How to Extract and Validate Data Across Related Documents

The chain of documents nobody checks completely

An invoice arrives. It references Purchase Order 4817. PO 4817 references Master Services Agreement MSA-2024-031. The MSA defines billing rates, payment terms, and scope of work. Somewhere in this chain, the invoice charges $185/hour for a role that the contract prices at $175/hour. The difference is $10/hour across 320 hours -- $3,200 of overcharges that nobody catches because nobody has time to trace the full document chain.

This is the cross-document validation problem. Business documents do not exist in isolation. Invoices depend on purchase orders. Purchase orders depend on contracts. Contracts depend on rate cards and statements of work. Each document in the chain is supposed to be consistent with the documents it references. In practice, consistency is assumed more often than verified.

The reason: manually tracing a document chain takes 20 to 30 minutes per invoice when done thoroughly. For a company processing 200 invoices a month, thorough validation would consume one full-time employee's entire month. So teams spot-check a few line items and trust the vendor to bill correctly. The exceptions -- rate overcharges, duplicate charges, unapproved scope additions, expired contract rates applied to new work -- add up to real money over time.

What cross-document validation actually requires

Validating data across related documents involves three distinct operations.

Extraction. Pull specific data points from each document in the chain. From the invoice: line items with descriptions, quantities, unit prices, and totals. From the PO: approved line items, quantities, and total authorized amount. From the contract: billing rates by role, payment terms, scope boundaries, and effective dates.

Matching. Link data points across documents. The invoice's "Senior Consultant -- 160 hours at $185/hr" needs to match against the PO's "Senior Consultant -- 160 hours at $175/hr" and the contract's rate card entry for "Senior Consultant -- $175/hr effective January 2024."

Validation. Compare matched values and identify discrepancies. Does the invoiced rate match the PO rate? Does the PO rate match the contract rate? Does the invoiced quantity fall within the PO's authorized quantity? Is the contract still in effect for the invoiced period?

Each operation is simple in isolation. The difficulty is performing all three across multiple documents with enough precision to catch real discrepancies without drowning in false positives.

Setting up cross-document validation in docrew

Here is how to extract data from a chain of related documents, cross-reference values, and identify discrepancies.

Step 1: Gather the related documents. Collect the documents that form the chain. For an invoice validation workflow, this means the invoice, the purchase order it references, and the contract or rate card that governs the engagement. Place them in a folder on your machine.

For this walkthrough, assume you have three files: an invoice PDF from Meridian Consulting, the corresponding purchase order as a DOCX file, and the master services agreement as a PDF.

Step 2: Describe the validation you need. Tell the agent what to extract from each document and how to compare the values. A strong prompt:

"I have three related documents: an invoice from Meridian Consulting, the purchase order it references, and our master services agreement with Meridian. Extract line items with descriptions, quantities, unit rates, and totals from each. Then cross-reference: does each invoiced line item appear on the PO? Do the invoiced rates match the PO rates? Do the PO rates match the contract rate card? Flag any discrepancies with the specific values from each document."

This prompt tells the agent what to extract, how to match across documents, and what to report. The agent now has a clear validation framework.

Step 3: The agent reads and extracts. The agent opens each document in sequence. It reads the invoice and extracts every line item -- description, hours, rate, extended amount, subtotal, tax, total. It reads the PO and extracts the approved line items with quantities and rates. It reads the MSA and locates the rate card, payment terms, and relevant scope definitions. All of this happens locally on your machine.

Step 4: The agent cross-references. With data extracted from all three documents, the agent maps each invoice line item to the corresponding PO item by description and role, then compares against the contract rate card. A typical output:

"Invoice line 1: Senior Consultant, 160 hours at $185/hr = $29,600. PO line 3: Senior Consultant, 160 hours at $175/hr = $28,000. Contract rate: $175/hr effective 1 Jan 2024. DISCREPANCY: Invoice rate exceeds both PO and contract rate by $10/hr. Overcharge: $1,600."

"Invoice line 2: Project Manager, 80 hours at $195/hr = $15,600. PO and contract rate: $195/hr. MATCH: All values consistent."

"Invoice line 3: Junior Analyst, 80 hours at $125/hr = $10,000. No corresponding PO line item. Contract rate: $120/hr. DISCREPANCY: Not authorized on PO. Rate exceeds contract by $5/hr."

Step 5: Review and act. You review the flagged discrepancies and decide how to handle each one. The Senior Consultant rate discrepancy goes back to the vendor for correction. The Junior Analyst line item needs investigation -- was it authorized verbally but not added to the PO? You have specific values and document references to support the conversation.

Scaling to multiple invoice chains

The single-chain validation takes the agent a few minutes. The real power emerges when you apply the same validation across a batch.

Batch validation prompt: "I have 25 invoices from various vendors. For each, find the corresponding PO and contract. Extract line items and cross-reference rates, quantities, and totals. Produce a summary report: vendor name, invoice number, invoice total, PO total, number of discrepancies, and total dollar amount of discrepancies."

The agent works through each chain systematically -- identifies the PO number, locates the PO file, finds the contract, extracts from all three, and validates. Twenty-five invoice chains processed with a consolidated exception report at the end. A week of manual spot-checking becomes an afternoon of reviewing exceptions.

Common discrepancies the agent catches

Rate escalation without authorization. A vendor increases their billing rate without a contract amendment. The agent catches this because it compares every invoice rate against the contract rate card, not just the PO.

Quantity overruns. The PO authorized 200 hours for a role. Cumulative invoices total 240. No single invoice exceeds the PO, but the running total does. Ask the agent to track cumulative quantities across invoices against PO limits.

Duplicate charges. The same work appears on consecutive invoices -- a time entry billed in both September and October. The agent flags duplicate descriptions and overlapping date ranges.

Expired rates. The contract rate card has effective dates. A vendor bills at the 2024 rate for 2025 work, or applies a 2025 rate increase to 2024 work.

Scope creep. An invoice includes "Data Migration Support" that appears in neither the statement of work nor any change orders. The agent identifies line items with no basis in the governing documents.

Math errors. Hours times rate does not equal the extended amount, or line items do not sum to the subtotal. The agent checks every calculation automatically.

Beyond invoices: other document chains

The same approach works for any set of related documents.

Insurance certificates against contract requirements. Your contract specifies minimum coverage -- $2 million general liability, your company named as additional insured. The vendor sends a COI. The agent extracts coverage amounts and named insureds from the COI and compares against contract requirements.

Lease terms against tenant correspondence. A tenant requests a rent reduction citing a lease provision. The agent reads both documents and provides a point-by-point comparison of the claim against the actual terms.

Regulatory filings against source data. A compliance filing references metrics from internal reports. The agent traces each reported figure back to its source and flags mismatches.

Proposal terms against signed contract. After negotiation, the signed contract should reflect the proposal. The agent flags every deviation between what was proposed and what was signed.

Structuring your validation workflow

Organize by chain. Keep related documents together or in predictable folder structures. Name PO files to include the PO number so the agent can locate corresponding documents quickly.

Define your validation rules once. The first time you describe your cross-referencing requirements, be thorough. The agent retains context for follow-up batches.

Run validation on receipt. Validate each invoice as it arrives while corrections are easier to obtain from vendors.

Track patterns. Ask the agent to summarize discrepancy patterns across vendors. "Which vendors have had the most rate discrepancies in the last six months?" This surfaces systemic issues that invoice-by-invoice review might miss.

Save the output. Have the agent write results to a spreadsheet, creating an audit trail of what was found and how it was resolved.

Studies consistently find that 1-3% of invoice value contains errors. For a company with $10 million in annual vendor spend, that is $100,000 to $300,000 in discrepancies per year. Cross-document validation with docrew removes the time barrier: every invoice gets full extraction, full cross-referencing against the PO and contract, and clear reporting of every discrepancy. The documents stay on your machine, the validation happens locally, and the output is a clear, actionable exception report.