QDocs

QDocs is QBO’s intelligent document processing. It takes the documents that arrive with your work (a loan file, a claim packet, a lease, an application), breaks packages into individual documents, works out what each one is, reads the data you need out of it, and sends anything uncertain to a person. Every result is saved as ordinary QBO data on the record the document belongs to, so your rules, workflows, worklists, and reports use it like anything else.

QDocs is built from QBO’s standard building blocks: Documents, Scoring, AI, Imports, Rules, and Worklists. This guide shows how they fit together for document processing; the platform pages cover each one in depth.

The pipeline

Each stage is an operation that a workflow step, a scheduled job, a menu option, or an API call runs on a document. You choose which stages run, in what order, and for which kinds of document.

Stage	What happens	Where the result lands	AI?
1. Ingest	Documents arrive by upload, drag and drop, the API, inbound email, or a watched sFTP or cloud folder.	A document on a record, in the storage you configured. See Document Management.	No
2. Split	A PDF that holds several documents is broken into one document per segment.	New documents on the same record, each linked to the original package. A single-document PDF is left as one document.	Optional
3. Read	OCR extracts the text of each document.	The document’s description (searchable text).	No
4. Classify	Each document is matched to one of your document templates. Labels can be added too.	The document’s template, and labels on the document.	Yes
5. Extract	The fields you define are read from the text, or from the original PDF or image.	A score on the document: one typed item per field, each with a confidence, plus a weighted confidence for the score. See Scoring.	Yes
6. Review	Rules look at the template, labels, and confidence, and send what needs a person to a worklist.	A task on the right worklist. See Worklists.	No
7. Act	Workflows and rules use the data: open or advance a case, generate a letter, notify a partner.	The next step in your process.	No

Splitting

QDocs can split a PDF in several ways:

• By AI. A multimodal model reads the PDF and returns where each document starts and ends, with a file name and document type for each segment. It can also classify each segment against your document templates and extract fields in the same pass.
• By bookmarks. If the PDF has an outline (bookmarks), each bookmark becomes a document.
• By page ranges you supply, or by size (file size or page count).

By default QDocs splits by bookmarks when a PDF has them, and uses AI otherwise. Each new document is linked to the original as its Document Package Source, so you can always trace a document back to the package it came from.

Confidence and review

Every value AI extracts carries a confidence between 0 and 1, reported by the model. A score’s overall confidence is the weighted average of its items (see weighted confidence). Because confidence is ordinary score data, you decide what happens with it: let confident results flow straight on, and route low-confidence or missing values to a review worklist with a rule.

Models and cost

• Providers. QDocs works with Anthropic Claude, Google Gemini, OpenAI, and Amazon Bedrock. Each task names an AI agent, and each agent names its provider and model, so you can use one model for splitting and another for extraction, and switch by configuration. See Agents.
• Batch processing. Large volumes can be submitted as provider batch jobs, polled on a schedule, and applied document by document as results arrive.
• Context caching. On Gemini, long instructions (such as a full catalog of document types) are cached and reused across requests instead of being re-sent each time.
• Token cost. Token usage can be posted to a ledger at per-model rates, split into input, cached input, output, and reasoning tokens. See Track token costs in a ledger.

Guardrails

• QBO orchestrates the model. A model never decides on its own what to read or change. QBO calls it for a specific document and saves the answer where you configured.
• Your security applies. AI operations run as the calling user, on documents that user can reach.
• Documents stay in your storage. Content is sent only to the provider configured on the agent.
• People stay in the loop. Confidence on every extracted value makes review targeted, and every change to a document is part of the record’s audit trail.
• Behavior is configuration. Document types, labels, and fields are templates with plain-English descriptions and prompts, not code.

Key concepts

Concept	What it means
Document	One file on a record. Split documents are linked to their source package.
Document template	A kind of document, such as Repair Estimate or Lease. Its description tells the model how to recognize it.
Label	A tag such as a language or a review flag, applied to a document when its description fits.
Score template	The set of fields to extract from a kind of document, each with a prompt and a value type.
Confidence	How sure the model is of each extracted value, from 0 to 1.
Agent	A named AI configuration: provider, model, and standing instructions.

Choose your view

Use the buttons under the page title. End user covers working with split, classified, and scored documents and the review queue; Power user covers document templates, score templates, agents, batch jobs, review routing, and cost tracking; Developer covers the operations and their parameters.

For outcomes and use cases by industry, see the QDocs product page. Security and certifications are covered in Security.