Two opportunities. Two tests.
An MSP using AI to summarise tickets is changing the economics of an existing service. An MSP selling AI workflow management is introducing a new customer proposition. Combining these into one return-on-investment figure hides the central question: which part of the business is expected to improve, and how will the improvement reach the income statement?
Kaseya’s April 2026 survey announcement reports both interest in AI and a gap between that interest and current service revenue. It is useful evidence of questions facing its respondents, rather than proof that every MSP has a ready market. Microsoft’s documentation also makes consumption monitoring and spending controls explicit for supported usage-billed AI experiences. A business case therefore needs continuing operating costs, not simply a licence estimate.
Source context: Kaseya, Microsoft Learn. See source dates and limitations below.
Start with a workflow baseline
Choose a bounded activity with observable inputs and a recognisable acceptable output. Record frequency, handling time, error correction, escalation and customer impact before changing the process. Averages alone can conceal the difficult cases that consume most supervision. Separate routine work from exceptions and retain a route back to human handling.
Measure the whole task. Faster first drafts have limited value if checking and correcting them absorbs the same time elsewhere. Include configuration, knowledge maintenance, permission reviews, usage charges, monitoring and the person responsible when the output is wrong. A quality deterioration can invalidate an attractive time calculation.
Treat saved time honestly
Consider an illustrative monthly workflow with 1,600 eligible tasks. Saving two minutes per task releases about 53 hours before review. If checking and exceptions consume 13 hours, the usable capacity is 40 hours. At an assumed loaded cost of £45 per hour, that capacity has a planning value of £1,800. Subtract £900 of recurring technology costs and £300 of monthly implementation amortisation: the indicative net capacity value is £600.
That is not £600 of cash profit. If payroll remains unchanged and there is no valuable work to absorb the capacity, the cash saving may be zero. The commercial benefit might instead be reduced backlog, better service or deferred recruitment. Name the mechanism and track whether it happens. Test lower task volume and higher review effort before committing.
Design a pilot that can fail usefully
Use a small, agreed test population with examples the team is authorised to process. Define what good output means before the trial starts: required accuracy, permissible omissions, turnaround and the escalation threshold. Include difficult cases, not only the examples on which the system is most convincing. If the pilot handles customer information, use the organisation’s applicable access and data-handling controls from the outset.
Keep a simple review log: task type, automated output, human correction, total time and outcome. Record why a task was excluded. An apparent improvement can disappear if the process quietly removes the hardest work from the sample. A sensible decision might be to automate preparation while retaining human approval, or to stop a use case whose exceptions are too consequential.
Before scaling, appoint an owner for changes in the workflow, knowledge sources and underlying product. A process that works in one pilot can degrade when inputs, staff or customer environments change. Budget for periodic checks and a fallback route. Review the business case after actual usage is available, rather than carrying the pilot assumptions into a permanent forecast.
Price the service you will actually operate
For a customer-facing offer, identify who approves the use case, controls access, monitors consumption and accepts residual operational risk. Define the recurring work: configuration, evaluation, exception handling, user support and periodic review. Charge for a supportable service boundary rather than assuming that access to a model is the whole product.
MSPs should pilot a specific workflow against a quality threshold and a realistic utilisation case. Vendors can improve adoption by exposing costs, failure behaviour and support responsibilities clearly. The strongest pilot conclusion may be to narrow the use case before expanding it.
A framework for the next decision
| Business case | Evidence required | Commercial test |
|---|---|---|
| Internal delivery | End-to-end time, quality and exception cost | Can usable capacity improve service or avoid a real future cost? |
| Customer-facing service | Willingness to pay, support load and consumption | Does the agreed fee cover the work and variability? |
| Scale-up | Stable performance across representative cases | Does value survive a lower-volume or higher-review scenario? |
The calculation is invented solely to demonstrate a method. It is not observed MSP performance, a product price, a savings forecast or a result from GMC research. Product billing and capabilities may change.
Sources & further reading
Explore the MSP programme
Connect practical analysis with peer exchange, relevant events and commercial insight.