Document Analysis and Structured Data Extraction

The problem

The information is in the document, waiting for someone to retype it

Supplier quotes, order forms and contracts arrive every day in different layouts. Someone has to read each one and re-enter the dates, amounts, parties, line items and special terms into a spreadsheet or system. It is slow, and when attention slips a figure gets mistyped or an important clause is missed.

• The same information sits in a different place in every document
• Manual data entry is slow and error-prone
• Important terms get missed in long contracts
• Comparing documents side by side is hard

The solution

A document comes in, the fields fill themselves, anything doubtful is flagged

Together we define which fields to extract for each document type — for a quote, say, supplier, validity date, line items and total. The assistant reads each incoming document, fills in those fields and shows where in the document each value came from. Fields it is unsure about, can't read or that break a rule are clearly flagged, and nothing flagged is passed on without approval. Checks such as whether line items add up to the total are done with fixed rules, not with AI.

Possible benefits

• Far less retyping by hand
• The source of every field can be checked in one click
• Uncertain fields are flagged, never passed through silently
• Documents in different layouts can be compared in one table