Billing Reconciliation

Billing is the legal AI surface where attorney-client trust meets economic reality. An AI billing system that gets it wrong does not just produce an incorrect invoice — it damages the firm's relationship with the paying client, often irrecoverably. This is why our billing-recon deployments ship with a mandatory human-in-the-loop checkpoint that we will not negotiate out of an SOW, regardless of how confident the firm is in their internal eval capability.

The surface drafts billing entries from time-tracker logs, matter activity in the case management system, and document-management activity (DMS edits, drafts, reviews). The draft includes the rate code, narrative, time entry, matter assignment, and client billing-guideline compliance check. An attorney reviews the draft (one-click approve, one-click edit, one-click reject) before any entry posts to the client invoice. Every approval, edit, and rejection is logged and fed into the regression set.

Casetrack ran billing-recon in production for 18 months. We had two clearly wrong outputs in the first year. Both were caught by the human-in-the-loop checkpoint we designed in from day one. Neither reached a client invoice. The takeaway is not that the model hallucinated — that is expected and bounded. The takeaway is that the HITL checkpoint did its job. We do not deploy billing-recon without the checkpoint, and we recommend no firm ever should.