The most useful question in an AI strategy is sometimes the simplest: could we change our minds?
Start with the decisions you need to own
Where will the data be processed? Who can access it? Can a model be replaced without rebuilding the application? These are architecture questions before they become procurement questions. Write down the decisions your organization must retain and use them to evaluate each option.
Treat residency as one part of the picture
Keeping a workload in a chosen region is a useful control. It does not, on its own, answer questions about administrative access, service dependencies, model licensing, or operational continuity. Map the full system: application, model endpoint, retrieval store, logs, backups, and support access.
Make portability a deliverable
Ask for documented interfaces, reproducible deployments, exportable data, and evaluations that can run against more than one model. Test an alternative on a small workflow. A practical exit path is more useful than a promise that switching will be easy.
Keep the knowledge in your team
A system is only as independent as the people who can operate it. Pairing, runbooks, architectural decisions, and hands-on training should be part of the engagement. OneFive’s role is to bring in expertise while helping your organization retain the ability to make its next decision.