Build ongoing client services into your AI consulting business

Practitioners describe automations that die after launch because nobody at the client owns them, and maintenance that was never priced into the project.

Both problems share a cause: the work after launch was never defined, priced or assigned. This guide is for established AI consultants and small consultancies that want that work to become a defined, paid service rather than a favor. It does not cover starting a consulting business from zero.

Which AI consulting business model fits your capabilities?

Many practices combine two of these models: a fixed project to build something, then a recurring agreement to keep it useful. The real question is which recurring role you can staff.

Many MSPs lead with a fixed-scope readiness assessment; published examples include scoped assessments from about £995, $1,700 excluding GST from an Australian provider, and $5,000 based on 50 users from a Canadian MSP. One published fractional AI officer retainer costs $1,950 a month and excludes hands-on build, training and managed operations. These show how scope is published, not market rates for your work.

Recurring models also create an obligation: someone must respond when something breaks.

What must you establish before taking a client?

Get written answers to these questions before you sign. If an answer is “nobody”, treat it as a scope problem, not a detail for later.

The decision owner and budget

Who approves the work, and which budget pays for setup and for every month after launch?

The problem and its baseline

Which recurring problem is in scope, and how is it measured today: volume, staff time, waiting time, repeat incidents?

The source owner

Who approves the documents and procedures the AI answers from, and who tells you when they change?

The operator

Who runs the service day to day after launch: the client's IT team, its MSP or you?

Access and approvals

Which systems will automations touch, with whose credentials, and who must approve each action?

The handoff

When the AI cannot finish, which person takes over, and during which hours?

The exit

If the engagement ends, what does the client keep, and can its operator run and change it without you?

The exit question matters. Gartner predicted in September 2026 that 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering by 2028. It is a forecast about vendor teams, not consultants, but the risk is the same: a system the client cannot run or change without its builder.

When should work remain a fixed project?

Not every engagement should become a retainer. Keep it a fixed project when:

• The deliverable is a decision or a document: an assessment, a roadmap, a policy or a vendor selection.
• The client already has an operator with the skills and time to run the result, and wants a clean handover.
• The system is stable: its sources rarely change and its automations touch few systems.
• You cannot staff ongoing response. A monthly service you cannot cover is worse than a good handover.

Then sell the handover properly: a runbook, a named owner for each upkeep task, tests the operator can rerun, and an optional paid review later.

What recurring work does the client actually need?

An AI assistant and its automations must change as the client changes. The upkeep is ordinary and easy to underestimate:

• Knowledge upkeep. Procedures and policies change. Someone replaces outdated sources and checks answers follow the new version.
• Access changes. New starters, leavers, role changes and new sites change who may request what, and which records the AI may show them.
• Approvals. Approvers leave and teams reorganize. Each new automation needs a tested approval path.
• Retesting automations. An application update, a changed web page, a new API version or an expired credential can break an automation that worked at launch. Retest after each known change and on a regular schedule.
• Monthly review of unanswered questions. Read what the AI could not answer or passed to a person. Each item is a missing source, a candidate automation, or a request that should stay with people.

What should a monthly review prove?

It should show that the service still works and what changed, not only activity counts. Bring:

• What the AI completed and what went to a person, reported separately.
• Unanswered questions and the action taken on each.
• Failed automation runs, the cause and the fix.
• Sources updated and access changes made.
• Hours used against the allowance, and requests that fell outside the scope.

A closed ticket does not prove the problem was solved. Check a sample with the people who asked, and agree in advance what would justify expanding, changing or stopping the service.

Who operates the service and handles failures?

There are three realistic operators. Choose one per client, in writing.

• The client's IT team. A fit when it has staff to update sources, manage access and respond during business hours. You build, train and hand over, and may keep an advisory retainer.
• The client's MSP. A fit when the client already outsources support. The MSP adds the AI service to its agreement and handles failures through its service desk, but it has to agree to run it. Many hear the demand: in Kaseya's 2026 State of the MSP report (vendor-sponsored, more than 1,000 MSPs), 48% ranked AI and automation as the top client need for 2026, while 13% were generating meaningful revenue from these services.
• You. Possible if you can cover response hours, absences and the client's systems. Small practices often underestimate this.

Write down what counts as a failure, how it is reported, the response target and who acts when the operator is unavailable. If the AI hands a conversation to a person, that person must be available when the client expects.

How do you scope and price upkeep without promising unlimited support?

Price a defined scope, not “support”. Unlimited support turns every new idea into free work. Give each recurring task an allowance and an owner on each side, and say what happens beyond it.

Service-scope worksheet

Fill in one per client and copy it into your proposal, or download the worksheet as a CSV file to edit in a spreadsheet.

New automations, new sources, setup and any service-desk migration usually belong in a separate project fee. Keep software subscriptions separate from your service fee so the client sees each cost.

An illustrative cost and price example

Illustrative only. The hours, cost rate and margin below are placeholders for the arithmetic, not market rates, recommendations or expected results. Use your own figures.

One client with one AI assistant and three automations might need this each month:

Knowledge source updates

3 h

Access and approval changes

1.5 h

Automation retesting

2 h

Review of unanswered questions

2 h

Monthly review meeting and report

1.5 h

Reserve for failure response

2 h

Total

12 h

1. Delivery cost. 12 hours at a placeholder loaded cost of US$90 an hour is US$1,080. Add any tools you pay for on this client's behalf.
2. Monthly price. Monthly price = delivery cost / (1 - target margin). With a 40% target margin: 1,080 / 0.6 = US$1,800.
3. Markup is not margin. Adding 40% to cost (1,080 x 1.4 = US$1,512) gives a margin of about 29%, not 40%. Use the formula.

Track actual hours. If the reserve is used up month after month, the scope or the price is wrong; change the agreement at the review rather than absorbing it.

How are service fees different from software commissions?

Your service fee is what the client pays you for the work in your scope. You set it, and it should cover your delivery cost on its own. A software commission is what a vendor pays you when your client buys its software. It comes from the vendor, depends on the client continuing to pay, and often lasts a limited period.

Keep the two apart:

• Price your service as if there were no commission.
• Disclose any commission to the client before it chooses the software.
• Recommend software on fit. A client who learns of a commission later will question every recommendation.

For example, if a client paid a vendor US$1,000 in a month under a 15% program, the commission would be US$150, paid by the vendor.

NetZen Partner Program: when it fits

NetZen is a service desk, RMM and AI automation platform that a client's IT team or its MSP runs. It fits when the client's need is IT and knowledge support: employees get answers from the client's approved knowledge, and requests are carried out by tested automations (browser tasks, API actions, device scripts and multi-step run books), with approvals the operator sets and a handoff to a person when the AI cannot finish. Employees use a portal or connected channels such as Teams or Slack. Adopting it means moving the client's service desk to NetZen; it is not a layer over another tool, and it is not for general model development.

The NetZen Partner Program follows the division of work above:

• You lead the client's decision.
• NetZen supports the demo, the proposal and the first rollout.
• The client's IT team or its MSP runs NetZen day to day.
• You earn 15% of the NetZen subscription each client pays, for 24 months. There are no fees and no certification.

Your own service fees stay separate from the commission. To plan the first workflow, use the readiness and handover playbook. If the client's MSP will operate the service, share how MSPs package and price managed AI.

Questions AI consultants ask

Sources are linked beside the claims they support. Published prices are examples, not benchmarks. Survey figures are self-reported; the Gartner figure is a forecast.