If you have 2000 different variations of invoice, how would you process this in AI builder with collections? Is it possible to create 2000 collections and train each variation of invoice? What's the impact on performance during processing of multiple variations within Power Platform or does it make sense to use Form Recognizer composed model, and publish it to containers.
Any insights?
At the moment, we support 200 different collections with the structured model.
The recomendation is to create an initial model and test each new vendor/layout against the model to check wherever a new collection is needed or not.
At some point, the model should be robust enough to manage the vast majority of your vendors/layouts without having to add more collections in the training set.
Hello @CedrickB,
I have a related question: How does a model handle invoices that are of the same format but that may have a different number of line items? Would these invoices go into the same collection or separate collections? For example, Invoice A has line items 'Amount', 'Consumption', 'City Tax', and 'Energy Solution Credit' but Invoice B has line items 'Amount', 'Consumption', and 'State Tax'.
Thank you.