A Manual Invoice Process Where Every Batch Meant Hours of Reading and Re-Keying
Invoice processing relied heavily on people reading source documents and transferring information into Xero. That made the process time-intensive, but the larger concern was the number of decisions and checks required before an invoice could safely enter the accounting system.
Manual PDF Reading and Xero Entry
Supplier invoices had to be opened, reviewed, interpreted, and keyed into Xero manually. A single batch could require approximately three to four hours of processing, with the same steps repeated each time new invoices arrived.
Different Invoice Layouts Across Suppliers
There was no single standard invoice structure. More than 20 supplier layouts had to be accommodated, with differences in where descriptions, quantities, rates, VAT information, references, and totals appeared. A generic extraction model was therefore not sufficient.
Risk of Incorrect Amounts and Accounting Codes
Manual entry created the possibility of transposed amounts, incorrect general ledger classifications, and missed UK VAT treatment. Even when invoice data was entered correctly, the appropriate accounting treatment still had to be determined separately.
Entity-Specific and Mixed-Cost Transactions
Some invoices contained charges relating to different operating entities. For example, the workflow needed to identify references such as inter-company vendor, apply the correct account treatment, and ensure the final file remained distinguishable during processing.
Duplicate Invoice Risk
Preventing the same supplier invoice from being processed more than once required more than checking the current batch. The system needed a way to compare incoming invoices against both batch activity and existing Xero AP records.
Limited Auditability
The manual process did not create a definitive, structured processing trail covering extraction, classification, validation, exceptions, and duplicate checks. This made it harder to establish exactly what had happened during an invoice-processing run.
Turning Invoice Processing into a Controlled Automated Pipeline
A keying-heavy routine became one automated pipeline that reads, checks, and prepares every invoice for Xero. It reflects the growing role of AI in finance operations, where speed only counts if accuracy and traceability come with it.
- Supplier PDFs read manually
- Invoice details keyed into Xero
- Supplier identification handled manually
- Generic or manually interpreted line items
- GL coding determined during manual processing
- VAT treatment manually reviewed
- Duplicate risk checked manually
- Mixed entity invoices handled individually
- Invoice data extracted automatically
- Xero-ready import file generated automatically
- Supplier matched against approved Xero contacts
- Bespoke line-item extraction for different supplier layouts
- GL classifications mapped using historical Xero coding
- Multi-layer VAT validation built into the workflow
- Three-tier duplicate detection
- Entity-specific processing logic applied automatically
Make AP Processing Faster Without Losing Control
Reduce repetitive invoice handling, strengthen validation, and prepare cleaner Xero-ready data through structured AP automation.
A Five-Stage Pipeline Built Around Real Invoice Complexity
The solution was designed around the client's actual supplier documents and Xero environment rather than using a generic invoice-processing template. Each stage addressed a different source of manual effort or processing risk.
Intelligent Invoice Extraction
- Supplier PDFs were processed using pdf to text and other concerned Python library.
- Preserved native tabular column alignment so figures stay in the right columns.
- Scored every extraction from 0–100% for OCR confidence, sending low scores to a human for visual sign-off.
- Pulled data directly from source files, removing the need to retype any figure.
Vendor Identification and Matching
- Used a two-stage recognition process to map each file to one of 300+ approved Xero contacts.
- Detected the second venue’s name in the invoice text and assigned its dedicated account code automatically.
- Appended an entity tag to the filename so separate entities stay separate through to Xero.
- Targeted the true invoice reference where suppliers print order and invoice numbers on the same page.
Supplier-Specific Line-Item Extraction
- Built custom regex extractors for 20+ vendor layouts to capture product descriptions, quantities, and rates.
- Skipped repeated page-break headers so no lines are dropped from multi-page tables.
- Split mixed-rate invoices into separate Xero import rows for each VAT rate.
- Re-calculated quantity × price independently and flagged any variance above £0.005.
GL Classification and VAT Validation
- Mapped accounts and tax types using AI-driven classification trained on 12 months of historical Xero data.
- Scored each GL assignment against actual coding history.
- Reconciled calculated VAT, PDF VAT, and import VAT, which must agree within £0.02.
- Ran a completely fresh extraction pass to verify the final invoice total independently.
Duplicate Detection and Controlled Output
- Applied a three-tier duplicate defence that cross-references live Xero AP history.
- Checked an append-only registry so the same invoice reference can never be processed twice.
- Generated a five-tab review workbook separating clean data, flagged items, credit notes, and blocked duplicates.
- Produced a multi-column ERP Bills import CSV with explicit per-line tax amounts to prevent rounding drift.
- Renamed source PDFs in a “Vendor Name-Amount-InvoiceRef" format for fast attaching in Xero.
- Issued a Markdown audit report giving each batch a definitive final status.
Faster Invoice Processing with Stronger Built-In Controls
The new workflow replaced a large amount of repetitive invoice handling while introducing a more structured validation and review process. The gain shows up in time saved, and just as much in the accuracy and proof behind every batch.
Key Takeaway for Finance Leaders
Three Principles Behind Reliable AP Automation
Automate Decisions, Not Just Data Entry
Reading an invoice automatically provides limited value if employees still have to identify the supplier, select the GL code, determine VAT treatment, check duplicates, and prepare the accounting import manually. The greater efficiency came from connecting these activities into one workflow.
Build Validation Separately from Extraction
Automation should not assume that extracted data is correct simply because the system successfully read the document. Independent amount calculations, VAT reconciliation, confidence scores, total re-reading, and duplicate controls create a stronger processing environment than relying on extraction alone.
Design Around Real Supplier Exceptions
AP automation performs best when it reflects the invoices the business actually receives. Mixed VAT rates, entity-specific costs, unusual invoice references, multi-page tables, and supplier-specific formats need defined handling rules. Otherwise, exceptions eventually push the process back into manual work.