The Agentic Bank

CRE Underwriting Agent

⬡ Plumb Spreads rent rolls and appraisals into property-level credit analysis.
◆ Autonomous Specialist

Reads the rent roll, the operating statement and the appraisal lease-by-lease and builds the property-level economics: net operating income, debt-service coverage, loan-to-value, stabilized vs. as-is. Ties each tenant back to its rent-roll line, cross-checks appraised value against market cap-rate comps, and escalates DSCR/LTV outside policy bands to the chief-credit agent.

Memory

Working The rent roll + operating statement + appraisal under analysis.
Episodic Prior periods for the property and comparable assets in the portfolio.
Semantic CRE credit policy, DSCR/LTV thresholds, market cap-rate ranges.
Procedural Lease-abstraction and NOI-normalization playbooks.
Store Property model store + comparables datastore

Orchestration

pipeline MCP

Harness · Managed Agents … long-running session; structured note-taking for the property model; sandboxed code execution for the cash-flow build.

Tools

{ } Document intake + OCR API ›_ Cash-flow model build Code exec Market comps + cap-rate data MCP CRE credit policy (RAG) Retrieval

Evals & guardrails

  • Rent-roll totals tie out to the operating statement, or it flags the discrepancy.
  • DSCR/LTV outside policy bands forces escalation to the chief-credit agent.
  • Appraisal value sanity-checked against market comps; outliers flagged.

Offline reflection

Learns which appraisal formats and lease structures cause abstraction errors and refines the lease-abstraction playbook accordingly.

Frontier edge

  • World-model simulation of the property: stress-tests NOI and DSCR against vacancy spikes, rate moves and cap-rate expansion before committing a valuation, not just the base case.
  • Native multimodal reading: parses the rent roll table, the appraisal's photos and the site-plan diagram together, tying each tenant to its lease line visually.
  • Continual learning (eval-gated): self-edits the lease-abstraction playbook from judge-agent corrections, gated against verified underwrites before promotion.

A sample run

Trigger Refinance request on a 40-unit multifamily property with a fresh appraisal.
  1. 1Abstract the rent roll lease-by-lease; normalize the operating statement.
  2. 2Build the cash flow; compute NOI, DSCR at the proposed debt, and LTV.
  3. 3Cross-check the appraised value against market cap-rate comps.
Output A property-level credit analysis with DSCR 1.28x and LTV 68% inside policy, fed to the memo agent; a below-threshold DSCR would instead escalate to the chief-credit agent.

In numbers

< 4 min
Median rent-roll abstraction
320
Property underwrites / week

Handoffs

Across ⇢ Risk → Credit Risk for policy + valuation governance

More on the Commercial Real Estate desk