The budget of an AI project in an application for support should reflect the actual course of the work, not just the price of the software. It is worth structuring it into analysis and design of the solution, data preparation and cleaning, the development or procurement of the solution itself, integration with existing systems, testing, user training, and support during roll-out. The most common mistake is underestimating data preparation and integration, which in practice consume most of the time, and, conversely, overestimating licences. Every line item must be backed by a quotation or a market survey, and its amount defensible as reasonable. Providers generally do not accept a contingency reserve for uncertainty, so risk is addressed through a more precise definition of scope rather than a mark-up on the price. Dividing the project into stages with their own outputs also makes later reporting easier and reduces the risk that the whole payment is put on hold because of a single disputed item.
See also: Data preparation as an eligible expense, Project budget, AI budget in a company.
