An AI subscription comes with a migration cost
Test whether one completed piece of work can move to another tool. A low subscription price leaves workflow rebuilding and continuing support costs to be assessed.
An inexpensive AI subscription can acquire a higher cost when the business work built around it is difficult to move.
The first comparison usually covers the monthly fee with the advertised features. Migration also involves customer records with their approval history. Custom instructions need to move, while integrations may have to be rebuilt. People also need to learn the replacement workflow.
Switching costs give me a way to assess that purchase. A small business can obtain a useful capability while increasing its dependence on the provider. The benefit needs to be evaluated beside the cost of replacing it.
Check what becomes easier to change
The OECD's April 2026 small-business survey describes ready-made AI use among more than 2,000 respondents in 12 countries. It reports uneven integration, with time or skills limiting adoption. Maintenance costs also create barriers. The sample is explicitly non-representative, so its findings describe those respondents. The 2026 survey summary
Ready-made software can reduce the work needed to obtain a capability. Replacing the resulting workflow has its own requirements, which need to be examined before purchase.
Imagine a shop using an assistant to summarize customer requests. The requests need a storage location, as do their summaries. Corrections should remain understandable beside the resulting decisions. If that history only makes sense in the provider's interface, the shop depends on the interface to keep using its own work.
The shop may accept that dependency in exchange for the benefit. An explicit account lets the owner decide whether the trade-off suits the business.
Include the commercial arrangement
The FTC's January 2025 cloud-and-AI partnership study identified potential concerns about computing access. Contractual commitments also featured in its findings, alongside switching costs. The report covers information examined during that period, with later partnerships requiring current evidence. The FTC's report summary
Those partnerships operate upstream of a shop's subscription. They still illustrate a mechanism relevant to the purchase: technical dependencies can make alternatives difficult to adopt, while contracts can add another barrier.
A model's benchmark result leaves those conditions unspecified. Two capable models can produce different operating costs when one requires an awkward contract or a replacement of the team's existing workflow.
Distribution affects adoption too. A feature beside a customer record has access to work already in progress. That convenience may encourage the business to build more steps around it, increasing the work required to leave later.
Date the proposal before interpreting its effect
In October 2026, the CMA proposed a strategic-market designation for Microsoft's business software, then opened consultation. Its announcement discussed bundling with integration, including customers' choice of AI tools. It also described cloud-provider steps on switching and interoperability. The October 2026 update
The announcement records a proposal under consultation. Establishing a completed ruling or a removed barrier would require the outcome of that process.
The proposal brings the software surrounding an AI feature into view. Selecting the feature can affect document access with permissions. Identity management also shapes how it connects to other applications.
Lower data-transfer costs can leave workflow-rebuilding costs unchanged. An export needs to remain understandable to the replacement tool before it provides portability the business can use.
Move one completed example
Consider a hypothetical reorder assistant for a shop. It takes a sales export before producing a proposed order. A person checks the proposal before sending it.
I'd try moving one completed example into another tool. The sales data needs to remain readable, with product identifiers that retain their meaning. The suggestion rule should be documented. Reviewers also need the changed quantities with an explanation of each change.
That test leaves the full business running in its existing system. It examines whether one piece of work can be understood outside that system's assumptions.
The test may show that a plain-file export carries enough information. It could instead expose missing approval history or identifiers that depend on the original integration. Either finding gives the shop evidence about a future move.
Operating costs need the same scrutiny. Manual reconciliation may consume a subscription saving. Staff training adds time, while a fragile connection can require repeated repair. Compare the full task before assigning the lower price a lower total cost.
Keep the process understandable
The file needs an explanation of which inputs the workflow uses. Someone should also know the intended output, with the review condition specified beside it.
That explanation helps a new colleague operate the process even if the provider stays the same. It also gives them a way to recognize output that no longer serves the task.
A small workflow may need a short note beside an input-output example. A consequential process needs more detail about permissions. Error handling requires an owner, with recovery steps that can be followed. The record should be large enough to explain the actual job.
An AI purchase can increase what a small business can do. Assess that capability against the continuing cost of operation, then test how the work could be moved. An exit that remains possible gives the business another option when its needs change.