AI Builder Model Types Explained

Document-focused
Document processing (custom): extracts specific fields you define from documents with a consistent layout, like invoices or purchase orders. Needs training on sample documents you tag by hand.

Invoice processing and receipt processing (prebuilt): the same idea, but for invoices and receipts specifically, and ready to use with no training at all.
Contract processing (prebuilt, preview): pulls key clauses and data points out of contracts.
ID reader (prebuilt): reads identity documents like passports and driver's licenses.
Category classification: sorts text into categories you define (custom) or a general set (prebuilt), useful for tagging support tickets or feedback.

Entity extraction: pulls specific named things out of text, like a product name or a date, custom if your entities are business-specific, prebuilt for common ones.

Sentiment analysis (prebuilt): flags whether text reads positive, negative, or neutral.
Key phrase extraction (prebuilt): pulls out the phrases that matter most from a block of text.
Language detection and text translation (prebuilt): exactly what they sound like.

Prompts: a GPT model you hand raw text and your own instructions, for anything freeform that doesn't fit a fixed extraction pattern, like summarizing an email in your own words.
Prediction (custom): trained on your historical data to forecast an outcome, like whether a supplier is likely to miss a delivery.
Object detection (custom): finds and counts specific objects in images, trained on photos you provide.

Business card reader (prebuilt): reads contact details off a photographed business card.