This piece goes through each deployment with its source, its scope and its caveats. It ends with a practical checklist for admins who have to turn customer-story glow into an actual business case.
Why these numbers need a careful read
Nearly all of the figures come from the vendors' own customer material. Microsoft's customer stories cover Bristol Myers Squibb, Williams, Regal Rexnord and KOHLER. Google Cloud and Google Workspace accounts cover Orange Spain, Santee Cooper, Equifax and Globe Telecom. None of them is an independent audit.
The metrics also differ from case to case:
- Adoption rate: the definition of "adopted" is not standardized.
- Self-reported time savings: these come from surveys or employee estimates.
- Agent counts: they show activity, not value.
- Forecasts: these are expectations, not outcomes.
Keep that in mind throughout. These cases show what organizations are attempting and how they measure it. They do not rank one platform over the other.
Microsoft 365 Copilot deployments
Bristol Myers Squibb: 96% adoption at 30,000 people
Microsoft's customer story, dated October 8, 2026, says BMS rolled out Microsoft 365 Copilot across its roughly 30,000-person enterprise and reports 96% adoption. It also cites more than 400 "AIQ Champions" and 25,000 agents in use.
The change-management details matter more than the headline number:
- BMS built a mandatory learning program called AIQ Academy.
- Employees get private monthly AI Adoption Scorecards showing their own usage.
- Teams share knowledge through Viva Engage, and Viva Insights helps spot which departments are ready.
- Before anyone receives a Copilot Studio license, they must complete an advanced course.
- Purview and Microsoft Agent 365 support governance, under what the company calls a human-in-the-loop model.
The "up to 20% time savings" figure applies only to project managers using one agent. That agent, called RABIT, analyzes meeting transcripts for risks and action items. The company describes the figure as an estimate. It is not a workforce-wide average.
Williams: single digits to more than 90%
Microsoft's September 9, 2026 story describes Williams, a Tulsa-based natural gas company with more than 6,000 employees. Adoption rose from single digits to more than 90%. Copilot went to frontline staff first, and employees built their own agents and linked them through an orchestrator.
Uses include generating plans, interpreting procedures, assessing regulatory requirements, and safety reporting. One operations technician built an anomaly detector with Copilot Cowork. It pulls equipment data reports from an Outlook distribution list and compares them with baseline stats. It flags signs such as rising thermocouple temperatures, falling oil pressure or increasing vibration.
Microsoft's account says "countless hours" were saved, which is not a measurement. Treat this as a strong adoption story and an interesting field-built workflow. It is not a quantified ROI case.
Regal Rexnord: the most concrete numbers
Microsoft's March 2026 Copilot Studio story offers the most specific metrics:
- An enterprise-wide employee agent, RRX GPT, handles more than 2,000 inquiries a month. Microsoft says that saves an estimated 2,400 hours a year.
- A customer-facing agent serves more than 1,000 weekly users, with satisfaction above 80% and often reaching 90%.
- A separate agent for service reps, used by more than 500 of them, is credited with an estimated 12% productivity improvement.
The story also contains a useful engineering detail. An earlier agent built on another platform answered only 40% of employee questions because of limited integration with knowledge sources. After the move to Copilot Studio, it connected to SharePoint and other Microsoft 365 sources. It also integrated with ServiceNow. For multi-tab Excel files, the team built an extraction process that converts spreadsheets to CSV, stores them in Azure Blob Storage, and vectorizes them with Azure AI Search.
Two lessons stand out. Grounding in clean, reachable knowledge sources drove the result. And the 2,400-hour figure is an estimate attached to one agent, not a general benchmark.
KOHLER: 98% in six weeks, with a split scorecard
Microsoft's August 10, 2026 story reports 98% Microsoft 365 Copilot adoption in six weeks. It says 5,000 employees report saving two to five hours a week, and that the company has created 3,000 agents, with hundreds active daily. KOHLER says it beat a goal of 95% adoption in nine months by hitting it within 40 days.
KOHLER's Chief Digital Officer also credits GitHub Copilot with a 25% improvement in code releases and a 75% cut in error defects. Those are developer-tooling claims. They should not be folded into Microsoft 365 Copilot productivity numbers. The reported drivers were executive sponsorship, tiered training and a champions network of more than 200 people.
Vodafone: historical, not 2026
The submitted story also cites Vodafone's 2024 trial of 300 employees. Legal users reportedly saved about four hours a week, and the company planned an expansion to 68,000 employees. I could not confirm those figures against a primary source for this piece. Treat them as older context, and treat the rollout as a plan, not a completed deployment.
Google Gemini deployments
Orange Spain: 1,000+ custom agents
Google Cloud's Gemini at Work 2026 announcement says Orange Spain is deploying more than 1,000 custom Gemini Enterprise agents. Employees in HR, IT, sales and customer service built them to automate incident triage, analyze network issues and retrieve corporate knowledge. Google says this helps non-technical staff automate daily work.
The Google material I reviewed does not quantify financial savings. It also does not confirm two claims in the secondary report: near-100% license activity within 30 days, and a specific BigQuery, Cloud Storage and RAG architecture. I have left those out.
Santee Cooper: a forecast, not a result
Google describes Santee Cooper as South Carolina's state-owned electric and water utility. It is deploying a specialized model built on Gemini Enterprise for financial planning and grid reliability. Teams use natural language to automate financial scenario modeling. They expect modeling speed to improve by up to 75%, in connection with a $10 billion grid-expansion budget.
The verb is "expect." This is a projection. The $3 million annual-savings figure in the secondary coverage is not confirmed by the Google source I reviewed, so I am not repeating it.
Equifax: trial results from 2025
Google Workspace's customer account, dated June 24, 2025, covers a Gemini trial run with SADA. It had more than 1,500 participants over six months. Results:
- 97% of participants wanted to keep their licenses.
- 90% reported an increase in work quality and quantity.
- Employees in nearly every business unit saved more than an hour a day.
Equifax said it planned a second phase for nearly 22,000 employees and contractors. This is trial-participant self-reporting, and the story dates from 2025. It is not a 2026 organization-wide measurement. The account does stress that Gemini inherits existing Workspace security settings, which matters in a regulated data environment.
Globe Telecom: adoption plus survey figures
Google's customer story reports 80% adoption of Workspace with Gemini. It says employees save about three to four hours a week on average, and that 87% of Gemini users report being more productive. A separate Google summary cites 92% of participating users feeling encouraged to explore new ways of working. These are survey indicators of different kinds, not equivalent measures of verified output.
Globe also built custom chatbots from NotebookLM knowledge bases. It notes that Gemini being included in all Workspace plans removed the overhead of assigning licenses.
What the cost math really says
The secondary report offers an illustrative calculation: $30 per user per month across 1,000 users comes to $360,000 a year. That arithmetic is right, but it is only the add-on.
Several pricing trackers report that Microsoft 365 Copilot for enterprise is $30 per user per month on an annual commitment. It sits on top of a qualifying Microsoft 365 license you must already own. One tracker also notes that a July 1, 2026 suite price update raised Microsoft 365 E3 from $36 to $39 and E5 from $57 to $60, with standalone Copilot SKUs excluded. On those figures, E3 plus Copilot comes to $69 per user per month, and E5 plus Copilot to $90. A separate small-business SKU, Copilot Business, is $18 per user per month promotionally for organizations up to 300 users, with a standard price of $21, through December 31, 2026.
These come from third-party trackers, not Microsoft's own price sheet. Confirm against your agreement, since pricing varies by region and contract. I did not establish a like-for-like Google price comparison. Google's own customer story says Gemini is included in all Workspace plans, so the two vendors' packaging differs. Gemini Enterprise agent technology is a separate product line.
That is the real budget point. Copilot-style licensing can be an add-on cost per seat. Bundled AI shifts the cost question toward agents, data integration, training and governance. The right comparison is total cost of ownership against measured outcomes, not sticker price.
A practical checklist for IT leaders
Use the customer stories as a pattern library, not proof. A few steps follow from the strongest cases:
- Define "active use" before you buy. BMS and KOHLER both tracked usage, but their definitions of adoption are their own.
- Baseline one workflow at a time. Regal Rexnord's numbers are persuasive because they attach to named agents handling counted inquiries.
- Measure quality alongside time. Regal Rexnord tracked customer satisfaction, and Equifax surveyed work quality.
- Fix the knowledge layer first. Regal Rexnord's early agent failed on integration, not on the model.
- Budget for training and champions. Every high-adoption case here had a structured program.
- Plan governance up front. BMS used Purview and Agent 365. Regal Rexnord separated sensitive domains such as HR, legal and finance into distinct Dataverse environments.
- Separate realized savings from forecasts. Santee Cooper's 75% is a projection.
Verdict
Neither vendor wins on this evidence. Microsoft's cases lean on adoption rates, change management and agent counts inside Microsoft 365-heavy workflows. Google's cases lean on employee-built agents, Workspace productivity surveys and specialized models. The different metrics make direct ranking impossible.
What these cases do show is consistent. High adoption follows structured training, visible executive support and peer champions. Measurable value follows from attaching AI to a specific, countable process. Buying seats alone does neither.