docRefinery vs Azure Document Intelligence

Models you assemble, or a workflow you configure

Azure Document Intelligence, now part of Microsoft's Foundry Tools, is a mature service with three families of models. Read and Layout extract text and structure. Prebuilt models cover invoices, receipts, ID documents, bank statements, pay stubs, contracts and a long list of US tax and mortgage forms, with field schemas Microsoft defines and maintains. Custom models extract your own fields after you label a training set, and custom classifiers identify document types before extraction.

Building a solution means choosing and composing those models, labelling samples and adding review and business logic around them. docRefinery packages that into one product for back-office teams: documents in, checked, structured data out.

Where Azure Document Intelligence is the better choice

• US forms out of the box. Prebuilt models for W-2, 1040, 1098 and 1099 variants, mortgage forms such as the 1003 and closing disclosures, health insurance cards and pay stubs go well beyond our catalogue.
• Microsoft estate and compliance. Microsoft lists Azure Document Intelligence in Foundry Tools as in scope for FedRAMP High, and Azure runs a SOC 2 Type 2 programme whose in-scope services are listed on the Service Trust Portal. docRefinery's SOC 2 Type II audit is in progress with no report yet, and SSO is planned for Enterprise but not available yet.
• Disconnected containers. Read, Layout, prebuilt and custom template containers can run with no cloud connection, after an access request and a commitment plan. That suits sites that cannot reach the internet.
• Low prices for text. Read lists at $1.50 per 1,000 pages, falling to $0.60 above a million, with commitment tiers for steady volume.
• Office formats. Read and Layout accept Word, Excel and PowerPoint files. Presentations and wider office formats are in development at docRefinery and not yet available.

What docRefinery does that Azure Document Intelligence does not, or does differently

docRefinery fits when you do not want to own a labelling and training cycle, and when review and governance should be in the box.

• No labelling cycle. Azure's guidance for custom models is to include every variation in the training set, separate visually distinct documents into their own models, and retrain with more labelled samples when confidence drops. With docRefinery, one sample is enough to design a workflow.
• Layout memory. Each new sender's layout is learned once and remembered, and re-trained when a form changes, with no model per variation for your team to keep.
• You define the output. Each field has a plain-language instruction, a "Return as" name you choose as its output key, a type and a required flag, rather than a schema set by the model.
• One output format for every document. Every field returns text, alias, field_confidence and location, and every table returns rows, so one parser handles a passport and a bank statement alike.
• Long files, segmented. Long, multi-page files are handled at scale and separated into the documents inside them, each classified into its type, without composing classifiers and models yourself.
• A review queue in the product. Microsoft's docs suggest using confidence to decide what to send to a person and leave that step to you. In docRefinery, values below a per-field threshold go to a queue. Our guide to confidence scores covers how to set them.
• Learns from corrections. A reviewer's correction becomes a suggested fix, tested on that layout's own documents before anyone applies it.
• Governed fields. Owner, admin and member roles decide which fields may be extracted, and a field missing from the page comes back null.
• Custom types at no charge. Document types outside our catalogue are trained, hosted and re-trained for you at no charge.
• Air-gapped on premises. docRefinery runs fully air-gapped on your own servers on Enterprise, on models that run inside your network.