David Bratslavsky on the Future of AI in Commercial Real Estate
David Bratslavsky shares his predictions for the next wave of AI applications in CRE — from underwriting to asset management.
Read post
Back to all posts
AI & Real Estate
David Bratslavsky on Why Multifamily Underwriting Was Broken — And How AI Fixed It
David Bratslavsky on Why Multifamily Underwriting Was Broken — And How AI Fixed It
David Bratslavsky has underwritten more multifamily deals than most analysts will see in a career — and he'll tell you, without flinching, that the traditional process is broken. Not "could be better." Broken. The fact that the industry treats 40-hour underwriting cycles as normal is, in David Bratslavsky's words, "an artifact of habit, not a law of physics."
This post unpacks what he means by that, and how the AI-first workflow inside QuickData.AI replaces the parts that were never adding value in the first place.
The three places underwriting actually breaks
When David Bratslavsky maps a typical underwriting workflow, he sees three failure modes. They show up at firms of every size, from one-person shops to billion-dollar funds.
1. The data-entry tax
Every deal starts the same way: an analyst opens a 60-page rent roll PDF, an Offering Memorandum, and a T12 from a property manager who clearly hates Excel. The next four to six hours are spent typing.
"The most expensive analyst on your team is the one re-typing a rent roll. They just don't know it yet." — David Bratslavsky
2. The reconciliation problem
Property managers categorize line items inconsistently. "Repairs & Maintenance" on one T12 is split across four ledger codes on the next. Underwriters spend hours reconciling categories before the model is even trustable.
3. The assumption drift
By the time the data is clean, the deadline is two days closer and the analyst is exhausted. Assumptions get rushed. The IC memo gets thinner. Decisions degrade.
The AI-first reframe
David Bratslavsky's thesis is simple: if a human can do it without thinking, a model should do it without typing. QuickData.AI was designed around that idea from day one.
- Rent rolls parse to unit-level data with 98% accuracy.
- T12 line items auto-categorize at 97% — and the platform learns each firm's chart of accounts.
- OMs extract directly into the underwriting model, including pro formas and rent comps.
The point isn't to replace the underwriter. It's to delete the four hours that were never adding signal.
What changes when the data layer is solved
David Bratslavsky has watched dozens of firms adopt the platform, and the pattern is consistent. Cycle time drops from 40 hours to 25. Pipeline capacity goes up roughly 40%. The IC memos get longer, not shorter, because the analyst now has time to think.
"When you remove the data-entry tax, analysts don't work less. They think more. That's the whole win." — David Bratslavsky
A note on accuracy
A common pushback David Bratslavsky hears: "97% accuracy isn't 100%." True. But the comparison isn't to a perfect world — it's to a stressed human reading row 412 of a rent roll at 11 PM. The error rate of a well-built model is lower than the error rate of a tired analyst, and it's auditable in a way human typing isn't. Every cell traces back to a source.
What's next
The next step, in David Bratslavsky's view, isn't smarter extraction — it's tighter integration. QuickData.AI is heading toward a world where the underwriting model is no longer a static spreadsheet but a live document that updates as new data lands. The data layer becomes infrastructure.
If you spend any part of your week typing numbers off a PDF, that future is worth paying attention to.
About the author: David Bratslavsky is the founder of QuickData.AI, a fractional CTO, and a Member of the Forbes Technology Council. Connect with him on LinkedIn.