The complete definition of Managed Intelligence, the five-pillar framework that makes it deliverable, and what it takes for a managed service provider to build the practice and bill for it every month.
Managed Intelligence is a recurring managed service in which an MSP operates, monitors, and governs a client's AI agents on an ongoing basis. It covers the full agent lifecycle: discovery, deployment, real-time monitoring, compliance governance, and continuous optimization.
It is a subscription service with a monthly operating cadence and a contract behind it. It is not a consulting engagement, and it is not a deployment project that ends at go-live.
Written by Tony Ferrigno, Founder and Managing Partner, AiT Advisory Group. Author of AI for MSPs: Managed Intelligence.
Deploying an AI agent is the easy part. Configuring a support triage agent or an invoice-processing agent is a matter of weeks, and the tooling gets simpler every quarter.
Everything after go-live is the hard part. Agents drift. Accuracy degrades when the underlying data changes. Integrations break silently. Staff route around an agent they do not trust, and nobody tells the provider. A compliance officer asks for six months of decision logs and discovers nobody kept them.
Those problems are continuous, and continuous problems need a continuous service. Nobody had built a repeatable, scalable managed service around that gap. That gap is where Managed Intelligence lives.
The parallel that matters to MSPs is cloud. Cloud did not destroy the managed services industry. It made the good providers irreplaceable, because someone had to own the migration, the governance, the cost management, and the day-two operations the client could not run alone. AI follows the same script on a faster clock, roughly three to five years instead of a decade.
Traditional managed services keep infrastructure available. Managed Intelligence keeps agents accurate, compliant, and productive. Three things change: the unit of management, the definition of the SLA, and what the client is actually buying.
| Dimension | Traditional managed services | Managed Intelligence |
|---|---|---|
| Unit of management | Device, user, server, endpoint | AI agent |
| Core SLA | Uptime and response time | Accuracy, business outcome, policy compliance |
| Failure mode | Something stops working and someone opens a ticket | Something keeps working and produces the wrong answer quietly |
| Governance | Background hygiene, rarely itemized | A named, billable pillar with audit trails |
| Growth path | More seats, more sites, more clients | More agents inside the same client, on a 90-day cycle |
| Client conversation | IT operations and budget owners | Operations, finance, compliance, and the CEO |
| Reporting artifact | Ticket volume and SLA attainment | Monthly AI Performance Report tied to business impact |
The last row carries more weight than it looks. An MSP that cannot produce a monthly report showing what the agents did, what they cost, and what they returned is not delivering Managed Intelligence. It is running unmanaged automation and invoicing for it.
The label "Managed Intelligence Provider" has spread quickly through the channel, and it is useful positioning. It is also being applied loosely. Most providers claiming it today are selling AI deployment projects with a new name on the invoice.
The test is simple and it has one question. What happens in month seven?
If the answer is "we check in and see if they want anything else," that is project work. If the answer is a defined monthly cadence covering monitoring, governance review, a performance report, and an optimization recommendation, that is Managed Intelligence. The name follows the operating model. It does not create one.
Before you rebrand as a Managed Intelligence Provider, write the month-seven deliverable. If you cannot describe what the client receives in month seven of the contract, you do not yet have a service to sell. You have a project you are hoping renews.
Five pillars, in this order: Discover, Deploy, Monitor, Govern, Optimize. Discover and Deploy are what most providers already recognize as work. Monitor, Govern, and Optimize are the three that turn it into a managed service, and they are the three most providers skip.
Every engagement runs all five pillars every month, in perpetuity. That is the structural point.
Assess the agent opportunity, identify the processes worth automating, and build the business case and deployment blueprint.
Configure agents, establish integrations, test against real conditions, and go live in stages.
Maintain visibility across every deployed agent covering health, uptime, performance, and business impact.
Enforce compliance controls, audit trails, lifecycle management, and policy guardrails across the whole agent estate.
Improve agent performance continuously and introduce new use cases every 90 days.
Govern sits fourth for a reason. Providers who bolt governance on after the agents are live spend the next two quarters rebuilding logging, retrofitting approval workflows, and explaining to a compliance officer why the first ninety days are undocumented. Govern is cheap to build at deployment and expensive to add afterward.
Governance has also outgrown a single pillar. It is the subject of the fourth book in the series, AI for MSPs: Governed Intelligence, now in final production.
A managed service is a promise about rhythm. This is what separates Managed Intelligence from automation someone installed once.
Notice what is missing: an annual review. Annual cadence is how providers lose AI accounts, because a year is long enough for an agent to drift, a champion to leave, and a competitor to run a discovery meeting the incumbent never heard about.
Managed Intelligence has two revenue components. Onboarding is a fixed-fee project. Ongoing management is a monthly fee tied to the number of agents under management.
Onboarding runs from roughly $5,000 for a starter engagement of one or two agents and two or three integrations, up to $150,000 or more for an enterprise estate of ten or more agents with an orchestration tier. Timelines scale with it, from three or four weeks at the low end to ten to twenty weeks at the top.
Monthly fees run from roughly $500 per month for one to three agents to $20,000 or more per month for sixteen or more agents. Recurring pricing assumes a 24-month minimum commitment. Month-to-month terms carry a 25 percent premium, and that premium is not a negotiating position. It prices the operational risk of a client who can walk before the agents are stable.
This is where the model earns its reputation. An additional agent riding on integration infrastructure you already built prices at roughly 60 to 70 percent of the standard onboarding rate for equivalent complexity. An additional agent that needs new integration work prices at roughly 80 to 90 percent. A tier upgrade with no new agents is the rate difference only, with no setup fee.
Run that against a 90-day expansion cycle and the account grows four times a year without a single new logo. Governance can also be sold standalone as a retainer in the $800 to $2,500 per month range, which is the cheapest way into a client who already deployed agents somewhere else and cannot prove what they are doing.
Figures as cited in AI for MSPs: The No-Fluff Playbook, Chapter 1. The gap between the third number and the fourth is the entire commercial argument.
Three failures account for most of what goes wrong, and all three are contract and cadence problems before they are technology problems.
The agent works. Staff route around it. This shows up as high uptime paired with flat business impact, which is exactly why monitoring needs a third layer. Uptime alone will tell you everything is fine while the client quietly concludes they wasted their money.
Output quality falls below the threshold the client was sold. The recoverable version is where you agreed a written acceptance threshold at deployment and can point to the trend. The unrecoverable version is where “accurate” was never defined and becomes a matter of opinion during a billing dispute.
The client expects agents or integrations the contract does not cover. Almost always traceable to a sales conversation where agent count was discussed and integration count was not. Price both. Name both in the agreement.