EY’s Microsoft AI Frontier Firm Claims: Copilot Metrics, PowerPost Evidence and Caveats

What "founding member" actually means​

EY says it is one of 14 founding members of the Microsoft-led Frontier Firm AI Initiative, run with Harvard. That is accurate, but the label needs some care.

Faculty at the Harvard Business School AI Institute announced the initiative on November 18, 2025, as a collaboration with Microsoft and its clients. The institute defines Frontier Firms as human-led, agent-operated organizations that put AI at the core of their strategy. The institute was formerly called the Digital Data Design Institute at Harvard (D³), which is why older coverage uses that name. The institute's own launch release notes it was known as D³ until April 2026.

The inaugural cohort takes part in large-scale, field-based experiments on AI-first work patterns. Members also join custom workshops meant to turn the findings into practical guidelines. The named members include Barclays, BNY, Clifford Chance, DuPont, Eaton, Eli Lilly, EY, GHD, Kantar, Levi Strauss, Lumen, Mastercard and Nestlé.

This is a research and learning cohort. It is not a ranking, and it is not a certification that a member has reached Frontier Firm status. In Harvard's own wording, the members are "embarking on the path to become Frontier Firms." EY's page presents membership as proof of arrival. The cohort's own description frames it as a starting point.

EY's account: a four-step build​

EY describes a sequence, not a single product rollout.

1. Cloud and data foundations. EY built EY Fabric on Azure as its global data backbone for itself and its clients. It then worked on making the data accessible, governed and trusted. Its example is EY Helix, where it says it unified its audit data supply chain with Microsoft Fabric. It also says this makes EY one of the largest Power BI users globally, but gives no figure for that.
2. A private generative AI assistant. Using Azure OpenAI, EY built EY.ai EYQ, a private enterprise generative AI environment. EY says 300,000 people had it before an equivalent enterprise product existed on the market.
3. Copilot and employee-built agents. EY rolled out Microsoft 365 Copilot and let staff build their own agents in Microsoft Copilot Studio.
4. Unification. EY says it combined intelligence, orchestration, data, workflows and governance into a bespoke agentic platform on Azure, with other ecosystem partners involved. The page gives no architecture, controls, model choices or measurement method for this platform.

The numbers, and how far to trust them​

EY's page lists these figures:

The page does not say what these are measured against. For most of them there is no baseline, time period or sample size, and no way to tell whether a figure is self-reported or measured in operations.

Be careful with the Copilot headcount. The page says Copilot was deployed to 365,000+ employees. Microsoft's May 2026 announcement says EY initially deployed Copilot to 150,000 users, saw a 15% productivity boost among them, and is now scaling across its roughly 400,000 people. So the 15% figure appears to come from the earlier, smaller group. Applying it to the whole firm would overstate what has been shown.

Where the detail is best: PowerPost​

PowerPost is the one claim with real implementation detail behind it, thanks to a Microsoft customer story from November 2025.

• The problem. The previous general ledger posting process was semi-manual and varied by region. The story also says it relied on a legacy third-party tool with substantial licensing and maintenance fees.
• The build. The app uses Power Apps. A pre-built SAP connector and the Microsoft data gateway link it to EY's on-premises SAP ERP. Users prepare journals in the tool or upload Excel templates. A Power Automate cloud flow loads the data into Dataverse, and users then submit for approval or posting.
• Controls. One flow handles several posting types, including foreign exchange postings. Approval workflows can be tailored by country, and controls restrict posting to authorized staff.
• The agent. A Copilot Studio agent guides users through posting and eases bulk uploads. Microsoft's story gives an example where submitting 20 journals takes about 15 minutes in the app and under a minute with the agent.
• Scale. PowerPost is used by more than 2,000 EY employees. The 95% lead-time reduction and 37% cost cut are tied to real-time data entry replacing a separate create-then-enter process, and to savings on processing time and third-party licences.

A lot of the saving here comes from retiring a legacy tool and redesigning the process. That is a different thing from "adding an agent" and should not be read as a general Power Platform result.

Corroboration, and its limits​

Microsoft's May 21, 2026 announcement describes EY and Microsoft investing more than $1 billion over five years in a new initiative. It pairs Microsoft Forward Deployed Engineers with EY industry professionals. That release repeats the PowerPost outcomes and describes the EY Canvas multiagent framework, which it says covers the workflows of 130,000 assurance professionals. It also reports Azure AI Document Intelligence on EY's Global Tax Platform, cutting manual workload by up to 90%.

Both companies profit from this story, so treat it as consistent messaging between partners. It is not independent validation.

What admins and IT leaders can take from it​

Here is my analysis, drawn from general industry practice and not from EY's page.

• Sequence matters. EY's order was data platform, governed data, a secure assistant, Copilot, then agents. Teams that start with agents on messy data tend to stall.
• Pick the workflow, then measure it. The best-documented win is one defined finance process with a clear before and after.
• Governance scales with citizen development. 80,000+ employee-built agents implies serious lifecycle, access and data-loss controls. EY does not describe its own. Anyone enabling Copilot Studio broadly should plan this up front.
• Ask vendors for baselines. When a case study cites percentage gains, ask what was measured, over what period and for which users.

Bottom line​

EY's story is a credible example of a large Microsoft shop deploying Fabric, Copilot, Copilot Studio, GitHub Copilot, Power Platform and Dynamics 365 at scale, and the PowerPost detail is genuinely useful. The headline claims are still company-reported, the 15% figure seems to apply to an initial subset of users, and "Frontier Firm" membership is a research cohort, not a seal of approval. Read it as a case study with a sales motive, and test its numbers against your own workflows.