Building end-to-end workflows with Microsoft 365 Copilot

Enterprises are beginning to treat generative AI not as an isolated productivity tool, but as the connective layer linking business applications, data, and human judgment. Nowhere is this shift clearer than in organizations using Microsoft 365 Copilot as part of a broader architecture that spans CRMs, low-code platforms, and specialized AI systems.

But according to Will McKeon-White, senior analyst at Forrester, it’s not always easy to build that architecture. “Integrations can prove difficult and usually require cooperation between subject matter experts and technical personnel to get them to work right and ensure the Copilot knows how and when to use different integrations,” he said.

Two early adopters — Monica Washington Rothbaum, COO of J&Y Law, and Patty Patria, CIO of Babson College — illustrate how successful implementations look. Their experiences show that the real value does not come from Copilot alone, but from deliberate decisions about integration, data design, governance, and change management.

From drafting assistant to orchestrated workflow: How J&Y Law re-engineered the case pipeline

When Rothbaum arrived at J&Y Law, she quickly realized the firm’s rapid growth in personal-injury cases demanded a more intelligent and consistent process. With a headcount of around 100 and the ambition to scale, the firm needed a way to manage high case volumes without sacrificing accuracy or human oversight. Her background in business growth and IT leadership helped her see an architecture forming across the firm’s systems.

According to Rothbaum, the personal-injury case workflow is a sequence of tightly connected phases: marketing, intake, pre-litigation, demand, settlement, negotiation, and — if needed — litigation. Each step generates critical data that must be captured, structured, and then reused downstream. To support that flow, legal operations platform Litify functions as the base CRM and “helps connect the dots for the case pipeline.”

On top of Litify, the firm layers specialized AI tools. Agentic AI and bots assist during intake, listening to conversations, reviewing written and oral communications, and flagging early “nuggets” of information that may influence liability or case strength later. Another AI platform, Foundation AI, ingests documents and files them in the CRM.

Internally built GPTs (custom versions of OpenAI’s ChatGPT assistant) then extract structured details from medical notes, client communications, and case files — data that is notoriously inconsistent in format. And as a case moves forward, EvenUp, a personal injury claims intelligence platform, evaluates key factors through the firm’s AI playbook, surfacing insights about the strength of the case.

The key, Rothbaum said, is that each tool “owns” a specific part of the pipeline and produces structured outputs that can feed the next stage.

Microsoft 365 Copilot’s role is not to replace these systems but to act as “the bridge between daily workflows and the firm’s structured case data and insights,” she said. Attorneys still work largely in Outlook, Word, SharePoint, and Teams, and Copilot connects those spaces with the structured case data residing in Litify and the insights coming from EvenUp and internal GPTs.

For example, drafting a demand letter used to take around six hours. With the integrated workflow, the process now takes about 45 minutes. Revision cycles have been reduced from “three or four rounds to one or two.”

Despite these efficiencies, Rothbaum insists that AI-generated materials must never appear automatic or unvetted — especially in a legal environment. The firm built a structured file-review system where attorneys receive comprehensive case summaries with internal hyperlinks that allow them to “run down a branch” of the evidence instantly. This system gives lawyers transparency into where information came from, ensures they can challenge or refine AI-derived insights, and helps prevent the appearance of boilerplate AI text.

“You can still tell when something is 75% AI-generated,” she said. “We never want to send that signal to an insurance company.”

While many workflows have been streamlined, two areas remain stubbornly human-dependent. One is medical-record acquisition — a process still largely dominated by phone calls and faxes to hospitals and doctor offices. Although some advances have been made in medical records follow-up, according to Rothbaum, AI cannot yet replicate the human judgment, nuances, and persuasion required.

The other human-dependent workflow is settlement negotiation — an interaction rich with nuance, psychology, and game theory. “AI can forecast or predict,” she noted, but the human-to-human bargaining remains essential.

Still, the orchestration achieved across the rest of the case lifecycle shows what is possible when data is structured, systems are chosen deliberately, and Copilot is positioned as the coordination layer rather than the engine itself.

Copilot in higher education: Babson’s institution-wide model for AI orchestration

While J&Y Law demonstrates an operational workflow, Babson College shows what end-to-end AI orchestration looks like in an academic environment that spans thousands of students, faculty, and staff.

CIO Patty Patria oversees all technology for the institution — from academic systems to analytics, security, and even the library. She was one of the first CIOs in higher education to deploy Microsoft 365 Copilot across both academic and administrative units, but she emphasizes that Copilot alone is not what transforms workflows.

“Copilot by itself is really meant to be a task helper,” Patria said. The shift toward workflow orchestration happens when organizations add Copilot Studio, Microsoft’s low-code platform for building bots, agents, and automation.

The university library is the best example. Traditionally, staff spent days coordinating student-worker schedules, especially during finals when the library operates 24/7. Now students submit availability digitally. Copilot Studio processes constraints, preferences, and staffing rules — and a schedule is generated automatically and submitted for final human approval. What once took days now requires minimal manual intervention.

IT and marketing have also seen major gains, particularly in integrations work, code generation, SEO optimization, video production, and analytics. In each case, Copilot acts as the user-facing interface that links people to the operational systems.

But Babson’s most ambitious project is its AI Entrepreneurship Console, a unified experience that helps students brainstorm business ideas, refine them, identify needed coursework, and connect with mentors. The system integrates more than 20 intelligent agents, the student-information system, CRM software, and tools that advise and coach. Students can go from concept to mentored project in a single interface. The agents include ethical guardrails as well — discouraging unethical strategies and guiding users to better alternatives.

Babson’s success did not happen by accident. Patria focused first on leadership training, starting with the president and cabinet. She then implemented institution-wide sessions on prompting, AI principles, and low-code development. Faculty participated in use-case workshops, while staff attended hands-on sessions where they built bots and agents.

“When you get people in a room together from all different disciplines with ideas, you really see the creativity spark,” she said.

According to Patria, the results were dramatic. Copilot interactions grew from 2,000 per week during the early rollout to about 20,000 per week today, she said. And surveys show over 80% of staff reporting increased effectiveness, one to ten hours saved per week, and significantly higher job satisfaction.

5 key lessons for IT leaders

According to Forrester’s McKeon-White, when enterprises connect AI agents to structured data sources, automation platforms, and operational systems, the results are transformative.

“One organization I spoke with, for example, changed how long a common process took from more than a day to around 13 minutes on average,” he said. “But it took a while to configure and scale, and required someone to know first that this was a problem or inefficiency in the first place, and second how to shorten it.”

Often the problem lies in implementation, McKeon-White said. Tech teams make their best guess and don’t engage subject matter experts to find out how the processes really work.

Enterprises seeking to integrate agents in workflows might gain from these learnings distilled from the experiences of J&Y Law and Babson College:

Start with one workflow: Rothbaum advises CIOs to resist the urge to “start with enterprise transformation.” Instead, choose one workflow with clear value, map the sources of truth, and tag data early. “If no one knows where the truth is, AI will amplify the chaos,” she warned.

Build AI literacy across the organization: Patria’s rollout worked because Microsoft Copilot wasn’t introduced in isolation. She trained executives, faculty, staff, and students systematically. The institution developed prompt literacy, low-code familiarity, and disciplined testing practices — skills that enabled rapid scaling.

Treat AI agents as systems of engagement: CRM systems and other legacy platforms continue to serve as the system of record, while specialized AI tools such as domain-specific bots supply the deeper insights. Low-code platforms provide the underlying automation, and the agent sits on top as the human-facing orchestration layer that ties everything together. This layered approach makes it possible to build coherent, end-to-end workflows.

Use governance to protect innovation: Both leaders emphasize that guardrails are essential — not to restrict creativity but to ensure workflow automation remains safe, compliant, and explainable. Structured review processes, classification frameworks, evaluation committees, and transparent communication help maintain trust in AI-supported decisions.

Let humans anchor the process: Whether it’s attorneys reviewing case summaries or faculty providing feedback through voice agents, people remain central. AI accelerates the workflow, but human judgment shapes the outcome.

For both Rothbaum and Patria, AI tools are no longer just assistants — they are now an integral part of critical workflows.

Related reading:

How IT leaders unlock productivity with Microsoft 365 Copilot

From chatbots to colleagues: How agentic AI is redefining enterprise automation

Microsoft drops M365 Copilot price for SMBs, upgrades free Copilot Chat

Microsoft unveils Agent 365 to help IT manage AI ‘agent sprawl’

Agentic AI – Ongoing coverage of its impact on the enterprise