Three stages.
One deterministic pipeline.
Most brokerages run on instinct and inbox triage. We run on a typed pipeline — every input, every market data pull, every back-out deduction is captured in an audit log. The same inputs always produce the same valuation, and the same evidence package goes to every capital partner.
Intake
- Address, purchase price (or existing loan), renovation budget if value-add
- Mid-FICO, investor experience, entity formation, and program preference (or "recommend one")
- Sponsor REO schedule and reserves position for permanent or agency files
- Submitted via apply or through a broker channel — the form normalizes both
Underwriting
- Live appraisal: geocode subject, pull FY 2026 HUD FMRs, CBRE H1 2025 cap rates with MSA overlay
- Stabilized pro forma built; back-out cascade applied (hard reno, soft 12%, lost rent, OpEx shortfall, marketing, EI at 15%)
- Pencil run against each program: which tier supports the requested leverage at the cleanest rate
- Borrower file scrubbed — credit report, REO schedule cross-checked, exit plan verified against the cascade
Close
- Term sheet within one business day; signed retainer locks the engagement
- Lender selection: deal goes to two or three matched capital partners simultaneously, best execution wins
- Coordinated docs: title order, hazard insurance binder, appraisal (when lender-ordered), payoff statement, entity guaranty
- Funding wired to title; deed recorded; loan number issued; servicing handed off to the lender's portal
A typed pipeline,
not a black box.
Connected Capital's underwriting runs on a three-agent architecture: a Research agent that pulls market data, a Logic agent that computes the valuation, and a Design agent that renders the report. Each agent has a typed input and a typed output. The orchestrator passes one agent's output as the next agent's input. Same deal, same inputs, always the same valuation — and every step is captured in an audit log we can play back to any lender's underwriter.
Pulls the market
- Geocode subject to county + MSA
- FY 2026 HUD Fair Market Rents by ZIP / county
- CBRE H1 2025 cap-rate survey + MSA overlay (Class A 4.75–5.00%, Class C ~5.38%, value-add ~5.20%)
- Parcel data (assessed value, year built, unit count when public)
- Returns a typed MarketContext
Computes the valuation
- Stabilized pro forma: GPR → other income → vacancy/collection → OpEx → NOI
- Direct cap at MSA-adjusted rate → As-Stabilized value
- Back-out cascade per Appraisal Institute method
- Sensitivity range surfaces low / point / high cap rates
- Returns a typed Valuation
Renders the report
- Cascade waterfall — every formula shown inline
- PDF for the borrower file and the lender package
- Per-deal audit trail (outputs/<deal>/audit.jsonl)
- Same render in dashboard and in the term-sheet package
- Returns a typed Report
Why this matters: when a lender's underwriter asks "where did the cap rate come from?", we can show them the exact CBRE table, the exact MSA adjustment, and the exact line of the engine that pulled it. That's the difference between a brokerage running on instinct and one running on a methodology.
The back-out cascade.
Eight lines, every time.
Value-add appraisal hinges on one move: take the As-Stabilized value and subtract a bundle of deductions to get back to As-Is. The Appraisal Institute lays out the bundle in The Appraisal of Real Estate, 15th ed., Chapter 29; MAI appraisal firms (CBRE, Integra, JLL Valuation Advisory) all structure it the same way. Our engine encodes that bundle as eight named deductions — none missed, none double-counted.
The waterfall card to the right is one real cascade run from the engine — a sample 20-unit Bronx walk-up, Class C, value-add. Every line shows its formula and the source data behind it. That same package goes to the lender.
Ready to run a deal?
Start with the appraisal engine if you want the pencil before the conversation. Start with apply if you have a specific program in mind. Either way, the file lands in the same pipeline.