Running it is probably costly, but there are papers on "influence analysis": Training data influence analysis and estimation: a survey (https://link.springer.com/content/pdf/10.1007/s10994-023-064...)

It would be easier to negotiate a fixed cost on using a particular datum per training of a model.