Read every document. Surface what matters.
An agent that reads contracts, drawings, RFPs, leases, and scanned PDFs end-to-end - extracts the fields you care about, cites where it found them, flags what is uncertain, and writes the structured result into your systems. Six hours of review becomes thirty minutes of confirmation.
Document Intelligence at Compass means a custom AI agent that reads your contracts, drawings, RFPs, and PDFs end-to-end, extracts the fields you care about, cites where it found them, flags low-confidence results for human review, and writes the structured output back into your Document Management System or matter file. Most builds cut document review time from six hours to thirty minutes. Built for Canadian SMBs in engineering, legal, real estate, and accounting. Engagements typically run $12,000–$35,000 CAD; scoped to your document types, volume, and integration depth during the Bearings call.
One document in. A structured answer out.
This is a preview of how a Document Intelligence agent built for your firm would behave. The data below is illustrative - the engine, the citations, and the flagging logic are real.
A commercial lease lands in the watched folder. Agent classifies it in under a second.
Four behaviours on every document.
The same pipeline whether you feed it a contract, a building plan, or a scanned invoice. Type-aware extraction, citation-backed answers, and a written record of what was read and decided.
Reads any document you throw at it
PDFs, scans, photos, Word files, technical drawings. Classifies the document type first, then runs the right extraction logic for that type.
Extracts with citations, not guesses
Every extracted field links back to the page and paragraph it came from. Click any value and see the source text highlighted. No black-box answers.
Flags what it is not sure about
Low-confidence extractions get flagged for a human pass instead of silently guessing. You see exactly what needs review and what is safe to trust.
Writes results back into your systems
Structured data lands in your DMS, matter system, CRM, or spreadsheet. No copy-paste. No re-keying. The agent finishes the workflow, not just the reading.
Same engine. Different document.
Every build is configured for the document type, the workflow, and the system you want the result in. Below is a sample of what we have scoped and shipped.
Contract review without the 6-hour markup
Drop a 40-page commercial lease into the agent. Get back: parties, term, rent escalation, options, indemnities, conditions precedent - with citations - in under 90 seconds. Lawyer reviews the structured summary, not the whole document.
Spec extraction from RFP documents
Feed it a 200-page municipal RFP. Get a structured matrix of every technical requirement, deadline, submission rule, and disqualifier - cited to the page. Decide go/no-go in a meeting, not over a weekend.
Lease abstract automation
Portfolio with 200 commercial leases? The agent reads every lease, builds a unified abstract (term, rent, options, restrictions), and flags anything non-standard. Three days of paralegal work becomes a coffee break.
Receipt and invoice OCR with categorization
Photograph receipts, drop in invoices, forward email attachments. The agent extracts vendor, amount, GST/HST, date, category - and posts to QuickBooks or Xero. Monthly bookkeeping pile shrinks 80%.
Operator-grade infrastructure. Canadian data residency.
The specific tools depend on your situation. The principles below hold on every build - we pick from a vetted bench and tell you exactly what is in the stack before scope is signed.
Document intelligence, answered plainly.
What kinds of documents can the agent actually handle?
Contracts, leases, NDAs, RFPs, technical specs, engineering drawings, financial statements, insurance policies, building plans, court filings, and any PDF or scanned image. The agent classifies the document type first, then runs the right extraction logic - it does not treat every document the same.
Does it cite where it found things, or do I have to trust it blindly?
Every extracted field links back to the exact page and paragraph in the source document. If the agent says "termination clause is 60 days notice", you can click the citation and see the original text highlighted. No black-box answers. This is the difference between an AI tool and an audit-ready system.
How accurate is it on scanned or handwritten documents?
Typed PDFs and clean scans run 97-99% accurate on standard fields. Older scans with skew or noise drop to 88-94% - the agent flags low-confidence extractions for human review instead of guessing. Handwritten documents need a human pass; the agent extracts what it can and clearly marks the rest.
Where does the document data live? Is anything sent to OpenAI or stored offshore?
Documents land in a Canadian-residency database (Montreal region by default). Processing runs through an enterprise-tier LLM with zero-retention. Nothing trains a public model. Nothing leaves Canada unless you explicitly opt in. PIPEDA (Personal Information Protection and Electronic Documents Act) compliant by default.
Does it integrate with our DMS (Document Management System), SharePoint, or matter management system?
Yes - common integrations are SharePoint, Google Drive, Dropbox, NetDocuments, iManage, Clio, PracticePanther, and any DMS with a REST API (Representational State Transfer Application Programming Interface) or webhook. The agent can watch a folder, process new files, and write results back into your system without anyone uploading files manually.
How much does this cost to build?
Every document intelligence build is custom-scoped. The bill depends on: document variety (one type vs. fifteen), volume per month, the destination system results land in (your DMS, matter system, spreadsheet, accounting platform), whether you need on-premise hosting, and how much process mapping is required to get the workflow right. Single-document-type pilots are the lightest entry and the way most engagements de-risk. Real numbers come out of the free Bearings call once we have mapped what you actually need.
Every contract you skim is a clause you missed.
The average commercial lease has 47 material clauses. The average lawyer reads it carefully once and skims it twice. The average paralegal abstract takes four hours. The agent on this page reads every clause, every time, on every document - and writes down what it found.
// Related capabilities
// Built for
Plot Your
Course.
A free 30-minute Bearings call. Two operators talking about your business. No deck. No pitch. Just an honest read on whether AI is the right tool - and if so, where I'd start.