How a Multi-Company Construction & Industrial Conglomerate Built a Suite of Microsoft Copilot AI Agents

Large, multi-subsidiary organizations face a growing AI decision: buy a third-party AI platform, or build intelligent agents on the Microsoft 365 infrastructure you already own? For one family-owned industrial conglomerate spanning construction, manufacturing, warehousing, and e-commerce, the answer became clear. And the results offer a replicable blueprint for any enterprise evaluating Microsoft Copilot AI agents.

Client Overview: Eight Companies, One Microsoft 365 Tenant

Our client is a family-owned conglomerate operating eight subsidiaries from a single campus, all running under a single Microsoft 365 tenant. Their portfolio spans a $700 million commercial contracting company, two manufacturing operations, three asset management and warehousing businesses, and an MRO (Maintenance, Repair, and Operations) e-commerce company that buys and resells industrial spare parts through their own site and eBay.

The IT team, a single director supported by a small internal staff and an MSP, is responsible for technology infrastructure across all eight companies. This unified Microsoft 365 tenant became the strategic foundation for an enterprise AI deployment that no third-party platform could match for security, ownership, or long-term cost efficiency.

The Problem: Three Manual Processes Tailor-Made for AI

Safety Knowledge Trapped in PDFs

The contracting company maintains a best-in-class safety rating, placing it in the top 1% of U.S. construction firms. That standard is backed by detailed, custom-developed safety manuals covering fall protection, PPE requirements, hazard communication, and incident reporting. These manuals existed as PDFs and printed copies distributed to ironworkers, crane operators, and skilled tradespeople in the field.

When a worker on a job site needs to know the exact tie-off requirement for working above 20 feet, or the correct procedure for handling a damaged harness, they were left flipping through a printed manual, calling their supervisor, or guessing. None of those outcomes were acceptable for a company that holds itself to a top-tier safety standard.

Manual Part Pricing in the MRO Business

In the e-commerce subsidiary, a small warehouse team handles the buying and reselling of industrial spare parts, motors, electrical components, PLC controllers, and similar equipment sourced from manufacturers and plant closures. Their pricing process: photograph each part, manually search eBay for comparable sold listings, review historical data, and estimate a listing price by feel.

This is exactly the type of task AI was built to automate. The research was slow, inconsistent, and only scalable by hiring more people to repeat the same manual work. Pricing decisions directly affected the profitability of the business.

Archaic Accounts Payable Workflows

The CFO identified accounts payable as a bottleneck, manual invoice routing, PO matching by hand, email approval chains, and exception handling at nearly every step. The team had been evaluating third-party AP software at $30,000–$50,000. Then they asked a different question: what if an agent could do this instead?

The Strategic Decision: Own Your AI Agents — Don’t Rent Them

When evaluating how to deploy agentic AI, the IT team compared two approaches. One was a third-party AI platform vendor with a polished demo, pre-built agents, and a proprietary web portal. The other was building natively in Microsoft 365 using Copilot Studio, with ESW Company as the technical and strategic partner.

The third-party platform offered multi-model routing, automatically selecting cheaper or more powerful LLMs based on prompt complexity, and a token-based pricing model. Leadership found the agent concept compelling. But the ownership and cost structure told a different story.

ESW articulated the core risk plainly: when you build on a third-party AI platform, you are beholden to their pricing, their platform decisions, and their continued operation. If they raise prices, change their model, or shut down, you have no recourse. With Microsoft-native agents, the client owns everything built — the agents, the configuration, the intellectual property.

• Data ownership: All AI interactions stay within the existing Microsoft 365 tenant — no documents, queries, or responses leave the environment
• Agent ownership: Every agent built is owned by the client, not licensed from a vendor
• Multi-model support: Microsoft Copilot now supports multi-model selection natively: Claude Opus, GPT-4o, Perplexity, and others — with automatic task-based routing, eliminating the key differentiator of third-party platforms
• Cost efficiency: A single service account with a Copilot license can power agents serving hundreds of users, the entire warehouse team, all field workers, the AP department, without requiring per-user Copilot seat licensing
• Governance built in: Microsoft 365 Purview, DLP policies, RBAC, and audit logging were already in place, no additional compliance overhead required

The Five AI Agents: What Was Built and Why

1. Safety Assistant Agent (Deployed)

All custom safety manuals, procedures, and policies were ingested into a Copilot Studio agent. Field workers ask natural language questions on any device “What’s the tie-off requirement on structural steel above 20 feet?” or “What do I do if I find a damaged harness?”  and receive accurate, cited answers sourced directly from the company’s own documents. Every response includes the source section and manual reference. Designed for mobile use on active construction sites.

2. Inventory Valuation Agent (Deployed)

Warehouse staff photograph a part or enter a model number. The agent identifies the component using AI vision, searches eBay and industrial marketplaces for comparable recent sales, calculates pricing trends over time, and returns a suggested listing price with a confidence score. What previously required 15–30 minutes of manual research per part now completes in seconds — with greater consistency and market accuracy.

3. Contract Review Agent (In Development)

The contracting company bids on dozens of projects simultaneously, receiving complex contracts from general contractors and project owners. This agent ingests contracts, flags risk clauses, summarizes terms in plain language, and highlights construction-specific concerns, accelerating legal review and enabling faster bid decisions. Claude Opus is used for complex contract reasoning tasks that require deeper analysis.

4. AP Automation Agent (Planned)

A CFO-prioritized initiative that automates accounts payable workflows end-to-end: invoice ingestion, PO matching, approval routing, and exception handling. Built on Power Automate and Copilot Studio with ERP connectors. This agent replaced a $30,000–$50,000 third-party AP software purchase the team had been evaluating, the AI agent handles the same workflows at a fraction of the total cost.

5. Executive Insights Agent (Planned)

The long-term vision: a natural language interface connected directly to the ERP, allowing construction leadership to ask operational questions without SQL or BI analysts in the loop. “How many jobs last year had a profit margin above $20 million?” “What are labor cost trends on our active projects?” The same agent framework will also serve HR queries across all subsidiaries, policy lookups, holiday schedules, benefit questions, reducing help desk load across eight companies.

Technical Architecture: Intelligent Model Routing on Microsoft Infrastructure

The entire solution runs on the client’s existing Microsoft 365 infrastructure. No data leaves the tenant. The AI processing strategy routes each request to the optimal model based on task complexity:

• Simple requests → GPT-4o Mini: Routine lookups, basic Q&A, and straightforward factual queries route to lighter, faster, lower-cost models
• Complex analysis → Claude Opus: Contract review, multi-document synthesis, and tasks requiring extended reasoning use Claude Opus for materially better outputs
• Enterprise knowledge → Microsoft Copilot: Queries requiring internal data: SharePoint documents, Teams conversations, HR records, ERP data  route through Microsoft Copilot with native graph-based access and full permission inheritance
• Routing → Automatic: Microsoft Copilot’s auto-select mode determines the optimal model at runtime, delivering cost optimization without manual configuration

Governance: Three Core Pillars from Day One

The client’s AI framework was built around three pillars: governance, security, and training & enablement  integrated into the architecture from the start, not added as an afterthought.

• All data stays within the Microsoft 365 tenant with zero external exposure
• Role-based access control scopes each agent to appropriate user groups, warehouse staff access the inventory agent; only safety officers can update safety manuals
• All AI interactions are logged in Microsoft 365 Purview for compliance review and audit trails
• Citation requirements on safety agent responses ensure outputs are grounded in verified internal documents
• Governance council established with cross-functional business leadership to drive AI adoption standards across all subsidiaries

The Phased Rollout: Moving Fast Without Skipping Governance

Phase one focused on managed Copilot deployment, removing unmanaged free Copilot from all machines, establishing governance frameworks, and rolling out licensed Copilot to leadership with structured onboarding. Phase two: agentic AI, ran in parallel, not sequentially. Waiting for governance to be “complete” before deploying any agents would have cost twelve months of competitive ground.

This parallel approach allowed the organization to deploy high-value, well-scoped agents immediately while governance infrastructure caught up, balancing risk management with operational urgency.

Key Outcomes

• Field workers have instant, cited access to exact policy sections on mobile devices, supporting a best-in-class construction safety rating
• MRO inventory pricing that required 15–30 minutes of manual research per part now completes in seconds with higher accuracy
• $30,000–$50,000 in AP software spend avoided by building a custom agent instead
• 100% of AI interactions remain within the Microsoft 365 tenant, no third-party data exposure
• All agent intellectual property is owned by the client, not licensed from a vendor

How ESW Company Can Help Build Your Microsoft Copilot AI Agents

If your organization is evaluating AI agents for construction safety, inventory management, accounts payable automation, contract review, or executive reporting, ESW Company delivers Microsoft-native solutions your team owns — built on infrastructure you already have.

• Microsoft Copilot and AI agent development — Copilot Studio agents designed for your specific workflows and data
• Power Apps consulting services — low-code applications integrated with your AI agent strategy
• Power Automate consulting services — workflow automation that connects agents to your ERP, AP systems, and line-of-business applications
• SharePoint consulting services — knowledge management and document libraries that power your enterprise AI agents
• Microsoft 365 consulting services — governance frameworks, licensing strategy, and tenant architecture to support enterprise-wide AI deployment

Our U.S.-based Microsoft specialists have delivered Copilot agent deployments across construction, manufacturing, warehousing, e-commerce, government, and professional services, always on infrastructure our clients own.

Ready to build Microsoft Copilot AI agents your organization owns and controls?

Contact ESW Company to schedule a free consultation and discuss your highest-value AI agent use cases.