Disclosure
Pacific Software Ventures (PSV) builds AI systems that read rent rolls and T12s for CRE firms, so yes, we have a stake in how you do this. The method below works in any tool, ours included. The free analyzer runs in your browser, no sign-up.
What is AI rent roll analysis?
AI rent roll analysis means an AI reads the rent roll in whatever shape it shows up (a PDF, a scan, a property management export), pulls every line into the same fixed columns, and flags what doesn't add up: vacant units showing rent, expired leases still billing, rents far off the rest of the property, the same unit twice. Spreadsheet formulas, not the model, compute occupancy, loss to lease, rent per square foot and the expiration schedule. A person checks the flags, ties the rent total to the T12 and decides what goes into underwriting. And before you trust it on a live deal, prove it on a deal you already closed.
What should AI extract from a rent roll?
No two rent rolls show up in the same layout. So the first job: turn whatever came in into the same columns every time, one row per unit or suite, with the source page or row noted next to each value. Write those columns down before the first run. A written schema is what makes this deal comparable to the last one.
- Keep vacant and down units in the table. A rent roll that only lists paying tenants overstates occupancy. Keep every space, mark its status, and leave rent blank where there isn't any.
- Multifamily and commercial rolls hold different things. An office, retail or industrial roll has fewer rows than a multifamily one and way more per row: rentable area, lease type, recoveries, escalations, options. Give each its own schema.
- Formats it should handle. Exports from Yardi, RealPage, Entrata, AppFolio and MRI, Excel files with merged headers, PDFs and scans. Have the AI mark every value read off an image and check those first.
- The rent roll isn't the lease. Escalations, options and termination rights live in the lease. When the two disagree, the lease wins, and the gap goes on your question list. AI lease abstraction
| Field | What it holds | What a person checks |
|---|---|---|
| Unit or suite | One row per space | Every space appears once, vacant and down units included |
| Unit type and area | Bedrooms and baths, or rentable square feet | Total area matches the property's stated size |
| Tenant name | The resident or tenant entity on the lease | Vacant rows don't still carry the last tenant |
| Status | Occupied, vacant, notice, down, model | Status agrees with the rent column |
| Lease start and end | The current term's dates | Occupied units whose end date already passed |
| Contract rent | Monthly or annual base rent on the lease | Monthly and annual aren't mixed in one column |
| Market rent | The asking rent for that unit type or space | Whose number it is: the seller's or yours |
| Concessions | Free rent and rent credits | Is rent shown before or after the concession? |
| Deposits | Security deposit held, letters of credit | Deposits held match the balance sheet |
| Other charges | Parking, storage, pets, utilities, CAM and tax recoveries | Recoveries kept apart from base rent |
| Balance and delinquency | Amount past due | Big balances on tenants marked current |
Which metrics should it compute, and who does the math?
Once the columns are clean, the metrics are just formulas. Have the AI write the table into a spreadsheet and let the formulas compute. A model can explain loss to lease perfectly and still add a column wrong. Every number in your summary should point to a cell.
- A worked example (illustrative). Take an illustrative 100-unit property with 95 units occupied. Say four residents are in a free month and three owe a full month. Physical occupancy is still 95 of 100. Economic occupancy isn't. That gap shows up in the T12, and the rent roll hides it.
- Loss to lease is only as good as the market rent. If market rent is the seller's asking schedule, loss to lease is just the seller's view of the upside. Use your own market rent.
- Expirations by year drive the hold. Group end dates by year and share of rent, plus weighted average lease term on a commercial roll. A heavy year needs a leasing plan and a reserve.
| Metric | How the formula works | The common trap |
|---|---|---|
| Physical occupancy | Occupied units (or square feet) divided by total units (or square feet) | Treating occupied as leased. Notice units and pre-leased vacant units change forward occupancy (exposure), and dropping down units from the total overstates it |
| Economic occupancy | Rent collected divided by gross potential rent at market | Using contract rent before concessions, which hides the free rent |
| Loss to lease | Market rent minus contract rent, summed over occupied units | Counting the seller's asking rent as found money |
| Rent per square foot | Annual contract rent divided by occupied rentable square feet | Mixing gross and net leases in one average |
| Expirations by year | Leases and rent ending in each year | Missing month to month and holdover tenants |
| Tenant concentration | Each tenant's share of total rent | Measuring share by area instead of rent, or missing that two tenant names are one credit (affiliates or a shared guarantor) |
What red flags should an AI rent roll review catch?
Most rent roll problems are contradictions: two columns on the same row that can't both be true. A machine checks those row by row and doesn't get tired on row four hundred. But a flag is a question for the seller. It doesn't settle anything.
- The extracted rent roll becomes the estoppel checklist. Each row tells you what the estoppel should confirm: rent, term, deposit and any default. Any difference goes on the diligence list. AI due diligence for CRE
| Red flag | How it shows up | What to ask |
|---|---|---|
| Vacant unit showing income | Status says vacant, the rent column has a number | A stale row, a pre-leased unit, or rent never collected? |
| Expired lease still billing | End date before the rent roll date, status still occupied | Holdover, month to month, a renewal not entered, or a tenant already gone? |
| Outlier rent | One unit far above or below others of the same type and size | Related party, employee unit, data entry error, or a lease the buyer can't repeat? |
| Duplicate unit | The same unit or suite on two rows | Two tenants on one space, a transfer not closed out, or a double count? |
| Occupied at zero rent | Occupied status, nothing in the rent column | Free rent period, a model, an employee unit, or a missing number? |
| Concession not shown | No concession column, while the T12 shows concessions | Is this face rent the property isn't collecting? |
| Abatement running past closing | Free rent on a commercial lease that ends after closing | Does the seller credit the free months the buyer inherits? |
| Open TI or leasing commissions | A recent lease with tenant improvement or commission costs not shown as paid | What's still owed, and is it credited at closing? |
| Recoveries that disagree with the lease type | A triple net lease with no recoveries, or a gross lease billing them | What does the lease allow, and is there a cap or base year? |
| Early termination or contraction option | A large tenant can leave or give back space early | When, for what fee, and is rent counted past that date? |
| Stale date | The rent roll date is weeks before the offering or the T12 month | Ask for a current one. Leases signed or lost since then change the answer |
How do you tie the rent roll to the T12?
A rent roll is a snapshot of what the leases say. A T12 is twelve months of what the property actually collected. They never match exactly. The gap is where the good questions are.
- Line up the dates. Match the rent roll date to the latest month of the T12. More than a month apart? Ask for a rent roll from the same month first.
- Compare contract rent to the latest month's rental income. Total contract rent on occupied units should land close to the T12's latest month of rental income after vacancy loss and loss to lease, but before concessions and bad debt. Don't use gross potential rent. It still carries vacant units at market, plus loss to lease.
- Explain the gap line by line. Concessions, delinquency and bad debt sit between that figure and rent collected. Other income sits outside both. The AI drafts the bridge. A person decides whether it's explained.
- Check occupancy both ways. Compare rent roll occupancy to what the T12's income implies. A rent roll that looks full next to a T12 that doesn't is the most common catch.
- Write down what's left over. Any gap you can't explain goes on the diligence list, with the rows behind it. Our reconciliation guide has a free tie-out calculator. Rent roll reconciliation
- Normalize the T12 the same way. The T12 needs its own pass: one-time items pulled out, categories mapped to your chart, management fee and reserves set to your policy. AI T12 analysis
How do you know the AI got it right?
Run it first on a deal you have already closed. You know the answer, so you can grade it.
- Pick a closed deal with a messy rent roll. One your analyst already did by hand, with known problems: a holdover, a concession, a duplicate unit.
- Give the AI the original file, not your cleaned version. The test is reading what the seller sent, in the format the seller sent it.
- Compare row by row. Every unit, status, rent and date against your analyst's table. Count the misses by column.
- Compare the flags. Did it catch what your analyst found? Did it raise anything your analyst missed?
- Fix the instructions, then run it again. When a column comes out wrong, change the instruction, not the output. Rerun it, add a second closed deal, and keep both as your standing test.
How to automate rent roll analysis in your own tools
You can run this in Claude, Microsoft Copilot or ChatGPT. The setup matters more than the assistant.
- A written instruction file. The schema, the status list, the red flags and the output format, saved once and reused on every deal. Skip it and every analyst gets a different table. Paste and edit the sample below. Our prompt library has more. ChatGPT prompts for commercial real estate
- Output to a spreadsheet, every time. Saved into your template, so the formulas compute and each flag sits next to its row.
- Files stay in your firm's workspace. Use your firm's business plan, in your own tenant, with training turned off. Anything under NDA follows the NDA. Is AI safe for deal documents
- Guides for each assistant. Setup and prompts for ChatGPT. ChatGPT for commercial real estate
- For Microsoft 365 firms. Running it right next to your SharePoint and Excel files. Copilot for commercial real estate
You are reviewing a rent roll for an acquisition. Output: one spreadsheet row per unit or suite, vacant and down units included, with these columns: Unit | Type or area | Tenant | Status | Lease start | Lease end | Contract rent (monthly) | Market rent | Concession | Deposit | Other charges | Balance due | Source (page and row) Status: use only Occupied, Vacant, Notice, Down, Model. Rules: - Leave a cell blank when the rent roll does not show the value. Never guess. - Cite the source page and row for every value. Mark any value read from a scan. - Subtotal and header rows are not units. Check your unit count against the totals row. - Do not compute occupancy, loss to lease or totals. The spreadsheet formulas do that. Red flags, each listed with its row: - Vacant unit showing rent - Lease end date before the rent roll date, status still occupied - Rent far from units of the same type and size - The same unit on two rows - Occupied at zero rent - Concessions in the T12 but none on the rent roll - Free rent or open TI or commissions running past closing - Recoveries that disagree with the lease type - Early termination or contraction options End with a list of questions for the seller, one per flag.
Try it free: the rent roll analyzer
Before you set anything up, paste a rent roll into our free analyzer. It runs in your browser, so the file never gets uploaded. It computes occupancy, in-place rent, loss to lease and near-term expirations, then lists the questions to ask. There's no model in it, just arithmetic. That makes it a good check on your AI.
- Free rent roll analyzer and blank template. Paste from Excel or upload a CSV, with a blank template to download. Open the rent roll analyzer
Examples: underwriting and deal work at real firms
Each line below is the client's own published result, linked to the full case study.
- Westlake Realty Group, underwriting in its own model. Westlake Realty Group cut per-deal underwriting from 6-8 hours of senior time to a 20-minute review, in its own model with every number traced to source. Read the case study
- Silverado Interests, screening and underwriting prep. Silverado Interests' AI analyst works deal screening, underwriting prep and investor reporting from one email address the Dallas team forwards work to. Read the case study
- Accomplish Collective, inside its own tenant. A real estate sponsor put four AI employees on deal sourcing, underwriting, asset management and investor reporting, inside its own Microsoft 365 tenant. Read the case study
What should a person still decide?
The market rent you underwrite to. Which concessions burn off. Whether a holdover tenant is staying. How much credit an above-market lease gets. AI reads every row and shows its work. Your analyst decides what the numbers mean.
Have it built, do it yourself, or buy software?
- Have it built around your model and checklist (ours). We build the rent roll and T12 pass around your schema, your red flags and your underwriting template, inside your tenant. Then we stay on as your AI team as your process and your deals change. How an AI analyst works
- Do it yourself in Claude, Copilot or ChatGPT. Cheap to try, and it works once the instruction file is written and tested on a closed deal. Your team keeps it current. Claude for CRE operators
- Buy rent roll software. Point products extract rent rolls and T12s in their own screens. Our list compares them. Best AI rent roll and T12 software
Common questions
- How does AI handle a scanned rent roll or one with merged headers?
- It can read both, but that's where the reading errors happen. Have it mark every value it read off an image, treat subtotal and header rows as not units, and check its unit count against the rent roll's own totals row. Review the marked values first.
- Can AI calculate occupancy and loss to lease from a rent roll?
- It shouldn't do the arithmetic itself. Have the AI extract the rows into a spreadsheet and let formulas compute physical occupancy, economic occupancy, loss to lease and rent per square foot. The AI's job is getting the inputs right and citing where each one came from.
- How do I automate rent roll analysis across every deal?
- Put the schema, red flags and output format into one instruction file. Send every rent roll through it into the same template. Keep two closed deals as your standing test.
- Should the rent roll be tied to the T12 before or after the AI pass?
- After. Extract and flag the rent roll first, then normalize the T12. Then compare contract rent on occupied units with the T12's latest month of rental income after vacancy loss and loss to lease. Bridge the rest with concessions, delinquency and bad debt. Whatever's still unexplained goes on the diligence list.
- Does AI rent roll analysis work for office, retail and industrial?
- Yes, with its own schema for rentable area, lease type, recoveries and escalations. Use the rent roll to find the questions and the lease abstract to answer them.
- How should I test an AI rent roll review before using it on a live deal?
- Run it on a closed deal your analyst reviewed by hand, using the seller's original file. Compare the tables and the flag lists, fix the instructions wherever it missed, and rerun.
- Where should rent rolls go when an AI reads them?
- Into your firm's own workspace or tenant, on a business plan with training on your data turned off, and only as far as any NDA allows. Keep data vendors' exports out unless you have their written permission. Work from the seller's rent roll, the T12 and your own files.