Case Study

MS365 Copilot OS

AI-designed. Zero-AI runtime. Corporate IT approved.

B2B Sales / M365 Automation / Deterministic Systems
The Problem

Corporate IT locks everything down.

CrowdStrike monitors every process. No admin rights. No external APIs. No AI tools allowed in production workflows. But the operator needs automation yesterday.

🛡

CrowdStrike Active

Endpoint protection monitors every executable, script, and network call. Nothing runs that IT doesn't sanction.

🔒

No Admin Rights

Standard user account. No software installation. No registry access. No system-level scripting. PowerShell runs user-scoped only.

No External AI

No Claude. No ChatGPT. No external APIs in production workflows. Corporate policy, full stop. The AI can never touch the network.

The Thesis
"We use AI to build systems that don't need AI."

AI designed the logic. Rules execute it. Zero AI at runtime. Zero hallucination risk. Zero probabilistic uncertainty. Fully auditable. Fully deterministic. Runs at greater than 6 sigma because it's pure logic.

0 AI calls at runtime
0 Hallucination risk
100% Deterministic execution
>6σ Process capability
The Deliverable

What was built.

A complete operational intelligence layer: 786-line blueprint + 259-line operating system. Runs entirely inside M365.

4

Power Automate Flows

Deterministic automation running on Microsoft's infrastructure. No external dependencies. No API calls outside M365. IT-sanctioned, enterprise-grade.

  • RFQ intake and routing
  • Quote submittal tracking
  • PO processing pipeline
  • Meeting follow-up automation
3

PowerShell Scripts

User-scoped data processing scripts. No admin privileges required. No installation. Pure logic transforming data inside the allowed perimeter.

  • Data extraction and transformation
  • Pipeline state management
  • Report generation
6

Outlook Rules

First layer of the email triage architecture. Rules fire before the operator reads a single email. Sort, flag, categorize -- deterministically.

  • Priority sender routing
  • RFQ/quote auto-categorization
  • Customer response flagging
  • Internal noise filtering
  • Action-required escalation
  • FYI deprioritization
7

Seeded Workflows

End-to-end processes documented, tracked, and ready for progressive automation. Each one moves through the iteration pipeline.

  • RFQ processing (9-step, 3-tier)
  • Quote submittal
  • Purchase order handling
  • Meeting follow-up
  • Pipeline review
  • Prospect research
  • Expense report
Deep Dive

The 9-Step RFQ Workflow

Three tiers of execution: manual judgment, Power Automate determinism, and Copilot prompts for edge cases. Each step classified by what runs it.

Manual

1. Receive & Classify

Operator classifies the RFQ: standard vs. engineered, stock vs. custom, budget range.

Automate

2. Log & Track

Power Automate captures metadata, creates tracker entry, assigns due dates.

Automate

3. Route to Vendor

Deterministic routing based on product category, vendor territory, and workload balance.

Manual

4. Vendor Follow-Up

Relationship management. Timing, tone, and escalation are judgment calls.

Automate

5. Quote Assembly

Pull vendor pricing, apply margins, format to company template. Deterministic math.

Manual

6. Pricing Review

Strategic pricing decisions: competitive positioning, volume discounts, relationship pricing.

Copilot

7. Cover Letter Draft

Copilot generates customer-facing language from structured data. Human reviews before send.

Automate

8. Submit & Confirm

Automated submission, delivery confirmation, tracker status update.

Automate

9. Follow-Up Cadence

Scheduled follow-up reminders at 3, 7, 14 days. Escalation rules for stale quotes.

Manual Judgment (3 steps)
Power Automate (5 steps)
Copilot Prompt (1 step)
The Method

Manual reps. Templatize. Automate.

Every workflow follows the same iteration pipeline. No premature automation. No theoretical flows. Real work first, then systematize.

01

Manual Reps

Do the work by hand. Understand every decision point, every exception, every edge case. Build the muscle memory that reveals the real process -- not the imagined one.

02

Templatize

Extract the pattern. Document the steps. Create templates, checklists, standard operating procedures. Make the implicit explicit.

03

Automate

Only now: build the automation. Power Automate flows, PowerShell scripts, Outlook rules. The automation encodes proven logic, not assumptions.

15 Systems Mapped

Every system the operator touches was inventoried with access status, automation potential, and constraint level.

Outlook Teams SharePoint Power Automate PowerShell Copilot Excel / OneDrive OneNote Edge Browser OneDrive Planner Power BI Corporate CRM Terminal (User) Task Scheduler
Results

The numbers.

786 Lines: Operational Blueprint
259 Lines: Operating System
4 Power Automate Flows
3 PowerShell Scripts
6 Outlook Rules
9 RFQ Workflow Steps
7 Workflows Seeded
15 Systems Mapped
0 AI Calls at Runtime

The Design Pattern

Design Phase Claude on MacBook Off-network. Designs logic, writes flows, authors scripts.
Human carries deliverable
Runtime M365 on Corporate Laptop Deterministic. Auditable. No AI. No external calls.

The AI never touches the corporate network. It designs the system. Humans deploy it. Rules execute it. The result: enterprise-grade automation that passes every compliance audit because there is nothing probabilistic in the runtime stack.

Your Move

Locked-down environment? We build for that.

VindexAI designs operational systems that run where your IT team says they should -- inside your approved stack, with zero external dependencies.