Article
Budgeting and governing an AI workflow after the prototype
I start with the full operating cost: review, support, failed runs and the work needed to keep source data usable. A cheap model call can still sit inside an expensive process, and that process needs an owner after the prototype is approved.
A prototype budget usually records the visible ingredients: a model, some development time and perhaps a data store. Routine operation adds work that the demonstration did not have to carry. Inputs arrive in poor condition, permissions change, reviewers handle uncertain results, failed runs need recovery, and somebody has to answer when a user disputes an action.
The cost model should follow the full workflow. It needs volumes, service prices and infrastructure, but also staff effort, exception rates, maintenance, controls and the cost of correcting a bad outcome. Governance then gives each of those parts an owner, a review frequency and a decision path.
This is a planning method, not a benchmark. Prices, labour rates and risk tolerances vary. Use current supplier terms and observed workflow data when filling it in, then keep assumptions separate from measured results.