OperationsWorkflow Documentation

Intelligence Workflow

From Dropbox audit to AI-queryable artifact. The full workflow: 6 phases, 10 prompts, 3 outputs, MCP server surface. Complete operational breakdown of how the Windfield Relationship Intelligence Report gets generated.

Author

IO Pipeline

Date

April 22, 2026

Phases

Sequential workflow

Prompts

Specialized chains

Outputs

Artifacts produced

Runtime

~45m

End-to-end

Workflow Overview

Dropbox Audit

Scan and inventory all transaction folders

~5 min

Extraction & Normalization

Parse transaction records from identified documents

~12 min

Analysis & Tiering

Calculate volume metrics, tier clients, identify patterns across geography, property type, and time

~10 min

Report Generation

Produce the structured intelligence report with 19 sections, data tables, tier cards, and ICP profiles

~15 min

The intelligence workflow runs in 6 sequential phases, processing Dropbox transaction records through extraction, normalization, analysis, and report generation. The final output is a structured intelligence artifact surfaced via the Windfield MCP server.

Phase 01

Dropbox Audit

~5 min

Scan and inventory all transaction folders. Map file naming conventions to extract deal metadata.

Recursive listing of Dropbox /Transactions/ directory. Identify closing documents, contracts, amendments, correspondence. Build a manifest of ~3,200 files across 250+ deal folders.

Phase 02

Extraction & Normalization

~12 min

Parse transaction records from identified documents. Normalize names, dates, amounts, property types.

OCR where needed, PDF text extraction, filename parsing. Standardize client names (e.g., 'Block & Co' → 'Block & Company'). Map property types to taxonomy. Deduplicate entries.

Phase 03

Analysis & Tiering

~10 min

Calculate volume metrics, tier clients, identify patterns across geography, property type, and time.

Client tiering by lifetime volume. Property type concentration analysis. Geographic heat mapping. Co-broker network graphing. Year-over-year trend analysis. Agent performance attribution.

Phase 04

Report Generation

~15 min

Produce the structured intelligence report with 19 sections, data tables, tier cards, and ICP profiles.

Prompt chain: executive summary → volume analysis → client tiers → property types → geography → agent performance → co-broker network → playbooks → ICP profiles → opportunities → risks → KPIs.

Phase 05

Review & Polish

~5 min

Fact-check numbers against source data. Verify all client names and contact details. Cross-reference totals.

Automated reconciliation: sum of tier volumes vs. total volume. Transaction count verification. Name standardization audit. Contact detail verification against CRM.

Phase 06

MCP Surface

~3 min

Publish report as queryable artifact. Register MCP tools for chat, search, and relationship lookup.

Store as structured JSON + rendered HTML. Register memory/query, memory/write, properties/get_memory endpoints. Enable Claude to answer questions about the report in conversation.

Output Artifacts

JSON

relationship_intelligence.json

Structured Data

Complete analysis in machine-readable format. 19 sections, all data tables, all metrics. Queryable via MCP.

HTML

RPT_windfield_relationship_intelligence.html

Rendered Report

Full visual report with data tables, tier cards, ICP profiles, property type bars, geographic rankings, and risk matrix.

relationship_intelligence_summary.md

Executive Summary

2-page executive brief. Key findings, top opportunities, immediate action items.

MCP Tools

memory/querySearch

Search the relationship intelligence knowledge base. Returns relevant sections based on natural language query.

memory/writeWrite

Add new intelligence or annotations to the relationship knowledge base.

properties/get_memoryRead

Retrieve stored relationship intelligence for a specific property or client.

properties/chatChat

Conversational interface to the relationship intelligence report.