Waxell

How MSPs run AI governance for their clients

MSPs run AI governance as a repeatable service: assess each client, apply one policy baseline, report regularly and manage all tenants from one place.

Logan Kelly10 min read

MSPs run AI governance for clients as a repeatable managed service: an assessment that finds the AI each client already uses, one policy baseline applied to every client, a regular report the client can read, and tooling that manages all client tenants from one place. What does not scale is doing each of those steps by hand, tenant by tenant, so the practice is built around doing each step once and applying it everywhere.

Your clients need this whether or not they have asked for it yet. In Microsoft and LinkedIn's 2024 Work Trend Index, 78% of people who use AI at work said they bring their own AI tools. And the MSPs trying to help are finding governance is the hard part: in a survey of 333 MSPs that Omdia ran for AvePoint, 51% named data governance and compliance as the main obstacle to their customers adopting AI, ahead of security, skills and budget.

This guide is for MSP owners and service leads building an AI governance practice. It covers what the service includes, how to run each part, what to report, and how to keep it from eating your team's time.

What does an AI governance service for clients include?

A complete service has five parts. You can sell them together or start with the first and grow from there.

  1. An AI assessment. Find which AI tools each client's staff use, on which accounts, and with what data.
  2. A policy baseline. A standard set of rules for approved tools, sensitive data and AI agents, applied to every client with a few per-client exceptions.
  3. Ongoing monitoring. Alerts when someone uses a blocked tool, shares sensitive data with AI or connects a new AI app to company accounts.
  4. Regular reporting. A monthly or quarterly report that shows the client what changed, what was stopped and what to decide next.
  5. Evidence for audits. Records the client can hand to an auditor or a cyber insurer when asked how they control AI.

The rest of this guide takes each part in turn, using the questions MSP owners ask most.

How do you run an AI assessment for a client?

An AI assessment answers three questions for the client: what AI is in use, what data it touches, and what should change first. A practical version takes four steps.

  1. Inventory. Combine what you can see from the network, the devices you manage, the AI apps users have connected to company email and files, and the accounts people sign in with. No single view shows everything, so use at least two.
  2. Interviews. Ask each department head which AI tools their team relies on and what they use them for. People will name tools you did not find, and tell you which ones they cannot do without.
  3. Data check. Note where sensitive data meets AI: client records, financial data, health data, source code. This sets which rules matter most.
  4. Findings and roadmap. Give the client a short list: what to approve, what to stop, and what to put in place in the next 90 days.

To give the findings structure, map them to a framework the client recognizes. The NIST AI Risk Management Framework is a common choice. It organizes AI risk work into four functions: govern, map, measure and manage. It is voluntary, and NIST notes it is being updated, so treat it as a structure for the conversation rather than a certification.

What should a starter AI policy baseline cover?

Keep the first baseline short enough that a client reads it and staff can follow it. Three or four rules cover what most clients need:

  • Approved tools and accounts. Which AI tools staff may use for work, and only with company accounts, not personal ones.
  • Sensitive data. What may never be pasted into, uploaded to or connected to an AI tool, such as client records, payment data or health data.
  • AI agents and connectors. Any AI that can act, send email or reach company files needs approval before it is connected.
  • What happens on a breach. Who is told, and how a tool is switched off quickly.

For each rule, choose how it is enforced: allow, warn the user, or block. Start with warnings for most rules and block only the clearest risks, so staff learn the rules before anything is blocked. The guide to AI policy enforcement goes deeper on writing rules that can be enforced, and keeping sensitive data out of AI covers the data rules.

Can you build one baseline and copy it across client tenants?

Yes, and you should. Write the baseline once, apply it to every client, and record each client's exceptions separately so the shared rules stay identical everywhere. When you change the baseline, the change reaches every client at once, and you can show each client exactly where they differ from it.

MSPs already work this way elsewhere. Microsoft 365 Lighthouse, for example, gives partners in Microsoft's Cloud Solution Provider program a default security baseline for small and medium-sized business tenants and multi-tenant views across all customer environments, from a single portal. Bring the same pattern to AI: one baseline, applied everywhere, with drift shown per client.

The alternative is what one MSP executive described in the research behind the AvePoint survey, as reported by Channel Insider: "We could build automations for every single one of our clients, but we'd have to do it 700 times."

What can you show a client each month or quarter?

The report is what the client sees of your work, so it should answer the questions an owner or board would ask. A useful structure:

SectionWhat it shows
AI in useTools found this period, new since last time, and which are approved
AccountsHow much use is on company accounts versus personal ones
Policy eventsWhat was warned or blocked, grouped by rule, with notable examples
Sensitive dataAttempts to share sensitive data with AI, and the outcome
Decisions neededTools to approve or retire, and rule changes you recommend

Keep it to a page or two, lead with the decisions, and show the trend over time. An owner should be able to read it in five minutes and know what to do next. In a quarterly business review, the "decisions needed" section is the agenda.

How do you manage AI governance across many clients from one place?

You need four things from your tooling, whatever you use:

  • One view across clients, so you can see every client's alerts and status without signing in to each tenant separately.
  • Strict separation, so each client sees only their own data, devices and reports.
  • Shared policies, so a rule written once applies to every client, with per-client exceptions.
  • Per-client reporting, so each report is produced from the same template without manual assembly.

Without these, every new client adds a full set of manual work. With them, a new client is mostly a deployment and an exceptions list.

Will you need someone watching this every day?

Not if the service is built around alerts and exceptions rather than watching dashboards. A workable rhythm looks like this:

  • Daily: triage the alerts that need action, such as a blocked attempt to share client data.
  • Weekly: handle approval requests for new tools and agents.
  • Monthly: produce and send each client's report.
  • Quarterly: review the baseline, retire rules that never fire, and walk each client through their decisions.

How much time that takes depends on the number of clients, how noisy the alerts are and how many exceptions you allow. Tune the alerts in the first month, because noisy alerts are what turn this into a full-time job.

How do MSPs package an AI governance service?

Package it the way you package your other managed services: a one-off assessment as the entry point, then a recurring managed service for monitoring, reporting and policy upkeep, with project work for anything bigger, such as rolling out an approved AI tool. Fitting it into the way clients already buy from you lets it slot into existing contracts.

Price from your own costs: the tooling, the hours your rhythm needs per client, and the value of the reports and evidence to the client. Be wary of published price points for AI services; this market is new and moving.

Which rules and frameworks should the service map to?

Map each client to what their industry already asks of them. The NIST AI Risk Management Framework works as a general structure. Clients in regulated industries, or selling into the EU, will have specific rules to meet. The guide to AI compliance and audit trails covers the records auditors expect.

What an AI governance practice is not

It is not a one-time project. AI tools change every month, so the value is in the ongoing monitoring and reporting, not the first assessment. It is also not the same as reselling an AI tool. Governance is the service that makes a client comfortable using the tools you sell them. If you are new to the subject, where to start with AI governance is a plain-English introduction you can share with clients.

How Waxell handles this

With reseller access in the Waxell partner program, you flip between client tenants from one console, and each client sees only their own devices, decisions and evidence. With Waxell Endpoints, you tune the Guard policy rules on one client, export them, read a dry-run plan of what would change on the next client, and then apply them.

Waxell Endpoints finds the AI already running on employee laptops. It deploys to client devices through the device management tools you already run: MDM, or RMM on Windows. The partner portal's roll-up dashboard shows every client workspace in one view, with usage and 30-day activity per client.

See how Waxell works for MSPs.

FAQ

What is an AI governance service for MSPs?

It is a managed service that helps clients use AI safely: finding the AI their staff already use, setting rules for which tools and data are allowed, monitoring for problems, and reporting regularly. Most MSPs sell it as an assessment followed by a recurring monthly service.

How do you run an AI assessment for a client?

Combine an inventory of AI tools from the network, managed devices and connected apps with short interviews of department heads. Note where sensitive data meets AI, then give the client a short list of what to approve, what to stop and what to put in place in the next 90 days.

What should be in an AI policy for a small business?

Start with three or four rules: which AI tools staff may use and only with company accounts, what data must never go into AI, which AI agents and connectors need approval, and what happens when something goes wrong. Keep it short enough that staff actually read it.

Can one AI policy baseline work for every client?

Yes, if you keep each client's exceptions separate from the shared rules. Apply the baseline everywhere, record per-client differences, and change the baseline in one place so every client gets the update. Microsoft 365 Lighthouse takes a similar approach to security, with a default baseline for small and medium-sized business tenants and multi-tenant views across all customer environments.

What should an AI governance report for clients include?

Show the AI tools in use and what changed, company versus personal account use, what was warned or blocked, attempts to share sensitive data, and the decisions you need from the client. Lead with the decisions and keep it to a page or two.

Sources

Want to see what running AI governance across your clients looks like? Book a 30-minute demo.