Case Study

A $60M Distributor. Four Months. The Entire Stack.

The client: a food distribution company with roughly $60M in revenue, 150 employees, and four facilities across the country. Growing fast, but running on spreadsheets, disconnected tools, and an expensive outsourced IT vendor. One operator rebuilt all of it.

The Starting Point

Growth Had Outrun the Systems

Finance Closed by Hand

Cash position lived in a manual Excel workbook, closed every Friday. Month-end accruals took a full day of one person's time. Nobody could see cash in real time.

Data Everywhere, Truth Nowhere

The ERP, CRM, bank, freight, and payroll systems each held a piece of reality. Reports disagreed with each other, and leadership had learned not to trust the numbers.

Vendor-Dependent IT

A managed service provider billed roughly $200K a year while core risks went unaddressed. ERP customizations cost $1,750 per field through another vendor. Cloud spend had no budgets at all.

Silent Failures

Customer invoices had been bouncing undelivered for years without anyone noticing. Scheduled reports failed quietly. Errors surfaced only when someone downstream complained.

The engagement in one sentence
"Rebuild the technology foundation of the entire company: finance, data, applications, AI, security, and the vendor stack."

Nothing was out of scope. That breadth is exactly what made it work: when one person owns the whole system, the pieces actually fit together.

What Got Built

Five Layers, One Integrated Stack

01: Financial Operating System

From a Friday Spreadsheet to a Live Finance Platform

The CFO's manual reconciliation workbook became a role-based financial operating system: live bank feeds over a secured treasury API, the ERP general ledger, and payroll, unified in one place the CEO and CFO use every day.

13-week rolling cash forecast on live data
Weekly check-run packet, automated and penny-exact
Month-end freight accrual: a full-day manual drill, now automated
Commission tracking pulled straight from the ERP, payout-gated on collections
Corporate income tax forecasting built into the same platform
02: Data Platform

One Warehouse, One Version of the Truth

A Microsoft Fabric data warehouse now ingests the entire ERP nightly, organized into a clean medallion architecture with per-department security. The pipelines monitor their own freshness and raise alerts when anything goes stale.

133 ERP source tables, roughly 15 million rows
Bank, payroll, CRM, and freight data landed alongside the ERP
Finance data layer that nets to zero against the general ledger
Git-based deployment pipeline, DEV to PROD
Capacity right-sized on day one, cutting warehouse cost
03: Application Fleet

25+ Applications, One Sign-On

Purpose-built web applications replaced spreadsheets and aging server scripts across every department: an operations platform for production and inventory, a sales intelligence portal, and a real-time technology cockpit for leadership. Existing staff-built prototypes were hardened into production systems instead of thrown away.

Operations platform: production breakdown, inventory, expiring lots, freight documents
Sales portal with live ERP data: reorder gaps and book-of-business per rep
Leadership cockpit: system health, costs, identity, and CI/CD in one screen
Company-wide single sign-on and role-based access
In-app feedback with AI triage across the whole fleet
04: AI Layer

AI Grounded in the Company's Own Data

Not a chatbot bolted on the side. Company-wide AI assistants answer from the ERP, the document library, and the CRM through custom-built connectors, with a hard policy wall keeping financial data away from staff queries. Even employee training videos are produced end to end by AI.

Org-wide AI assistant grounded on ERP, documents, and CRM
Finance-scoped AI agent for leadership questions
Custom ERP connectors purpose-built for AI access, read-only and guarded
App onboarding video written, narrated, and rendered entirely by AI
Self-healing loop: user feedback becomes AI-authored fixes for review
05: Audits, Security & Vendor Takeout

Found Money and Closed Risk

With the data unified, the audits got easy. Every disbursement, identity, and vendor contract was examined. Some findings paid for months of the engagement on their own; others closed risks the company didn't know it was carrying.

$30M+ of disbursements audited against bank and ledger records
Recoverable payment errors found in the first pass
$2M+ of commission exposure flagged before payout
$2M cyber insurance policy bound, with the coverage-voiding gap caught and fixed
$200K+/yr in vendor and infrastructure savings identified
Years-old silent invoice-delivery failures surfaced and closed
The Complete Build

Five Layers Was the Summary. Here's Everything.

The engagement delivered more than 80 discrete systems, integrations, audits, and automations. The full manifest, grouped by discipline:

Finance & Treasury

  • Financial operating platform (live bank + ERP + payroll)
  • 13-week rolling cash forecast
  • Automated weekly check-run packet, penny-exact
  • Month-end balance-sheet reconciler
  • Month-end freight accrual engine (18/18 parity tests)
  • Commission system, collection-gated payouts
  • Corporate income-tax forecasting engine
  • Canonical metrics layer (44 metrics, one source)
  • AR collections scoring engine (88% accuracy)
  • Outstanding-checks reconciliation, automated
  • Corporate-card spend visibility
  • Card-payment (gateway) reconciliation
  • Fully-loaded margin-basis methodology

Banking & Payments

  • Bank treasury API integration (mTLS + OAuth)
  • Webhook activity feed + 12-month backfill
  • Payroll API integration (OAuth2 + mTLS)
  • Payment-gateway feed & reconciliation
  • Spend-management platform evaluation

Data Platform

  • Microsoft Fabric medallion warehouse (133 tables, ~15M rows)
  • Clean-slate rebuild across 5 departments
  • Data control plane (catalog + lineage + telemetry)
  • DEV to PROD Git deployment pipelines
  • Nightly refresh + self-monitoring freshness watchdog
  • PySpark incremental ingestion pipeline
  • Capacity right-sizing (~$525/mo saved)
  • Seven source-system ingestions
  • Warehouse reconciled to the general ledger
  • Five semantic models + five AI data agents

ERP Engineering

  • Custom ERP-to-AI server (14 guarded tools)
  • Per-user ERP OAuth connectors
  • Read-only items gateway (1,797 items, rate-limited)
  • ERP upgrade outage root-cause + vendor fix
  • Weekly inventory report rescue
  • Custom ERP extensions (tiles + embedded BI)
  • Financial-data policy wall for AI queries
  • Six-role ERP security model
  • In-house customization (replacing a per-field vendor)

Custom Applications

  • Operations platform (production, inventory, freight docs)
  • Sales intelligence portal (reorder gaps, book-of-business)
  • Real-time technology cockpit (11 sections)
  • Enterprise sales-ops board rebuild (SSO + MFA)
  • Inherited app fleet hardened to production
  • Legacy tool stewardship + migration playbook
  • Consolidated app-fleet hosting (~$25-45/mo)
  • In-app feedback + AI-triage release system

AI Agents & Automation

  • Org-wide AI assistant (grounded on ERP + docs + CRM)
  • Enterprise-search agent for every seat
  • Finance-scoped conversational agent
  • Composed buyers/reps agent (eval-gated ≥90%)
  • Production agent fault diagnosis + fixes
  • Email-mining to AI-tool roadmap
  • Self-healing feedback-to-fix loop
  • Multi-agent overnight orchestration
  • Org AI workspace + email add-in rollout
  • Enterprise search indexes (16,749 docs + email)

AI Content & Media

  • HTML-to-video production pipeline
  • Fully AI-produced onboarding video (zero human editing)
  • In-app tutorial + first-run onboarding system

Security & Compliance

  • $2M cyber-insurance policy bound + warranty-gap fix
  • Money-out fraud audit ($30M+ disbursements)
  • Tenant identity & access audit, over-provisioning stripped
  • Credential remediation (secrets rotated, vaults isolated)
  • GitHub org security hardening
  • DNS + email hardening (DMARC to reject)
  • Data-loss-prevention forensics + restoration
  • Policy work (T&E, AI/privacy, CCPA notice)

Diagnostics & Forensics

  • Margin-variance study ($377K variance traced)
  • Bank-fee forensics ($50-70K/yr opportunity found)
  • Invoice-delivery forensics (silent failures since 2023)
  • Inventory discrepancy forensics + guardrails
  • Report regression triage
  • Business operations mapping from ERP change logs

Vendor, Cost & IT Ops

  • MSP replacement program (~$204K/yr targeted)
  • Legacy server decommission (~$9,900/yr)
  • Microsoft licensing audit (~$77K/yr mapped)
  • Azure cost governance (budgets from zero)
  • Vendor deliverable validation vs shipped code
  • Platform evaluations (scanning, freight APIs, B2B commerce)
  • Ongoing break-fix + sysadmin (~38 items/mo)
  • ERP support stream (~32 items/mo)
  • Onboarding/offboarding SOPs
  • Headless mail automation
  • Platform access recovery from departed staff

Strategy & Advisory

  • Technology transformation thesis (exit-readiness)
  • Market-context research (SMB succession + AI adoption)
  • Digital sales-channel design
  • Owner briefing / strategic dashboard

The Platform Behind It

  • AI-generated invoicing (516 signals to 196 items/cycle)
  • Machine-independent AI agent platform
  • Autonomous overnight operations
  • Cost-governed ops console + research pipeline
  • Persistent AI-maintained knowledge base (761+ pages)

Client details anonymized. Every item above traces to a system running in production or a deliverable in hand.

How One Person Does This

The AI Agent Platform Is the Team

This engagement wasn't staffed by a bench of consultants. It was one operator directing a platform of AI agents: agents that build, test, and ship while I review; agents that run overnight against the backlog; agents that watch the systems we shipped and file their own fixes.

Overnight throughput

Autonomous agent sprints shipped reviewed, production-ready pull requests while the humans slept. The backlog moved seven days a week.

Systems that heal themselves

User feedback flows into AI triage, becomes a proposed fix, and lands as a reviewable change. Data pipelines watch their own freshness and raise their own alerts.

Even the invoices are AI

The engagement bills itself: collectors gather the month's real activity across email, calendar, code, and cloud, and AI synthesizes it into a detailed engagement report. The client sees exactly what they paid for.

Time to Full Stack 4 months
Custom Applications 25+
Warehouse Rows ~15M
Live Integrations Bank · ERP · Payroll · CRM · Freight
Operators Required One
Your Business Next

This Playbook Transfers

Distribution, wholesale, manufacturing, services: the pattern is the same. Unify the data, automate the finance operations, build the tools your team actually needs, and put AI where it pays. Client details above are anonymized; specifics available in conversation.

Start the Conversation See the Practice Areas