Economics
Total cost of ownership (AI)
Also called: TCO
Definition
Total cost of ownership is the full multi-year cost of an AI capability — inference, integration, evaluation, governance, review labour, and the re-work triggered every time the model underneath changes.
Pilot budgets capture the build and almost nothing else. The recurring costs — maintaining evals, re-validating after provider updates, staffing the exception queue, meeting audit obligations — are what determine whether the capability is affordable at scale, and they rarely appear in the initial slide deck that got the project funded. TCO is also the honest comparison against a vendor product, which bundles those costs into a licence fee that a build estimate typically excludes entirely.
The recurring line items nobody budgets
Eval maintenance, drift monitoring, prompt and connector reviews, incident response, and re-validation after every model change are all ongoing costs, not one-time setup work, and none are optional once the system touches real decisions with real consequence attached. Teams that budget only for the build routinely discover, a year in, that the true run-rate cost of keeping the system trustworthy is a multiple of what the original project approval assumed, because these categories were simply never included in the first estimate.
Build versus buy resolves here
Buying looks expensive against a build estimate and cheap against a build TCO, because the build estimate almost never includes the ongoing categories a vendor's licence fee is quietly pricing in on your behalf. Run the comparison over three years, including the people who keep it honest — the eval owner, the governance reviewer, the person staffing the exception queue — and the build-versus-buy decision often looks materially different than it did in the first-quarter estimate.
TCO should be revisited, not calculated once
A TCO estimate produced at the start of a project is a forecast, not a fact, and it should be checked against actual spend at regular intervals as the system matures and volumes change. Programs that never revisit the original estimate lose the ability to notice when actual costs are diverging from plan, which is exactly the early warning that would let a leadership team intervene before a program that looked economical at launch quietly becomes a persistent drain on the budget.
In practice
- Budget ongoing evaluation and governance as a permanent operating line
- Model three years, not one, before choosing build or buy
- Attribute review labour to the system that generates the reviews
- Revisit the original TCO estimate against actual spend at regular intervals
