SyncBook's AI pipeline combines Google Cloud Vision OCR with a fine-tuned large language model to automate bookkeeping tasks. The system performs structured data extraction, transaction categorisation, anomaly detection, and cash flow forecasting — reducing manual accounting work by up to 80%.
MODEL PERFORMANCE METRICS
Document Processing Pipeline
End-to-end AI extraction workflow
Live AI Demonstrations
Interactive examples of each AI capability
The AI analyses vendor name, description, and amount to assign the correct UK Chart of Accounts category. Transactions below 70% confidence are flagged for accountant review.
⚠ 1 transaction flagged for review (confidence < 70%). Accountant action required.
Confidence Scoring System
How the AI rates its own certainty
Auto-approved. No accountant review required. Field is reliably extracted.
Accepted with caution. Accountant may wish to spot-check the value.
Flagged for mandatory review. Document appears in the Review Queue.
Confidence scores are computed per field (vendor, amount, VAT, date) and aggregated into an overall document score. The threshold of 0.70 is aligned with UK accounting practice standards for automated data entry systems.
Technical Architecture
Stack and integration details