Data preparation as an eligible expense covers the activities without which an AI project has nothing to build on: cleaning and unifying records, filling in missing data, migrating from legacy systems, labelling samples for model training, and setting up ongoing quality control. This is an expense that companies systematically underestimate in their applications, even though in real projects it consumes the most time. It can be recognised both as an external service and as a wage expense of the company's own employees, if proven by a work record. It is important to describe it as a separate activity with its own output, for example a prepared data file or a documented cleaning procedure. Dissolved into the "implementation" item, it tends to be the first candidate for reduction during a check. It is therefore worth stating in the application an estimate of the volume of records processed, because that is precisely what makes the proposed price defensible.
See also: AI and Company Data Quality, Data audit, Wage costs and timesheets.
