AI cut commercial real estate underwriting from a week to 90 minutes.

A recent case study in commercial real estate: an agentic AI system took deal underwriting that used to run more than a week down to about 90 minutes, with an analyst spending another 20 minutes reviewing the output. Same analysis, a fraction of the time.

The headline is the speed. The point is what the speed lets you do. So here's how the underwriting actually ran.

How it works

Underwriting a deal is really a stack of document work, done in sequence. The agentic version runs that stack and hands a person the result to check.

1. Pull the rent roll apart. It reads the rent roll and normalizes it: who's paying what, on what term, with what escalations, so the income picture is built from the actual leases, not a summary someone typed.

2. Read the leases. The same system read three complex retail leases in under seven minutes, pulling the uses, rents, escalations, and renewal options, and flagging the unusual clauses. CBRE research puts that work at four to eight hours and $150 to $350 per lease when a person does it by hand.

3. Run the comps. It pulls the comparable sales and rents and lays them against the deal, so the analyst is checking a comparison rather than assembling one from scratch.

4. Draft the summary. It produces a structured write-up the investment team can act on: the numbers, the risks, the terms that stand out.

5. Hand it to a person. The 20 minutes of review is the point, not an afterthought. The judgment, whether to bid and at what, stays with the team. The assembly moves to the system.

That's the mechanism. No new data the team didn't already have. The work that used to take a week just stopped taking a week.

You can do this for your own workflows too

Underwriting is the visible example, but the move is the same in every workflow that runs on someone processing documents and keeping track of what matters. Lease abstractions. Rent rolls. Offering memoranda. Cap rate comps. Due diligence checklists. Each one is a job that leaks hours when it runs on analyst time alone, and each one an AI can run end to end. If your work lives in documents, that's the place to point it first.

And it compounds. Every deal your team runs builds a record: what the numbers looked like, what the comps supported, what you bid and why, what closed and what didn't. A team that runs this for a year reads its own criteria more precisely than a team working from memory. The record gets sharper with use.

The part the winners get right

JLL's 2025 survey found that most of the industry is already trying this: 88% of investors, owners, and landlords have started AI pilots, and 92% of occupiers have too. But only 5% say they've achieved all their goals. The gap isn't the technology. It's integration. A pilot bolted onto one task in isolation gives you a faster task. The teams that get more out of it tie the analysis into the actual deal flow, including the relationships behind every deal.

Because faster underwriting means more deals in play, and every deal is a web of people: the broker, the seller, the equity partner, the lender, the local contact who knows the block. The teams that win aren't just quicker on the math. They use AI to hold those relationships at a pace a person can't sustain by hand, so nothing gets dropped while the pipeline grows.

Where Sonzai fits

Sonzai is the relational intelligence layer for real estate. It keeps a living record of every deal, every contact, every commitment, and every follow-up, and puts an AI workforce on top of it that runs the document work and the relationship work together, at whatever volume the pipeline demands. You point it at the workflow that leaks the most, and it runs.

If you want to see what your deal pipeline looks like when nothing slips, book a demo. Bring your current process. You'll leave with a clear picture of what the same team can do with the right infrastructure.