365 Business Finance — Machine learning that learns from underwriter decisions — automating transaction categorisation while improving accuracy over time.

Transaction categorisation at scale requires underwriters to read, interpret, and classify thousands of records manually. Manual overrides are common, time-consuming, and represent valuable institutional knowledge that is lost once the decision is made. The business needed automation that could replicate that knowledge and get smarter as it received feedback.

• Two-phase architecture: rule-based categorisation first, ML-powered classification second
• AWS Textract extracting transaction description text from PDFs, receipts, and unstructured documents
• SageMaker ML model trained on historical transactions and manual underwriter corrections — improving accuracy with every cycle
• Underwriter interface for manual review and override — every correction feeds the training dataset
• Secure data ingestion via API or batch upload to S3 with encrypted storage and transmission
• FCA and data protection regulatory compliance maintained throughout processing and storage

• Built Phase 1 rule-based categorisation — Lambda-based text matching rules applied to transaction descriptions extracted by Textract.
• Designed and built the underwriter override interface — allowing corrections to be applied easily and captured for model training.
• Trained and deployed Phase 2 SageMaker classification model on historical transaction data and accumulated manual corrections.
• Implemented automated retraining pipeline — model accuracy improves progressively as the corrections dataset grows.
• Provisioned secure infrastructure: encrypted storage, encrypted API transmission, IAM access controls, and FCA-aligned data governance.
• Validated accuracy and performance through rigorous testing under production-representative transaction volumes.

• Transaction categorisation automated — underwriters review exceptions rather than processing every record manually.
• ML model continuously improving — every underwriter correction makes the next categorisation more accurate.
• FCA-compliant data handling in place — secure, encrypted, and auditable throughout.
• System handles growing transaction volumes without performance degradation.
• Underwriters freed from routine categorisation to focus on complex decisions that require human judgement.