Manulife’s Company-Wide AI Rollout: Copilot Access, Employee Training and Governance

The Core Claim: Don't Make Everyone Wait​

McDonald describes a familiar corporate playbook for generative AI. A central team runs pilots, a specialist group gets the tools first, and everyone else waits for a policy, a training course and a business case. His argument is that this leaves most employees spending a year imagining what AI might do to their jobs, instead of finding out what it can do for them.

Manulife did the opposite. It made generative AI tools available across the company from the start. These included Microsoft Copilot and ChatMFC, its internal GenAI chat assistant. Training, support and guardrails were built alongside use rather than ahead of it. McDonald treats AI as "a people transformation more than a technology transformation."

His strongest claim is that this decision did more for colleague confidence than anything leadership said about AI. That is a belief, and the piece offers no comparison group or survey data to test it.

What Manulife Says It Built Around Access​

The account lists several practices that sit alongside the tools:

• Role-based training. Separate programs exist for users, leaders and practitioners. McDonald's example is that an underwriter and a software developer each learn what their own work requires.
• AI Champions. This is a network of strong users across Manulife's markets who share what works and set an example.
• Ongoing learning. Workshops, enablement programs and AI communities of practice run throughout the year, tied to day-to-day work rather than one-off courses.
• Leadership participation. AI is a standing item in team and senior leadership meetings, where leaders practise with the tools and discuss them.
• Internal campaigns. These encourage colleagues to try new uses and share what they find.

The piece gives no participation counts, training hours or assessment results. These are company-described practices, not measured outcomes.

Psychological Safety as an Adoption Control​

The least technical section may be the most practical for IT and HR teams. McDonald says colleagues arrive at very different levels. Some are expert users, some have barely started, and some are reluctant to admit they aren't using AI.

Manulife's stated answer is that no question is too basic. It is fine to challenge what the tools produce and to learn from mistakes, and everyone gets the same chance to learn.

Encouraging staff to question AI output is also a quality control. If people feel unable to admit confusion, they are less likely to flag a wrong answer.

The Numbers, and What They Do and Don't Show​

McDonald reports that Manulife has made GenAI capabilities available to 100% of its workforce. He also says more than 82% of colleagues globally were actively using GenAI tools as of September 2026. The piece doesn't define "available" or "actively using," the measurement window, or the workforce denominator. Availability should not be read as every employee holding the same licences or entitlements.

Other Manulife figures use different geography, dates and measures, so they shouldn't be merged with the 82%:

• A Manulife announcement on August 18, 2026 says 80% of Asia colleagues were actively using AI tools as of June 2026, and that 5.2 million prompts were recorded in Asia in 2025.
• The same announcement says Manulife ranked as the number-one life insurer for AI maturity in the 2026 Evident AI Index for Insurance, for the second consecutive year.
• The Asia Consumer Insurance Awards profile of Manulife Asia's win says the group has moved 129 generative AI use cases into production, with 138 more in development. It also cites an AI Champions Network across eight business functions and a global community of practice of more than 12,000 colleagues.

All of these figures are company-reported or come from award materials. They corroborate what Manulife says about itself but are not independent audits.

The Microsoft Context Behind the Copilot Mention​

McDonald's piece doesn't say which Copilot edition or licences were deployed. Other reporting adds some context. Manulife announced a five-year expanded agreement with Microsoft in July 2026. The agreement includes expanding adoption of Microsoft 365 Copilot to over 30,000 employees and deploying Microsoft Agent 365, described as a control plane for governing, monitoring and securing AI agents at enterprise scale.

Manulife also adopted Microsoft 365 E7 Frontier Suite capabilities as part of the deal. Agent 365 is meant to serve as a registry for AI agents across the enterprise, with a single interface to observe, govern and manage them. Azure and Microsoft Foundry continue to support Manulife's enterprise AI platform for developers and data scientists.

That matters for how to read the opinion piece. "Everyone gets AI" at Manulife sits inside a large, contractually committed Microsoft stack. The "30,000" figure is a Copilot target in a press release. It isn't McDonald's 100% availability claim, and the two shouldn't be treated as the same measure.

Why Open Access Needs Rules: The Work Trend Index​

McDonald argues that people will use AI whether or not an employer provides it, so approved tools are the more responsible course. He supports this with Microsoft and LinkedIn's 2024 Work Trend Index. Microsoft's report confirms the figures he cites:

• 78% of AI users were bringing their own AI tools to work, and only 39% of people globally who use AI at work had received AI training from their company.
• The survey covered 31,000 full-time employed or self-employed knowledge workers across 31 markets. Edelman Data & Intelligence fielded it between February 15 and March 28, 2024.
• BYOAI was even more common at small and medium-sized companies, at 80%.
• 52% of workplace AI users were reluctant to admit using AI for their most important tasks, and 53% worried it made them look replaceable.

There are caveats. This is a survey of knowledge workers, not all employees, and it dates from early 2024. It supports the general logic of providing approved tools. It doesn't show that shadow AI has stayed at 78%, or that Manulife's approach reduced it.

Microsoft's report also says BYOAI puts company data at risk, at a time when leaders' top concern is cybersecurity and data privacy. That is the admin-side reason behind the "open access with rules" argument.

Governance: What Is and Isn't Specified​

McDonald says Manulife's Responsible AI Principles require human oversight of AI systems and additional review for higher-risk uses. He says they sit within governance and risk-management frameworks covering how AI is assessed, approved, deployed and monitored across its lifecycle. These frameworks address privacy, transparency and accountability. Colleagues are also trained to recognise where human judgment is essential.

The piece doesn't list the principles or the risk-review process. The award profile adds that six Responsible AI principles are published openly, and that lifecycle controls are shaped by risk, legal, compliance and privacy teams. Beyond that, there is no operational checklist to report, and none should be assumed.

Practical Takeaways for IT and HR Teams​

This is a company-reported model, not a proven formula. Still, the pieces can be compared with your own rollout:

1. Pair access with learning. Manulife describes training, champions, communities of practice and meetings as parallel tracks rather than prerequisites.
2. Make training role-specific. The underwriter-versus-developer example suggests generic courses aren't the aim.
3. Use peers. Champions networks spread practical habits faster than central announcements, in McDonald's experience.
4. Give permission to question output. Challenging AI results is framed as expected behaviour.
5. Offer approved tools to reduce BYOAI. The 2024 survey shows unapproved use was widespread, but approved tools only help if people understand the guardrails.
6. Define your metrics. Manulife's "available" and "actively using" are undefined in the piece. Your own program should say what counts as active use, over what period, and for which population.

A Balanced Reading​

The case for the approach is that waiting for perfect policy leaves AI use to happen unofficially. It also leaves anxious employees without hands-on experience. Manulife's numbers, even as self-reported, suggest broad uptake is possible.

The case for caution is that this is one executive's first-person account in a trade outlet. The adoption data is undefined and unaudited, and the piece shows no causal link from open access to confidence. A financial services firm with a five-year Microsoft agreement, a dedicated AI organisation and published principles has resources that a smaller organisation may not.

Open access is also not free of risk. Broad Copilot or chatbot availability makes data governance, permissions hygiene and monitoring more important. Manulife's wider Microsoft deal, with an agent registry and governance tooling, suggests it sees those controls as part of the package.

The takeaway is not that open access alone builds confidence. Manulife's account is that access, role-based learning, peer champions, leadership practice, psychological safety and lifecycle governance together build it. The company says it started with the tools and built the rest around real use.