FeaturedEquipment Finance — Commercial Loan Servicer
Foundation Protocol Implementation
How a $481.6M commercial loan servicer went from zero documented processes to a fully specified AI-ready operation in 8 weeks.
$481.6M
Portfolio Size
8 weeks
Engagement Length
44
Triggers Cataloged
175
Terms Defined
42
Agents Designed
"This is the foundational use case: it proves you can convert tribal knowledge into automation-ready specifications before you deploy automation. It's the missing layer most AI vendors skip."
Snapshot
- Client
- Commercial loan servicer (equipment finance)
- Portfolio
- $481.6M
- Engagement
- 8 weeks, completed March 2026
- Starting point
- Zero documented processes, zero automation
The Problem
The client managed a $481.6M portfolio with no documented operational layer. Every process lived in someone's head. There were no SOPs, no trigger definitions, no workflow diagrams, and no terminology standards. The operation ran on institutional memory — which meant it couldn't scale, couldn't train, and couldn't automate.
- →Tribal knowledge dependency — no SOPs, trigger definitions, workflow diagrams, or terminology standards across any department
- →Deal completeness failure — 10–12% of deals triggered rework loops; each rework cycle cost 35–55 minutes
- →Multi-system re-entry — agents toggled across 3 systems for basic inquiries; duplicate entry for address changes and schedules was standard
- →Visibility gap — no dashboards; leadership decisions driven by intuition, not data
- →No customer portal — balance inquiries and payoff requests required live calls; payoff quotes tracked in spreadsheets
- →Platform debt — multiple LOS platforms and workflows across servicing clients; complexity compounded with every new client added
What Was Built
The Foundation Protocol doesn't deploy automation. It builds the operational specification layer that makes automation possible — and defensible. Every deliverable below was produced from transcript analysis, process interviews, and workflow observation. Nothing was invented; everything was extracted and structured.
45 Task Specifications
Each task captured: trigger, inputs, owner, decision logic, outputs, system dependencies, and exception handling — built to the NOTBO standard across the full equipment finance lifecycle.
8 Production SOPs
CS & Payoff Inbox Maintenance, Assumption Request Processing, and all 5 DPD-stage Collections voice-agent scripts — each validated and ready for agent deployment.
44 Triggers Cataloged
EVENT, THRESHOLD, TIME, and EXTERNAL trigger types mapped across Origination, Underwriting, Funding/Docs, Accounting/Boarding, Titling, Collections, Servicing, and Insurance/Compliance.
Foundation Protocol Dictionary v2.0
175 definitions across 7 categories — acronyms, outcome codes, system/agent terms, collections/servicing terms, operations terms, compliance terms, and platform/vendor references. Sourced from transcript analysis, not invented.
42 Agents Designed
4 agents fully specified to production-ready requirements: Payoff Quote Processing Agent, Morgan (Collections Voice Agent), Customer Service Agent, and UW Data Aggregation Agent.
2 Production Dashboards
Operations Performance Dashboard and Portfolio Risk Monitor (PRM) — single-file HTML/CSS/JS, deployable immediately without additional infrastructure.
Results
0→8
SOPs (NOTBO-validated)
1 (partial)→44
Triggers cataloged
0→175
Terminology definitions
0→42
Agents designed
0→4
Production-spec agents
0→2
Real-time dashboards
Why It Worked
The protocol forced translation before automation design. Institutional knowledge became structured specifications. Undefined triggers became a cataloged inventory. Inconsistent vocabulary became a standardized dictionary. Every decision tied back to an SOP; every escalation tied to a named role and SLA.
The dashboards weren't a separate initiative — they became possible because the underlying operational definition layer existed. You can't build a risk monitor if you haven't defined what risk looks like in your operation. The Foundation Protocol creates that definition layer first.
Client name suppressed per policy.