Journal entry automation

Problem

A US natural gas operator's accounting team was building journal entries from a continuous stream of incoming bank statements by hand. The work was slow, repetitive, and error-prone — and because the workflow sits inside SOX scope, every entry had to be traceable, validated, and defensible in an audit. Automation was only acceptable if it strengthened the controls rather than bypassing them.

Approach

I led the engineering team and owned the architecture from requirements through production rollout. The pipeline ingests bank statements as they arrive, uses AI-driven parsing to extract and classify transactions, then pushes everything through a validation and controls layer before any number reaches the ERP. Validated transactions are consolidated into ERP-ready journal entries, with a full audit trail preserved at every step.

The solution was deployed across dual Azure tenants, keeping client and delivery environments cleanly separated. Alongside the build itself, I managed the project lifecycle, oversaw client deployments, and maintained direct client communication throughout — requirements, UAT, and production support.

Architecture

Outcome

The pipeline runs in production, removing hours of manual statement-to-entry work while keeping accountants in control of what posts. Every generated entry is validated and fully traceable, so the automation strengthens the audit position instead of complicating it — exactly what a SOX-scoped workflow requires.

Stack

• Azure
• AI / LLM
• Python
• ERP integration
• SOX controls

Have a similar problem? Let's catch up →