← All case studies Custom AI · Case study

An AI Underwriting Agent for Multifamily Acquisitions

How a multifamily acquisitions firm replaced manual deal entry with an AI underwriting agent, cutting underwriting time from 30 minutes to under 5 per deal.

Client
Pre/3
Industry
Headline result
30 → 5 min
Faster Underwriting
90%+
Extraction Accuracy
100+
Broker Formats Handled
About the Company

A multifamily acquisitions firm evaluating hundreds of deals a year.

Pre/3 sources and underwrites multifamily acquisition opportunities at volume, reviewing hundreds of deal packages a year. Every deal arrives as a mix of PDFs and spreadsheets, and almost every broker sends them in a different format.

The Problem

Manual data entry was capping underwriting throughput.

Each deal package had to be read by hand and re-entered into a standardized Purchase Template. Pulling figures from T12s, rent rolls, and offering memorandums ran 25 to 30 minutes per deal, and with roughly 100 brokers sending files in their own layouts, the work never got faster with volume. Throughput was limited by how fast analysts could type, not by how many deals were worth a look.

The Solution

An AI underwriting agent built around Pre/3's own process.

Custom AI Studio built an AI underwriting agent that reads raw deal documents and returns an analysis-ready Purchase Template. Analysts upload OMs, T12s, and rent rolls through a web dashboard. The agent extracts the financials, maps expenses to Pre/3's 12-category logic, and exports to Excel with the template's formulas intact.

Anything the agent isn't sure about gets flagged instead of guessed, so a person confirms the edge cases before export. A memory layer records those corrections and learns each broker's formatting, so the system gets more accurate the more deals it sees. Every processed deal is stored in a proprietary database for later comparison and benchmarking.

The Impact

Real estate underwriting automation that scales with deal flow.

Underwriting that took 25 to 30 minutes now takes 2 to 5, recovering roughly 104 analyst hours a year. Pre/3 can evaluate more opportunities without adding headcount, and accuracy keeps tightening as more broker formats accumulate in memory.

Under the hood.

The end-to-end workflow behind the Pre/3 engagement: how a deal package moves from upload through extraction, categorization, human review, and Excel export as a single coordinated system.

PRE3 Underwriting flow / deal package → purchase template Rev 01 · 2026 Pre/3 · 12-category logic Knowledge inputs Prompts & principles K · 01 Pre/3 categories 12-category map K · 02 Template spec Purchase Template K · 03 Extraction schema T12 · RR · OM fields K · 04 Validation rules confidence + flags Source Deal package OM · T12 · rent roll 00 · Gate Intake parse · OCR · classify Main underwriting flow 4 stages · sequential 01 Extract financial line items 02 Categorize Pre/3 12-category logic 03 Validate confidence check 04 Populate Purchase Template → Deliver Export formulas intact Connected systems Live data & actions Subflow · A Memory read · write Subflow · B Persistence store · log Letta memory corrections Broker patterns learned formats Cloud storage raw docs Deal database every deal Sink Template out analysis-ready — Critical path · gate → deliver Deal package → purchase template

Results.

  • 30 → 5 min — Underwriting time per deal
  • ~104 hrs/yr — Analyst time recovered
  • 90%+ — Extraction and categorization accuracy
  • 100+ — Broker formats handled

Ready to become
AI-Native?

Book a 30-minute conversation. We'll map the highest-leverage workflows in your business and tell you whether AI is the right answer.