Marketing Already Uses Microsoft Copilot. Is Your Health System Ready?

Your team is already using Microsoft 365 Copilot on patient stories, CRM segments, and service line plans. Know what Copilot can see before a governance gap becomes a brand story.

A healthcare CIO client of mine received an increasingly common mandate from his CEO last spring: begin an AI-driven digital transformation, with Microsoft Copilot at the centre of it.

My client wasn’t skeptical of the mandate. He was committed to it, and to extending his commitment to Microsoft to support it. He stopped the project before it started anyway.

His reasoning fit in a sentence: he didn’t want to find out the wheels had come off right as they hit warp speed. He wasn’t convinced the organisation’s governance could survive a Copilot rollout — and he wasn’t willing to bet the organisation’s reputation on a hunch in either direction.

He needed evidence, fast. That’s where I came in. Most of my career has been spent inside the Microsoft ecosystem — including Avanade, the consultancy Microsoft co-owns with Accenture, plus Rightpoint and Whereoware — with the last several years focused on AI governance. His brief was three lines long: tell me what Copilot can already reach in our tenant, show me proof, and do it in weeks, not quarters.

Accepting his challenge, I embarked on a rapid assessment that paired traditional consulting practice with the speed of AI, auditing his Microsoft tenant across five governance lenses: configuration, licensing, data governance, Power Platform, and Copilot telemetry.

The first thing we found was that the question was overdue. Copilot hadn’t stayed in its pilot. Invite-only access for a handful of directors had spread into ad hoc use across most of corporate — quietly, with nothing governing it. Nobody had broken a rule. There wasn’t a rule.

Almost nothing else we surfaced was about AI. It was about permissions and sharing inside the ordinary workspaces the business ran on every day: shared drives, team sites, folders someone opened up years ago for a reason that made sense at the time. Copilot hadn’t created any of it. Copilot had just made it searchable.

That engagement — and a career spent mostly inside healthcare marketing organisations — is what convinced me this wasn’t a one-off problem. In a health system, marketing produces the highest volume of documents, campaign data, and patient-facing content of any department, which makes it the most exposed to this exact gap, and usually the last to hear about it. That combination — urgent, and not unique to one CIO’s tenant — is why I’m writing this for you instead of your CIO, and why we turned the bespoke process from that engagement into a repeatable audit. What follows is what we found, how the audit works, and how to run it against your own organisation.

1. Why marketing is the highest-exposure function in a health system

Copilot, and AI in general for that matter, doesn’t create a new hole in your environment — it magnifies an existing one. It just gets a lot better at using what’s already there. AI leverages whatever context it has access to, and for Copilot that’s basically everything your organisation keeps in Microsoft: SharePoint, Teams, OneDrive, Outlook, you name it. All of that information was already sitting there. It was just hard to find. Now it’s one plain-language question away. The old “out of sight, out of mind” approach to permissions and file sharing used to be a workable risk posture. It isn’t anymore — with AI, everything is in sight, and in mind.

Now consider what lives in marketing’s footprint at a multi-site health system:

• Patient stories and testimonial files — consent forms, unredacted intake notes, clinical detail forwarded by a service line to “help with the story.” The most brand-valuable content you produce is also the most PHI-adjacent.

• CRM segments and campaign lists — exports staged in SharePoint for an agency or a media partner, often with diagnosis or service-line proxies attached, often long past their useful life.

• Service line growth plans — volume targets, contribution margin, downstream revenue models, and the honest internal assessment of which sites are underperforming.

• Physician recruitment and provider brand work — comp discussions, named candidates, and which practices are being courted or wound down.

• Unannounced news — de novo openings, closures, service line exits, affiliations, rebrands, and the crisis comms holding statements written for the scenario you hope never happens.

• Agency SOWs and media budgets — rates, retainers, and performance by partner, sitting in a folder shared to a link that once said “anyone in the organisation.”

None of that is a Copilot problem. It’s a permissions problem that Copilot turns into a discoverable one. And when it surfaces, it doesn’t surface as an IT incident. It surfaces as a physician asking why a coordinator saw the oncology growth plan, a partner asking how their rates got quoted internally, or — worst case — a patient asking why their story was in a system they never consented to.

2. So what is really at stake for healthcare marketing teams?

Brand trust, which is the whole asset. Healthcare marketing runs on the premise that the institution can be trusted with the most personal thing a person has. The scrutiny that has landed on tracking pixels and patient data flowing into ad platforms is the same scrutiny that will land on AI tooling. A governance failure involving patient data is not a compliance line item — it’s a paid-media pause, a press cycle, and a trust rebuild you didn’t budget for.

Velocity, which you lose either way. When nobody can prove the environment is governed, legal and compliance default to no. Every AI-assisted workflow your team wants — campaign variants, patient-comment analysis, physician bio production, translation across markets — gets held at the door, not because it’s risky but because it’s unprovable. Meanwhile the shadow usage continues, so you carry the exposure without capturing the productivity.

Budget you’ve already spent. Governance work has a habit of surfacing paid licenses on accounts nobody uses, security tiers already purchased but never enabled, and access that predates two reorgs. In a multi-site system with acquired entities, that accumulates quietly and materially. Some of it is recoverable spend.

3. Why this keeps happening

The surprising part: Microsoft 365 already has nearly everything you’d need to govern Copilot safely. Sensitivity labels, Conditional Access, data loss prevention, SharePoint governance, Power Platform controls, audit logging deep enough to satisfy a regulator. It’s enterprise-grade, and it’s sitting in a tenant you’re already paying for.

The problem is that it’s a lot. Microsoft’s governance surface spans a dozen admin centres, licensing tiers, and compliance concepts, each with its own logic, and almost none of it is configured out of the box. Capable internal IT teams get lost in it — not for lack of skill, but because staying fluent in all of it isn’t their full-time job. The traditional answer is a Microsoft partner engagement. Those are priced and scoped for enterprise budgets, which puts real help out of reach for exactly the mid-market systems that don’t have a deep internal security bench either.

So the tenant sits there: fully capable, fully unconfigured, while everyone waits for someone to either turn Copilot on and hope, or freeze the initiative out of caution. Neither is a plan. That’s the choice my CIO client was staring down — and it’s why he called before flipping the switch on Copilot, not after.

4. So what do these risks look like in real life?

In the case of my client, the organisation had grown substantially through affiliation and acquisition, which meant his tenant looked like most multi-site care organisations’ tenants: layers of inherited sites, permissions, and licensing from entities that used to be separate companies, each with its own idea of how sharing should work.

Here’s what surfaced that a marketing leader would care about. I’ve generalized it to protect the client. None of it is hypothetical, and none of it was exotic.

Strong walls, unlabeled contents. Identity and conditional access controls were in reasonable shape. What didn’t exist was classification. No sensitivity labels applied at any scale, no automated tagging of PHI or PII. Meanwhile, clinical staff were pulling records out of the EHR to share ad hoc, and outside vendors were emailing reports containing patient data straight into inboxes. All of it landed in the same file stores everyone else works in — marketing included. Copilot cannot be careful with a file nobody has told it to be careful with.

A long tail of team sites nobody owned. SharePoint was the primary file store. The official departmental and location sites were governed by directory groups and reasonably well kept. Next to them sat a large population of sites end users had spun up themselves, with no formal cleanup cadence and only occasional informal review. That is exactly where campaign folders, agency collaboration, and one-off project sites accumulate — and exactly what nobody inventories until an assistant starts reading it.

External sharing with no expiration date. Anonymous links were blocked, which is the right call. But sharing to named external addresses was permitted, because the organisation worked with too many partners for domain whitelisting to be practical. Reasonable — except nothing expired and nothing was reviewed. Every agency, freelancer, and media partner ever granted access to a folder was, in all likelihood, still holding it.

Licenses nobody was auditing. Copilot access began invite-only for directors, then expanded ad hoc to essentially every corporate user who asked for it. Seats were eyeballed manually at each monthly renewal. There was no automated seat reclamation and no formal licence audit. Money moving in both directions: paying for seats nobody touched, while adoption outran governance.

Automations running on a person instead of the organisation. A handful of Power Platform flows and apps had been built under individual user accounts. At least one broke when the person who built it left. If your team leans on a form handler, a list export, or a routing automation someone resourceful put together, it may be one resignation away from failing quietly.

And the finding nobody expected. Copilot usage was concentrated in Teams and email, and ran roughly half that in Word, Excel, and PowerPoint. The likeliest explanation wasn’t preference — it was that nobody had been trained. Which means the functions producing the most documents in the building, marketing chief among them, were getting the least out of a tool the organisation was already paying for. The governance gap and the value gap turned out to be the same gap.

None of that required a breach to matter. There was no zero-day, no misconfigured firewall, nothing that would have surfaced in a penetration test. It was ordinary operational drift in a tenant that had grown faster than anyone’s ability to keep it tidy. Which is precisely why it deserves your attention — if it were exotic, you could reasonably assume you don’t have it.

What the CIO handed the CEO wasn’t a reason to stall the AI mandate. It was a prioritised map of exactly what needed fixing — and a rollout date he was genuinely confident in.

5. How we made that repeatable, rapid, and affordable

That engagement is where the Copilot Readiness & M365 Governance Audit came from. Two things became clear while building it.

First, almost none of the work was specific to that organisation. Unlabeled sensitive files, end-user sites nobody reviews, external sharing that never expires, licenses nobody reclaims, automations tied to a departed employee — none of that is a senior care problem or a Copilot problem. It’s what accumulates in any tenant that has grown for a few years, especially one that has grown by acquisition, without someone dedicated to keeping it clean. The findings change. The five places you have to look don’t.

Second, the reason work like this normally carries a six-figure price tag is that it’s normally done by hand. So we automated the part that should be automated. The audit starts with a read-only scan against the tenant via Microsoft Graph — nothing is changed, nothing is enabled or disabled — which produces comprehensive coverage across the whole environment in days rather than the weeks a manual review would take. Then targeted stakeholder interviews fill in what a scan can’t know: not that a site is shared with the whole organisation, but that it’s shared that way because a service line needed it in 2019 and nobody revisited it.

The human half is what makes the output usable. Every finding comes back with a screenshot showing exactly what was found and where, written in plain business language, with step-by-step remediation instructions attached. Not “improve data governance.” Which setting, in which admin centre, changed to what, in what order. Automation without interpretation is a spreadsheet nobody reads. Interpretation without automation is a budget nobody approves.

The other thing automation buys you is that it isn’t a one-time snapshot. Governance drifts, quietly, as users are added, sites are shared, and licenses are bought. Most organisations only discover their controls have eroded when something goes wrong. You can re-run the scan and check instead.

Each engagement since has tightened the process further — which is the point of productizing it rather than treating every tenant as a blank page. Three weeks, fixed fee, no six-figure statement of work.

6. What that looks like for your marketing organisation

Three weeks. Fixed fee. Read-only scan, stakeholder interviews, findings in plain business language, every issue backed by a screenshot and a step-by-step fix. Delivered as a live audit site your team can navigate rather than a static PDF that gets skimmed once and archived, plus a prioritised remediation roadmap and a live walkthrough session so nothing gets lost in translation.

For a CMO specifically, the deliverable does something a typical security assessment doesn’t: it produces one document that marketing, legal, compliance, and IT can all read and agree on. That’s the difference between arguing for AI adoption on faith and walking into the executive team with a documented answer. It’s also the one thing that unblocks your roadmap — a defensible reason for legal to say yes.

And the findings are rarely all bad news. The same pass that tells you which patient-story folders are overshared usually also tells you which licenses you’re paying for and not using, and which capabilities you already own that would make your team faster the moment someone turns them on.

7. Frequently asked questions

7.1. Is Microsoft Copilot HIPAA compliant?

Microsoft 365 and Copilot can operate within a HIPAA-compliant configuration when the right licensing, data classification, and access controls are in place — but compliance isn’t automatic. It depends entirely on how your specific tenant is configured, which is exactly what a governance audit is built to verify before you scale Copilot across a health system.

7.2. Can my marketing team use Copilot on patient stories and testimonials?

That depends on whether the files those stories live in are labelled and permissioned correctly today. Copilot reaches whatever the user can already reach, so the question isn’t really about Copilot — it’s whether consent forms, intake notes, and clinical detail sitting in marketing’s SharePoint are classified and access-controlled. A readiness assessment answers that with evidence rather than assumption.

7.3. Isn’t AI governance IT’s job, not marketing’s?

IT owns the controls. Marketing owns the consequences — and, in most health systems, a large share of the adoption. If your team is generating content, analysing patient feedback, or working campaign data with AI assistance, marketing is a stakeholder in how that environment is governed, not a bystander. The organisations that get this right have the CMO in the room early.

7.4. Is Copilot a security risk on its own?

No. The risk comes from oversharing and access sprawl that predates Copilot — files, sites, and permissions nobody has reviewed in years, often inherited through acquisitions. Copilot can reach anything a user can reach, so any existing gap becomes visible the moment someone asks the right question.

7.5. What does a Copilot readiness assessment actually check?

Five workstreams: tenant and Copilot configuration, licensing inventory and gap analysis, data governance and access controls, Power Platform governance, and Copilot telemetry and adoption — with every finding backed by a screenshot and a step-by-step fix, and the ability to re-run the scan later to catch drift.

If this sounds familiar — an AI mandate from leadership, a marketing team already using Copilot unofficially, a tenant nobody has fully audited, and no clean answer when compliance asks how you know patient data is safe — let’s talk. The Copilot Readiness & M365 Governance Audit runs three weeks, fixed fee at $10,000, and gets you a clear, evidence-backed picture of where you stand before you commit to a rollout date.

Reach out directly, I read every message myself: john.schneider@chameleon.co.