How to Underwrite a Multifamily Deal With AI: A Practical Walkthrough

The Short Answer

Underwriting a multifamily deal with AI follows four steps: extract the rent roll and T-12 into your model, reconcile and flag what does not tie out, build a first-pass return, then hand it to an analyst who owns the assumptions. The AI does the mechanical work; the judgment stays human. This walkthrough runs the full sequence on a sample deal.

The discipline matters because most AI efforts never reach a reliable workflow: JLL’s 2025 Global Real Estate Technology Survey found 88% of real estate investors, owners, and landlords piloting AI, yet only 5% report achieving all their program goals.

You underwrite a multifamily deal with AI by feeding it the rent roll, T-12, and offering memorandum, then letting it extract the line items, normalize them, and flag what it cannot verify. The AI populates the model and surfaces risk. You, the investor, still judge the deal and decide.

That is the whole trade, and it is worth being precise about it. The machine types and adds. Your team prices the asset. Below is the workflow I run for a first-pass multifamily underwrite, stage by stage, with a clear line between what the AI owns and what you keep.

The most expensive hour in acquisitions

Every first-pass underwrite starts the same thankless way: a fresh offering memorandum, a rent roll, a T-12, and a long stretch of pure data entry before anyone has an opinion worth having. At real deal volume, that is most of a working day lost to typing instead of thinking. AI is finally good enough to take that hour, and the timing is not academic. JLL's Global Real Estate Technology Survey found the share of CRE firms using or piloting AI climbed from under 5% to 92% in three years (JLL survey). When nearly every institutional peer is piloting AI, owning a clean underwriting workflow stops being an edge and starts being table stakes.

The workflow, stage by stage

Here is the map. Each stage hands the mechanical work to the model and keeps the judgment with you.

Ingestion and normalization

The first bottleneck in multifamily underwriting is always the rent roll. Sellers hand it over in whatever shape it happens to be in: a clean export sometimes, a scanned PDF that has survived three ownership transfers more often. The model extracts it into a consistent table and, more importantly, flags the things that hide behind a headline occupancy number: units on month-to-month leases, expirations clustering in the near term, and concessions quietly inflating gross income. If you are still comparing tools for this layer, our guide to the best multifamily underwriting software lays out the field.

Rebuild an honest NOI

This is the stage that separates a real underwrite from a pretty one. Have the model rebuild in-place income the way you would if you had the hour: drop the model and employee units, strip the concessions, reset the tax line to your basis instead of the seller's, and normalize the expense load. Now the seller's headline number and the honest one sit side by side, and the gap between them is the entire conversation. The AI does the rebuild. You set the assumptions it rebuilds on.

Market context and the pro forma

With a clean rent roll and an honest NOI, the model assembles market context from your own comp sources and populates a multi-year pro forma and a sensitivity view, so you see a range of outcomes rather than a single-point estimate that is almost certainly wrong. It proposes the mechanics. You own the inputs that matter: the cap-rate assumption, the growth assumption, and the exit. Wired into your stack, this is exactly what an AI underwriting copilot should do for an acquisitions team, speed without the invented numbers.

The rule that makes it trustworthy

Here is the part people skip, and it is the one that turns a handy tool into a real liability. A model will fill an empty cell with a confident, plausible number every single time you let it, and in an underwrite a made-up rent is worse than a blank one, because it looks finished. So you hand the model one hard rule: use only what is in these documents, and where a number is missing, write a flag that names exactly what to go verify. Never a guess. Built that way, every figure in the output is either sourced or flagged for diligence, and the pursue-or-pass call stays entirely yours.

Where this fits your team

None of this replaces your analysts. It moves their hours off data entry and onto judgment, which is the only place their time was ever worth much. A custom workflow also does not start as a giant platform build: a first scoped build lands around 5,000 dollars and earns its keep the first week it hands the typing back. If you want to see where an AI-assisted process fits your portfolio strategy, start with how we work with investment teams, or compare the wider field in our guide to the best AI tools for CRE underwriting.

Frequently asked questions

How do you underwrite a multifamily deal with AI?

You feed the model the rent roll, T-12, and offering memorandum, and it extracts the line items, normalizes them into a clean structure, rebuilds an honest in-place NOI, and flags anything it cannot verify. It then populates a multi-year pro forma and a sensitivity view from those inputs. What it never does is decide. The pursue-or-pass call, the assumptions, and the judgment on which flags matter stay with your investment team. The point is simple: the model proposes, your team decides.

What does the AI actually do, and what does the human still own?

The AI owns the mechanical work: reading the documents, structuring the rent roll, normalizing expenses, assembling comparable rents, and populating the pro forma and sensitivity view. It also flags what it cannot source instead of inventing a plausible number to fill a blank. The human owns judgment: which comps are genuinely comparable, which assumptions to run, which flags change the thesis, and the final decision. AI removes the typing so your analysts spend their hours on the thinking that actually prices the deal.

Will AI just make up rents and cap rates when a number is missing?

It will if you let it. A model fills an empty cell with a confident, plausible figure every time, and in an underwrite a made-up rent is worse than a blank one because it looks finished. The fix is a hard rule: use only what is in the documents, and where a number is missing, write a flag that names exactly what to go verify. Built that way, every figure in the output is either sourced or flagged for diligence, never guessed.

How much does an AI multifamily underwriting workflow cost to build?

It depends on scope, but a first scoped build lands around 5,000 dollars rather than the cost of a full platform. Starting small is deliberate: a first-pass underwriting workflow earns its keep quickly by handing your analysts back the hours they lose to data entry on every deal. You expand from there once the workflow is proving itself inside your process.