Managed AI for MSPs is a monthly service where the MSP, with a delivery partner if needed, picks which client workflows get AI, redesigns them with a person approving the output, and reports measured results every month. Vendor agents are the tools inside that service.
By Sagar Pandya, Co-Founder and CEO, Spej. Exited MSP founder.
The agents are already in your stack
ConnectWise has made its RMM, Voice, Teams and Email AI agents generally available. Its RMM agent prepares remediation steps inside the PSA ticket, where a technician reviews and approves them before changes are made (ChannelE2E).
Lemhi has also announced general availability of its managed AI services platform for MSPs, with a certified ConnectWise PSA integration (Channel Insider). In Lemhi's own platform data from its partners' assessments, 64% of employees at companies already using AI tools personally use at least one tool their company hasn't adopted (Channel Insider). At companies where Microsoft Copilot is the only AI tool, 63% of employees use ChatGPT on their own (Channel Insider). The data comes from Lemhi partners' AI readiness assessments of 183 organizations (Channel Insider).
My read is that agents and governance tooling are becoming commodity features in the MSP stack. I'd build the service around decisions a client needs help making: which work deserves attention, who can approve what, and whether the change paid back. An agent license leaves those questions with the client.
What to sell on top of AI agents
Workflow selection
Sell the work of choosing where AI belongs. Have the vCIO and the client's process owner review a real workflow together. Record where work waits, what gets re-entered and how mistakes are corrected. Check whether the necessary data is accessible and whether someone can own the result.
The deliverable should be a written choice with a reason: this workflow is worth changing, this is how we'll measure it, and these prerequisites need attention first. For a mid-market client, a narrow, frequent task with a willing owner is a sensible place to start. Put attractive ideas with inaccessible data or no accountable owner back on the backlog.
Workflow redesign with a human in charge
Draw the workflow from input to final action. Decide what AI prepares, what the reviewer sees and what happens when the reviewer rejects the output. Name the person who approves it and give them enough context to make the decision. Include a manual fallback for failures.
The ConnectWise RMM approve-in-ticket pattern is a useful model for that handoff. Apply the principle to client work: keep the proposed action and its supporting information where the responsible person already works. A review step that requires hunting through another application adds friction you need to account for.
There's evidence for bounded AI in manufacturing. In a Rootstock-commissioned survey conducted by Researchscape, 73% of manufacturing professionals preferred human-directed AI or AI operating within defined limits with escalation; only 13% preferred fully autonomous AI (Rootstock). Use that as a reason to ask about approval boundaries, while confirming what each client's workflow requires.
Measured results, reported monthly
Before the build, agree on a baseline and where the evidence will come from. You might track handling time and rework for the selected workflow. Include the time a person spends reviewing AI output, and keep software and service costs visible when discussing the return.
In the monthly report, show the starting measure beside the current measure and explain what changed. Record usage so the client can see whether people are using the workflow. If the evidence doesn't support a benefit yet, say so and agree what to change or stop. A usage chart alone won't tell a service leader whether the work improved.
Package it as a monthly managed AI line
Write the monthly scope around the workflows you support. Specify who handles incidents, how exceptions reach the process owner and when access permissions get reviewed. Include time to help staff use the workflow, review results with the client and prioritize changes. Separate larger builds from routine maintenance so both sides know what the retainer covers.
For manufacturing clients, there may be a budget conversation to have: 77% of manufacturing professionals in Rootstock's survey plan to increase budgets or resources for AI-powered tools over the next 12 to 18 months (Rootstock). A stated plan doesn't give you an approved budget. Ask the process owner what they can fund and what result would justify continuing.
For the starter offers to put on your service menu, see how MSPs can add AI services. For the delivery decision, see managed AI platform, your own engineers or a partner.
Spej's published pricing is an AI Roadmap starting at $10,000, Design & Build starting at $20,000, and AI Office starting at $10,000 a month. Final scope and pricing are agreed for the engagement. Set your client offer around the responsibilities you're taking on and the delivery terms you've agreed.
What AI governance evidence should your managed AI line keep?
An acceptable-use policy is a starting point for AI governance for MSPs. The monthly service should also keep evidence of the approved tools, the data they may access, the actions they may take, and who reviews the results. Insurers, auditors and client security questionnaires are starting to ask for that proof. The practitioner recommendations here are from Kevin Williams, Smarter MSP, November 10, 2026. Use that evidence to prepare for those questions, then confirm it against the request you actually receive.
Cyber insurance AI questions vary by insurer. Milad Morgan, a broker at Atlas Risk Partners, told Smarter MSP that he points to Aon's published underwriting questions on AI use, human oversight, and data and intellectual property controls. He says requirements still vary by insurer. A governance package can support underwriting but cannot override an exclusion or establish coverage by itself. Morgan flags the MSP and client split as the detail most likely to create problems when a claim is filed: the provider says it configured the tool, while the customer assumes the provider is reviewing every output. Put that split in writing, show the permissions, approval records and exception logs that support it, and confirm the requested evidence with the insurer or broker. This is not coverage advice.
John Gormally, former digital marketing manager at Coeus Consulting, a Phoenix-based MSP that sells compliance services, described a five-page package. It starts with an AI tool register that lists each tool, its owner, the data it is allowed to access, and whether it operates under business-grade terms, plus an acceptable-use policy every employee signs each year and a rule that client data, PHI and CUI stay out of consumer AI tools. Before any AI system that can read company files is turned on, someone reviews access, and unapproved tools are blocked or logged. The package also includes a callback rule for any payment or banking change, and a named owner who reviews it each quarter. The package includes a loose alignment with the NIST AI Risk Management Framework so an auditor can recognize the framework. Spej makes no certification claim for that alignment. The package is preparation for questions that still differ by insurer, auditor and client.
Gormally uses Copilot oversharing to explain the access review. A rollout can expose existing oversharing quickly when files are broadly shared, old accounts remain active, or sensitivity labels are missing. Review those permissions before you enable an AI tool that reads company files.
On the monthly managed AI line, include register updates, approval and exception records, access changes and the work of keeping that evidence current. Keep the quarterly owner review and the annual policy signature in the scope, and write down who approves a change and who holds the evidence. Dara Gibson, CEO and CISO of Cybersecurity Readiness Advisors, warns that clients may assume the MSP covers it, and the MSP may assume the client or insurance covers it, and then nobody holds the evidence. Williams reports her implementation timeline as 30 to 60 days for the documentation, tool inventory, acceptable-use policy, vendor review, fraud controls, incident response addendum and training. Treat that range as a practitioner estimate, and keep it separate from any Spej delivery commitment.
MSP GLOBAL's Summer 2026 State of the Industry Report, from the July 2026 survey wave and published in August 2026, is context for where MSPs say they deploy AI and automation. It doesn't measure demand for governance packages. Among the n=126 respondents asked where they deploy AI and automation, 70.6% use it for internal operations, up from 54.9% in April 2026 (n=164), and 42.1% sell client-facing AI-enabled services. The category shares on that table add up to more than 100%, and the report flags these AI deployment figures as small-n results (MSP GLOBAL Summer 2026 report).
Spej can run the workflow and the evidence trail behind the MSP through white-label delivery. Direct partnerships and AI Office embeds are also options, once you agree the scope and who is responsible. Spej offers white-label AI delivery and AI Office embeds, so we have a commercial interest in this work. The evidence work sits on the service menu in how MSPs can add AI services, and partnerships covers delivery behind the MSP and the responsibility split. Bring one client workflow and its evidence gaps to Spej to discuss delivery scope.
How Spej fits behind you
Through white-label delivery, you lead the offer under your brand and own and manage the client relationship. Spej supplies security, architecture, governance and delivery behind the scenes on agreed terms. You bring the client context and help decide what work deserves doing.
Spej also works directly with MSPs and holding companies as an embedded AI Office, working alongside the team. AI Office is a monthly retainer that maintains what's built, puts a new initiative into production each month and reports on it. An MSP owner or a holding company's operating partner can use that support to develop the service with their team. Agree who owns support and how knowledge passes to your staff before delivery begins.
Disclosure: Spej offers white-label delivery and an embedded AI Office, and we have a commercial interest in this work.
Where this falls short
Buying a platform won't stop shadow AI on its own. Staff still need an approved way to do the work they're using outside tools for, and someone needs to understand why they're going outside the approved setup.
A redesigned workflow nobody uses won't pay back. Make adoption part of the monthly review and ask the people doing the work where it slows them down. Results also need a baseline before the build; without one, you'll have a harder time separating an improvement from a good impression.
Some clients aren't ready. Missing permissions, unreliable data or a process with no owner can make delivery premature. Put the prerequisites in writing and revisit the decision when the client can meet them. Keep that judgment in the service even when it delays a sale.
FAQ
What is managed AI for MSPs?
Managed AI for MSPs is a monthly service that selects client workflows for AI, redesigns them with a person approving the output, and reports measured results. The MSP owns the service and can bring in a delivery partner. Vendor agents provide tools within the workflow, with responsibilities agreed for each client.
How can MSPs offer managed AI services?
Define the workflows you'll support and agree who owns each one at the client. Set an approval point, record the baseline before building, and include support, adoption and results reviews in the monthly scope. An MSP can deliver with its own team or agree delivery responsibilities with a partner.
If my PSA/RMM already has AI agents, what is left to sell?
You can sell workflow selection, workflow redesign and monthly evidence of results. A client still needs someone to decide where AI belongs, assign a reviewer and check whether the change helped. Package that work with support and adoption so the client knows what the recurring service covers.
Do clients want fully autonomous AI?
Most manufacturing professionals in Rootstock's 2026 survey did not: 73% preferred human-directed AI or AI operating within defined limits with escalation, while 13% preferred fully autonomous AI. Ask each client what AI may do, who approves its output and when it must escalate. Agree a manual fallback before delivery.
How does Spej work with MSPs?
Spej can deliver behind an MSP under its brand, with the MSP owning and managing the client relationship. Spej also works directly with MSPs and holding companies as an embedded AI Office alongside their teams. Agree scope, support ownership and commercial terms for the work before starting delivery.
What AI governance evidence should an MSP help clients keep?
Keep an AI tool register with owners, allowed data and terms, plus records of access reviews, human approvals and exceptions. Add a signed acceptable-use policy and a named review owner. Practitioners quoted by Smarter MSP recommend this evidence package. Insurer, auditor and client questionnaire requests can differ, so confirm the specific request. Smarter MSP, November 10, 2026.
Does an AI governance package guarantee cyber insurance coverage?
No. Broker Milad Morgan, quoted by Smarter MSP, says a governance package can support underwriting but cannot override an exclusion or establish coverage by itself. Requirements vary by insurer. Put the MSP and client responsibility split in writing and confirm the requested evidence with the insurer or broker. This is not coverage advice. Smarter MSP, November 10, 2026.
If you're defining a managed AI service for your clients or across MSPs in a holding company, talk with Spej about partnership options. Bring the workflow you're considering and the responsibilities you want your team to keep so we can discuss a delivery scope.