Crowe brings audit-ready AI to lease accounting with Copilot Studio and Azure

When “finding the clause” becomes the job

When lease accounting teams need answers, they usually start the same way: open a PDF and start hunting. The terms that matter most—effective dates, payment frequency, renewal clauses—are often buried across hundreds of pages, sometimes in multiple languages, with amendments layered on top. For knowledge workers, it’s a scavenger hunt and it painfully steals time from the work that requires judgment.

Crowe operates in a world where proof matters. For Crowe, the difference is applying AI with an auditor’s mindset: building workflows that can stand up to scrutiny, not just generate an answer. As a public accounting and consulting firm delivering audit, tax, and consulting services, accuracy and evidence aren’t optional—they’re the job. And as part of the Crowe Global network of 200+ independent firms across 130+ countries, that standard must hold at scale.

That’s why, for Crowe, AI can’t just be impressive, it must be defensible. The goal was to replace PDF scavenger hunts with continuous audits teams can trust. Workflows that surface what matters and make it easy to verify the “why” behind every answer.

From experiments to habit

Crowe’s leadership treats Microsoft 365 Copilot as the change‑management lever -because when AI lives inside the tools people already use every day, it stops feeling “extra” and starts feeling normal. Doug Schrock, Managing Principal, Artificial Intelligence at Crowe, has a straightforward playbook: train people, show them examples and put AI where the work happens: “Copilot puts AI where the work happens. People are already in Word, Outlook, PowerPoint, and Excel. Use AI in the flow of that work. It’s a little button—you click it, and you’re using it. That’s how it becomes normal.”

He also describes the current moment as a narrow window: a chance for organizations to move from tinkering to real transformation and build durable advantage. Crowe approached this like a transformation, not a point solution—aligning leadership on outcomes while giving technical teams the guardrails to build in a way the business can trust.

At Schrock’s level, “AI culture” isn’t a poster on a wall. It’s what changes in the day‑to‑day. The fundamentals—customer service, respect for people, delivering with quality, staying true to your promise—don’t change because AI shows up. What changes is behavior: teams stop doing work manually “because that’s how we’ve always done it,” and start using AI to get to a strong first draft faster, improve consistency, and raise quality.

Introducing Crowe’s lease audit AI agent, built on Microsoft Azure

Lease abstraction had become too complex to verify manually at scale. Leases were long, full of amendments, and often spread across multiple languages. Crowe needed a way to confirm that critical terms had been extracted correctly, because a single missed clause could flow into incorrect lease accounting.

That’s where Crowe’s build team stepped in to tackle the problem. Luis Lopez Garay, who leads the firm’s AI forward deploy engineering (FDE) team, teamed up with Thomas Callaghan, who heads up applied AI and machine learning, to create a lease audit AI agent using Microsoft technology. With Azure, the tool can read through lease agreements, pull out the key details, and double-check the payment schedule that drives the accounting.

Instead of making teams comb through every single lease, the AI auditor flags the ones most likely to have issues. It then generates a clear, actionable report so teams can spend their time where it matters most—focusing on higher-risk leases rather than treating them all the same.

At the center of the solution is a reusable capability Crowe calls Crowe Capture, its internal retrieval system designed to pull the right sections of a document quickly and accurately. Lopez describes Crowe Capture as the main building block behind the lease auditor, because the heart of the use case is reliable retrieval: getting the right clause fast and doing it consistently across a large document population.

Callaghan describes Crowe Capture as a document understanding API that ingests unstructured and semi‑structured files, performs OCR and structure extraction, and enables retrieval‑augmented question answering—what he calls document interrogation. The crucial part is trust: it returns answers with citations so users can verify what those sources are.

Why Azure: an enterprise‑ready foundation for document AI

For Callaghan, the decision to build on Azure wasn’t just “because cloud.” It was about having an enterprise‑ready foundation for document AI end‑to‑end: “We chose Azure versus other stacks because Azure provided an integrated, enterprise-ready foundation for document AI.” Instead of stitching together disconnected tools, Crowe built on an integrated foundation so extraction, retrieval, security, and operations could scale as one system.

That foundation matters because this isn’t a one‑off extraction task—it’s a repeatable workflow that needs to extract structure reliably, keep retrieval grounded in evidence, and run with the observability required for client‑facing systems. Lopez summarizes the choice simply: “it’s the enterprise readiness of it.”

Behind the workflow

Crowe Capture runs on Azure services that each play a distinct role. Callaghan explains they use Azure Document Intelligence in Foundry Tools for OCR and table extraction, and Azure AI Search for vector retrieval in their RAG flow. That evidence-first design enables finance and audit teams to move from searching for clauses to validating only the leases that truly require human judgment.

Built for trust:

Crowe emphasizes that platform choice is inseparable from security requirements. Lopez calls out tenant boundaries as critical for data privacy: “Azure lets us keep everything in the client’s tenant. From a data privacy perspective, that matters. Their data stays on their tenant, and many other solutions don’t offer that.”

Built to scale:

Document processing is unpredictable. Latency varies and volume spikes. That’s why Crowe designed for resilience and asynchronous processing. Callaghan explains they need retry and back pressure, so they use scalable workers to process long running jobs reliably at scale, hosted on Azure Container Apps with auto scaling.

He adds that processing asynchronously is a “real key part” of achieving scalability, alongside tuning replicas and compute resources and publishing rate limits to handle high load without sacrificing reliability.

Built for all:

To turn the Azure-based lease auditor into something teams could use day to day, Crowe leaned on Power Apps, Copilot Studio, and Azure Logic Apps as the workflow and integration layer.

Rather than building and maintaining custom integrations, Lopez explains they used Azure Logic Apps and its out‑of‑the‑box connectors to connect to external systems, making connectivity “a lot simpler” without having to build a connector from scratch.

Because the lease auditor runs asynchronously and delivers a report, Copilot Studio is part of the workflow layer—helping Crowe operationalize and integrate the solution quickly, rather than serving as a “chat with your leases” interface. Lopez adds “On the Copilot and Power Apps side, we chose them primarily so we didn’t have to worry about infrastructure. We wanted to spend time solving the business problem—not troubleshooting infrastructure—and that was the right choice. It’s simply speed to value.”

What’s next:

Crowe’s lease audit AI agent delivers impact where it matters most: risk reduction and audit confidence at scale. Instead of asking teams to manually comb through hundreds of pages per lease, the agent continuously reviews lease documents in the background, extracts and validates key terms, and surfaces discrepancies before they become downstream accounting issues. Lopez describes it as “a continuous audit over our lease population,” designed to “find the errors before it’s too late.” And because it produces an actionable report, teams can prioritize attention—focusing on the subset of leases most likely to contain errors—rather than treating every document the same.

Looking ahead, Crowe is evolving Crowe Capture with confidence scoring, stronger vision for charts and images, and Graph RAG to better understand relationships across documents. Doug says, “Redefine what’s possible. That doesn’t happen through just doing things the way we used to do but trying a little bit harder… you’ve got to have trust in stepping into the unknown where you haven’t gone before.” For teams buried in PDFs, the promise stays the same: less hunting, more verification—and AI that delivers outcomes with proof.