AI Vendor Lock-In Is a Purchase-Time Decision
Lock-in feels like a renewal problem. It is created on the day you sign, and it is cheap to prevent then.
Erin Moore
Fractional Chief AI Officer
Lock-in feels like a renewal-time problem and is created at purchase. By the time a vendor raises prices or sunsets a product, your leverage is gone — the tool is integrated, your team depends on it, and the assets that would let you move live inside someone else's system.
The four assets to keep
Your evaluation set. The examples you use to judge whether a system performs. If this lives in a vendor's platform, you cannot compare a competitor fairly, which is most of what switching requires.
Your prompts and business logic. The accumulated encoding of how your company actually works. This is genuinely yours and is routinely surrendered by default.
Your data, in a portable format. Including derived artefacts — embeddings, indexes, fine-tunes. Most contracts cover raw export and say nothing about the rest, which is where the real stickiness sits.
The integration boundary. If your systems call a thin internal interface that talks to the vendor, swapping is a contained change. If they call the vendor directly from forty places, it is a project.
Where lock-in is fine
Not all lock-in is worth avoiding. Deep integration with a system that is genuinely better and priced fairly is a reasonable trade, and abstracting everything "just in case" has a real cost in complexity and time.
The judgement is about consequence: how bad is it if this vendor doubles their price or disappears? For a marginal internal tool, tolerable. For something in your billing path, not. That is the same reversibility axis used in how to prioritize AI use cases.
What to negotiate
Exit format, timing, and treatment of derived artefacts — plus a renewal price cap. All four are cheap to secure before signature and close to impossible afterwards, which is the argument in AI contract negotiation and question twelve of the security questionnaire.
The build-versus-buy connection
The middle path most companies should take — buy the model and platform, build the thin layer encoding your business logic — is largely a lock-in strategy. The commodity underneath stays replaceable; the part that reflects how you work stays yours. That reasoning is set out in build vs buy AI.
Frequently asked questions
How do we avoid AI vendor lock-in? Keep four things in your own control: your evaluation set, your prompts and business logic, your data including derived artefacts, and a thin integration boundary so swapping is contained.
Is lock-in always bad? No. Deep integration with a genuinely better, fairly priced system is a reasonable trade. Judge it by consequence — what happens if the vendor doubles the price or disappears.
What do most contracts miss? Derived artefacts. Raw data export is standard; embeddings, indexes and fine-tunes usually are not addressed, and that is where switching costs accumulate.
When should we negotiate exit terms? Before signature. After integration you are requesting a favour rather than negotiating a term.
Further reading
Tagged with:
Ready to Automate Your Business?
Let's discuss how AI automation can deliver measurable ROI for your organization in 90 days or sooner.