AI Agents for Managed Service Providers

How the money actually works here

An MSP's economics are a contradiction: revenue is recurring and roughly fixed per client, while cost is ticket volume, which is not. Adding a client adds predictable revenue and unpredictable load, so profitability per client is decided after the contract is signed by how many tickets they generate. Headcount is the only lever most MSPs have, and it is a step function — you are either comfortably staffed or hiring in response to a queue you are already failing. Anything that reduces tickets rather than handling them faster changes the shape of the business.

For scale: the median firm of this size does $280,611 per employee a year, on 2022 Census figures. The revenue per employee benchmark has the same figure split by headcount band. These are averages rather than medians, so the largest firm in the band pulls them upward.

The rhythm this has to fit

There is no season, but there is a pattern: ticket volume spikes on Monday mornings, after any change window, and whenever a client rolls out something they did not tell you about. The load is therefore lumpy in a way that makes staffing genuinely hard — average volume is comfortable and peak volume is not, and you staff for something in between and absorb the difference in response times. Continuous first-line triage flattens exactly the peaks that headcount cannot economically cover.

Where the repetitive load sits

Three roles absorb most of it. Median wages are from the US Bureau of Labor Statistics, grossed up by the 1.43× private-industry loaded-cost ratio:

That is roughly $201,358 of annual compensation covering work that is, in large part, the same sequence repeated. 41% of that — the share Slack’s Workforce Lab found desk workers spend on work that is “low value, repetitive or lacks meaningful contribution to their core job functions” (n=10,281, six countries) — is about $82,557 a year.

Read across those three roles, the work itself is order management, accounts receivable, expense management, customer onboarding, customer support and quote to cash. Each of those is a loop with its own failure points, and the page for it says where the loop breaks rather than where the job title sits.

Against BreathingRoom at $30,000 a year, recovering just 36% of that covers the engagement.

What we would build first

• First-line ticket triage and routing
• Recurring maintenance reporting
• Contract and licence renewal tracking

Which goes first is decided in the audit rather than in advance. Anyone who tells you the order before looking at your volumes is guessing.

How this goes wrong in managed service providers

First-line triage is the right build and the failure is misclassified severity. An agent that routes a genuine outage into the standard queue because the ticket was written calmly has cost a client an SLA and you a credit. Users under-describe emergencies routinely — the person whose system is down often writes a short, polite message. The guard is that severity is inferred conservatively and never downgraded automatically: the agent may raise a priority on its own but only a person may lower one, and anything mentioning an outage, security or data loss goes straight to a human.

When you should not do this yet

Below roughly two hundred tickets a month this is hard to justify — the triage overhead approaches the triage saving. It is also the wrong move if your ticket volume is concentrated in one or two clients with a genuinely non-standard estate, since the value here comes from patterns repeating across clients. And if your documentation is thin, fix that first: an agent answering first-line tickets from a knowledge base that is out of date will confidently give clients your old answers.