What "founding member" actually means
Faculty at the Harvard Business School AI Institute announced the Frontier Firm AI Initiative on November 18, 2025. It is a collaboration with Microsoft and its clients, and it aims to deepen understanding of how to build Frontier Firms. The inaugural cohort is described as organizations "embarking on the path to become Frontier Firms." That cohort includes Barclays, BNY, Clifford Chance, DuPont, Eaton, Eli Lilly, EY, GHD, Kantar, Levi Strauss, Lumen, Mastercard and Nestlé.
Membership is a research and executive-learning arrangement. Members take part in large-scale, field-based AI experiments on AI-first work patterns. They also join custom workshops designed to turn the findings into practical guidelines. Harvard's own initiative page says the effort aims to uncover what works, "not by declaring who's ahead." That quote comes from Harvard's site as captured in the evidence packet.
So membership is not a certification. EY is an inaugural participant that describes its own journey as a Frontier Firm transformation. Harvard has not awarded it a verified performance grade.
EY's account: cloud and data first
The excerpt EY submitted covers the foundations, and the fuller EY page adds the rollout and results. EY's story starts with plumbing, not chatbots.
- EY Fabric on Azure. EY describes this as a unified cloud platform that serves as the global data backbone for EY and its clients.
- EY Helix. EY says it unified its audit data supply chain with Microsoft Fabric and applied analytics across its services. It also says this makes EY one of the largest Power BI users globally.
- The stated logic. Data has to be accessible, governed and trusted before it can power AI.
This is sensible and familiar to anyone who has watched a Copilot pilot stall on messy SharePoint permissions. EY does not publish dates, architecture, data residency details or any measure of "largest Power BI user," so those claims are EY's own descriptions.
The wider rollout, as EY reports it
The full EY page adds several figures. All of them are EY-reported:
- EY.ai EYQ. It was built on Azure OpenAI and put in the hands of 300,000 people.
- Microsoft 365 Copilot and Copilot Studio. EY says it rolled out Copilot and let employees build their own agents. It reports Copilot deployed to more than 365,000 employees and more than 80,000 employee-built Copilot Studio agents.
- EY Fabric scale. EY says more than 70,000 clients use it and that it hosts 3,000 business apps.
- Productivity. EY attributes a 15% boost across the organization to agents working on tax and legal research, deal management, HR and document drafting.
- Development. EY reports more than 16,000 developers using GitHub Copilot and more than 18 million lines of code accepted. It says its Software Lifecycle Assist Agent drove 70% faster test-case and script creation.
- Finance. EY says PowerPost, built on Power Platform and integrated with its SAP instance, delivered 95% faster lead times and a 37% cost reduction.
- Sales. Dynamics 365 Sales was deployed to 100,000 users with embedded AI sales agents.
- Assurance. EY Canvas, built on Azure, Microsoft Foundry and Fabric, has enhanced more than 160,000 audit engagements and processes 1.4 trillion lines of journal-entry data a year.
Reading the numbers with care
The page gives no baselines, measurement periods or definitions for "productivity," "faster" or "enhanced." That matters for several reasons.
The Copilot counts don't line up. In a May 21, 2026 announcement, Microsoft said EY initially deployed Copilot to 150,000 users and recorded a 15% productivity boost. Microsoft said that gain was reinvested into client delivery and learning. It also said EY was scaling Copilot through Microsoft 365 E7 to its more than 400,000 people. EY's page says 365,000+. These figures reflect different dates and scopes, so don't merge them into one adoption number. The 15% figure is also the same headline number reported at different stages, not a second confirmation.
Agent counts are not value. Eighty thousand agents show that many people built things. They don't show how many are used, trusted or worth their licences.
"Agent-operated" is a framing. EY doesn't name its bespoke agentic platform, diagram it, or explain where humans approve actions or how governance is enforced. Treat it as an organizational ambition, not proof of autonomous decision-making.
Both parties have a commercial interest. The Microsoft release describes the same figures as the basis for a joint offering. A separate WindowsForum analysis of EY's metrics reached a similar conclusion about the missing baselines.
Two separate things not to conflate
The Harvard–Microsoft research cohort is one thing. The Microsoft–EY commercial initiative is another. Microsoft's May 2026 announcement describes more than $1 billion invested over five years. It pairs EY practitioners with Microsoft Forward Deployed Engineers, with initial focus on Finance, Tax, Risk, HR and Supply Chain. It also names several industry sectors. That is a services offering for clients, while the Harvard work is research.
What Harvard's early guidance adds
The most useful counterweight is Harvard's own early insights. They come from an upcoming case by Iavor Bojinov, Raffaella Sadun and Shunyuan Zhang.
- AI adoption needs a structured, frontline experimentation phase. It should be followed by adaptation and scaling. Employees should not be assumed to self-discover good uses.
- Managers must decide what to test, whom to involve, how to codify what they learn, and how to set incentives.
- Harvard's Skill–Will framework sorts employees by AI literacy and motivation. Evangelists, Skeptics, Learners and Resisters need different handling.
- Agents make this harder. They require onboarding, including codifying tacit knowledge and building agent-management skills. They can also create sharper winners and losers.
EY's four principles are consistent with this. They are building AI into the existing ecosystem, redesigning work rather than just automating it, making human–agent collaboration the foundation, and pairing governance with cultural change. Those principles are EY's own.
Takeaways for Microsoft admins and IT leads
- Start with data governance. EY's story puts Azure and Fabric ahead of Copilot, and that ordering is the part most likely to transfer.
- Treat rollout as an experiment. Define what you test, with whom, and how you'll capture what works before scaling licences.
- Measure beyond adoption. Ask for baselines, time periods, quality checks and cost boundaries before accepting any "X% faster" claim, including this one.
- Plan for agent sprawl. If employees can build thousands of Copilot Studio agents, you need ownership, lifecycle and approval rules. EY doesn't describe its own controls, so you can't borrow them.
- Keep the vendor lens in view. A Client Zero story is a reference case, not an independent audit.
Bottom line
EY has built a large Microsoft-centred AI estate, and its data-first sequencing is credible advice. Its Frontier Firm label is self-described, and its outcome numbers are unaudited. The membership shows it is taking part in Harvard's research. It doesn't show that the figures have been checked.