dispute. Underbill and you leave money on the table. Miss an exclusion and you have a potential lease default argument.

AI can now handle most of the heavy lifting in this process. Not all of it — we will be clear about where human review is still necessary — but enough to turn a multi-day exercise into an afternoon.

Why CAM Reconciliation Is Hard (Even for AI)

Before walking through the workflow, it helps to understand why this task is more complex than it looks.

The difficulty is not the math. The math is straightforward — pro-rata share multiplied by recoverable expenses, adjusted for caps and exclusions. The difficulty is that every lease defines "recoverable expenses" differently. Tenant A's lease excludes capital expenditures above $10,000. Tenant B's lease caps CAM increases at 5% per year over a base year. Tenant C's lease excludes management fees entirely. Tenant D has a gross lease with a base year stop that converts to NNN treatment after Year 3.

Each of these provisions can live in a different section of a different lease document, written by a different attorney, using different language to describe the same economic concept. The reconciliation is really a translation exercise — mapping legal language to accounting categories — and that is where AI earns its value.

The Workflow

Step 1: Extract CAM Provisions from Every Lease

This is the foundation. Before you can reconcile anything, you need a standardized view of every tenant's CAM terms.

If you are using a general-purpose agent like Claude Cowork, you can point it at your lease folder and ask it to build a CAM provision matrix:

"Read every lease PDF in the 'Leases' folder. Create an Excel workbook called CAM_Provisions.xlsx with one row per tenant and the following columns: Tenant Name, Suite Number, Lease Commencement Date, Pro-Rata Share (%), CAM Cap (if any — note whether it's cumulative or non-cumulative), Base Year (if applicable), Excluded Expense Categories (list each one), Management Fee Cap (if any), Capital Expenditure Threshold for Exclusion, and any other non-standard CAM provisions worth noting. For each extracted term, note the page number in the source lease."

The page number reference is important. When a tenant disputes a reconciliation, you need to point to the exact clause. AI that provides source traceability is meaningfully more useful here than AI that just gives you the answer.

Purpose-built lease platforms like Prophia and Kira handle this extraction natively and will hyperlink each data point back to the source paragraph. For a general-purpose agent, you get the extraction but should verify the complex provisions manually.

Step 2: Categorize the General Ledger

The second input is your actual operating expenses for the year. This typically comes as a GL export — a CSV or Excel file with hundreds or thousands of line items, each coded to an expense category.

The problem: your GL categories rarely map cleanly to the CAM recovery categories in your leases. "Repairs & Maintenance" in the GL might include items that some leases classify as capital expenditures. "Professional Fees" might include legal costs that certain tenants' leases exclude from recoverable expenses.

This is a good task for AI because it requires reading every line item and making a judgment call:

"Review the general ledger export in GL_2025.csv. For each line item, classify it into one of the following CAM recovery categories: Common Area Maintenance, Utilities, Insurance, Real Estate Taxes, Management Fee, Capital Expenditure, or Non-Recoverable. Flag any line item above $5,000 that could reasonably be classified as either CapEx or OpEx — I need to review those manually. Also flag any line item that appears to be a one-time, non-recurring expense."

The flags are the key. You are not asking the AI to make the final call on ambiguous items — you are asking it to surface the items that need your judgment. This is the same approach a good property accountant takes, except the AI processes 500 line items in the time it takes a human to process 20.

Step 3: Apply Tenant-Specific Provisions

This is where the two inputs come together. You now have a standardized CAM provision matrix (Step 1) and a categorized expense ledger (Step 2). The reconciliation is applying each tenant's specific terms to the actual expenses.

"Using CAM_Provisions.xlsx and the categorized GL, calculate the CAM reconciliation for each tenant. For each tenant:

1. Start with total recoverable expenses.
2. Remove any expense categories that tenant's lease excludes.
3. Apply their pro-rata share percentage.
4. If the tenant has a CAM cap, apply it (note whether the cap is cumulative or non-cumulative and calculate accordingly).
5. If the tenant has a base year stop, calculate the recovery as the excess over the base year amount.
6. Compare the calculated recovery to the estimated CAM charges billed during the year.
7. Output the over/under amount for each tenant.

Save the results in CAM_Reconciliation_2025.xlsx with a summary tab and a detail tab for each tenant showing the full calculation."

A well-structured prompt like this will produce a working reconciliation for most standard lease structures. The output should show the math clearly enough that you can trace any number back through the logic.

Step 4: Audit the Edge Cases

AI handles the standard provisions well — flat pro-rata shares, simple percentage caps, straightforward exclusion lists. It struggles with the provisions that require interpretation:

Cumulative vs. non-cumulative caps. A non-cumulative 5% cap means CAM cannot increase more than 5% over the prior year. A cumulative 5% cap means CAM cannot increase more than 5% compounded annually over the base year. These produce very different numbers over a 10-year lease, and the language distinguishing them in leases is often ambiguous. Verify these manually.

Controllable vs. non-controllable expense splits. Some leases cap only "controllable" expenses while passing through "non-controllable" expenses (taxes, insurance, utilities) without a cap. The definition of "controllable" varies by lease. AI can extract the language, but a human needs to confirm which GL categories fall on which side of the line.

Gross-up provisions. If the building is not fully occupied, many leases allow the landlord to "gross up" variable expenses to what they would have been at a specified occupancy level (often 95%). The gross-up calculation is straightforward, but determining which expenses are variable (and therefore subject to gross-up) versus fixed requires judgment.

Audit rights and dispute windows. Many commercial leases give tenants 90–180 days to audit the CAM reconciliation after receiving it. Knowing which tenants have audit rights — and whether those windows are still open for prior years — is critical context that the AI can extract from the lease but that you need to track operationally.

Step 5: Generate Tenant Statements

Once the reconciliation is finalized, you need to produce individual statements for each tenant showing their charges, payments, and the amount due or to be credited.

"Using the finalized CAM_Reconciliation_2025.xlsx, generate a one-page CAM reconciliation statement for each tenant as a PDF. Each statement should include: Property Name, Calendar Year, Tenant Name, Suite Number, Total Recoverable Operating Expenses, Tenant's Pro-Rata Share, Tenant's Share of Expenses, Less: Estimated CAM Billed During the Year, and the Net Amount Due (or Credit). Include a line-item breakdown of the major expense categories. Save each statement as CAM_Statement_2025_[TenantName].pdf."

This is the kind of repetitive document generation that AI handles cleanly. Fifteen tenants, fifteen statements, each with different numbers but the same format.

The Tools

Purpose-built platforms (MRI Contract Intelligence, Yardi, Prophia) are the best choice if you manage a large portfolio and do CAM reconciliation annually across dozens of properties. They integrate lease data, GL data, and billing in a single system and automate most of the workflow natively.

General-purpose agents (Claude Cowork, ChatGPT) are well-suited for firms that do CAM reconciliation on a smaller scale or as part of an acquisition due diligence process — verifying the seller's CAM reconciliations before closing. The flexibility to work with whatever file formats you have, without importing into a dedicated platform, is the advantage.

Spreadsheet-only approaches still work if you are comfortable building the logic yourself. AI can accelerate the extraction steps while you maintain the reconciliation model in your own Excel template.

Where Human Judgment Still Matters

AI compresses the time on CAM reconciliation dramatically. It handles the extraction, the categorization, the math, and the statement generation. But the decisions that carry financial and legal weight — how to classify an ambiguous expense, whether a cap is cumulative, which costs are subject to gross-up — still require a human who understands the lease and the property.

The right way to think about it: AI does the work of the first pass. You do the work of the final pass. The first pass used to take days. Now it takes minutes. That is a real gain — but only if the final pass still happens.

This is also where the distinction between general and purpose-built tools matters most. A general agent will extract CAM terms and do the math, but you have to teach it the nuances every time. A purpose-built CRE coworker — like those from Lumetric — already understands the difference between cumulative and non-cumulative caps, knows to flag gross-up provisions, and produces reconciliations in the format your property management team expects. Not a tool you configure for CRE work. A coworker that was built for it.