AI Managed Services: What Changes When the MSP's Product Is Resolution

Managed service providers have spent twenty years getting good at a specific thing: absorbing a client's IT operations and running them more cheaply than the client could, mostly by pooling scarce expertise across many clients.

The pooling logic is sound and it is not going away. What is changing is the unit being pooled.

When resolution requires a person, the MSP's product is access to people at a better ratio than you could staff yourself. When a large share of resolution can be performed by software, the product becomes the configured, governed system that resolves, and almost every commercial assumption built around the first model misprices the second.

Why the three standard pricing models break

Per-device or per-endpoint. Built on the assumption that support load scales with estate size. It roughly does, for a human-staffed desk. It does not once resolution is automated, because the automated share does not consume the resource the price was proxying for. The buyer pays for growth that costs the provider almost nothing.

Per-ticket. The most direct conflict of interest in the industry, and it becomes acute here. A provider paid per ticket has no reason to eliminate ticket categories. If the provider deploys agentic resolution under a per-ticket contract, it is deliberately shrinking its own revenue.

Per-user, fixed monthly. The closest to workable, with one serious exposure: if the underlying AI cost is consumption-based and the client contract is fixed, the provider has written an unhedged option. Usage grows, cost grows, price does not. ServiceNow's Q2 2026 disclosure, subscription gross margin down 250 basis points to 80.5%, attributed to accelerating AI consumption. Is the same mechanic visible at platform scale.

The model that survives is some version of pricing the outcome: a committed resolution scope at a fixed price, with clearly defined boundaries for what falls outside it. That requires the provider to have a real view of its own cost per resolution, which most do not yet have.

The asset nobody has on the balance sheet

Here is the part that should make MSPs more optimistic than the framing usually allows.

An agentic service desk is only as good as the knowledge it is grounded in. Building that knowledge is the hard, slow, unglamorous part of every deployment, and it is precisely what an MSP has been accumulating for years across dozens of clients.

The runbooks. The known-error database. The accumulated understanding that this vertical always has this problem in the first week of the quarter. A provider that has supported forty mid-market manufacturers has seen the failure modes of mid-market manufacturing more times than any single client ever will.

That knowledge has historically been trapped: in senior engineers' heads, in inconsistent documentation, in tickets nobody mines. The thing that makes it valuable now is that it can finally be operationalized: encoded once, applied across every client, improved from every resolution.

The MSPs that do well here will be the ones that treat accumulated cross-client knowledge as the product. The ones that treat AI as a cost-reduction exercise on their existing labour model will compete on price against providers who have changed the model.

There is a real constraint attached: client data is client data. Knowledge derived from one client's environment cannot leak into another's answers. The distinction between pattern and particulars has to be architectural, not a policy statement.

If you are buying managed services

Ask for autonomous resolution rate by client tenure. Aggregate deflection across the book tells you nothing. It is dominated by whichever clients have been on the platform longest. Ask what it looks like at three months, at twelve, for clients your size.

Ask who owns the knowledge base at the end of the contract. This is the new lock-in, and it is more consequential than the old one. Tooling is replaceable. Three years of accumulated, structured, environment-specific knowledge is not. Get portability in writing.

Ask what the agent is permitted to do in your tenant. A provider running agentic resolution has write access to your identity provider, your endpoints, and your ticketing system. That is a larger grant than the old model and it deserves a larger review. See governing enterprise AI agents.

Ask how AI cost is passed through. Fixed today, metered later, or metered now? A provider absorbing consumption cost at a fixed price is either very confident about its unit economics or has not looked closely.

Ask what happens to the price when volume falls. If the engagement succeeds, ticket volume should drop substantially. A contract that does not contemplate that is one where your success funds the provider's margin rather than your budget.

If you are selling them

The uncomfortable question is what you are charging for in three years.

If the answer is still "a team of engineers at a better ratio than you could hire," the ratio is the whole business and someone will offer a better one. If the answer is "a configured, governed resolution system informed by everything we have learned across four hundred clients, with accountability for the outcome". That is defensible, and it is not something a client can assemble alone.

The transition is genuinely hard, because it means deliberately reducing the billable hours that currently fund the business. Providers that wait for clients to force the issue will make the transition under margin pressure and on someone else's timeline.

The ones moving now are doing it while they can still choose the pace.