Field Materials AI uses computer vision in addition to OCR to understand visual document layout and extract text within the layout. Then the visual layout together with the extracted text is fed into a Large Language Model trained on tens of thousands of construction vendor documents to do inference.
LLM infers the vendor and extracts all the line items with descriptions and pricing. It also reads notes, understands units of measurement and conversions, and even infers cost codes wherever appropriate.
Field Materials AI is over 90% accurate. 100% accuracy is highly improbable considering that some documents can have information that would be hard to interpret even for humans. But most customers see a 90%+ reduction in manual data entry. The AI gets smarter over time too, learning from your corrections and getting better at reading your specific vendors' formats.
For invoices and delivery slips, the AI compares received and invoiced quantities against the original purchased quantities. It will flag if you receive more than requested, or price is higher than originally quoted, or you’re getting invoiced for items you haven’t received, or the invoice includes items that weren't on the original PO. Field Materials AI will help you identify the backordered items and items that are getting double billed. That's where a lot of the cost savings come from.