The problem

Bank reconciliation is a large amount of repetitive matching punctuated by a small number of items that genuinely require judgment. Doing all of it manually spends most of the effort on the part that does not need a human, which leaves less attention for the part that does.

The approach

I separated the work into what can be matched deterministically, what can be proposed with reasonable confidence, and what must be decided by a person. Clear matches are handled automatically. Probable matches are proposed with the reasoning attached, and a reviewer accepts or rejects them. Anything that does not fit either category is escalated untouched. The control point is preserved: the automation never closes out an exception on its own, and every automated action stays on an audit trail.

Tools

  • n8n
  • Claude API
  • NetSuite
  • Bank data feeds

What it enables

  • Reconciliation effort concentrates on genuine exceptions rather than routine matching.
  • Every automated action is logged and reviewable, so the control environment is unchanged.
  • Reconciliations complete earlier in the close sequence.

My role

I defined the automation boundary and the review points, built the workflows, and documented the control design.

Related

Back to all projects