Legal AI is generating measurable efficiency gains but firms cannot locate the financial return because they are looking in the wrong place

A detailed analysis published on 24 August argues that the gap between legal AI's proven operational effectiveness and its absence from law firm profit-and-loss accounts is not a measurement problem but a structural one: firms are looking for the return in billable practice work, where it is hardest to capture, rather than in non-billable operational overhead, where it is clearest. The piece, authored by Alejandro Castellano, CEO of legal AI company Caddi, argues that law firms effectively run two separate economies simultaneously. The first is the billable practice of law, tracked obsessively by utilisation and realisation rates. The second is the non-billable business of law: intake and conflicts, matter opening, docketing, time capture, billing, collections, and document management. The latter is barely measured and, precisely because it is unmeasured, is where inefficiency accumulates unchecked. The analysis identifies a structural problem with applying AI to billable work under hourly billing models: compressing a billable task compresses a billable hour, making the firm's core product cheaper to produce but harder to charge for. The honest response, the author argues, is a transition to fixed-fee or outcome-based pricing, but that is a multi-year renegotiation process, not a return in the current financial year. A further finding is that most firms lack an honest inventory of their own operational processes. Chief operating officers can list their departments and systems but cannot describe which processes run across the firm, how often, who performs them, how many tools each one crosses, or what the aggregate cost is. The result is that automation decisions are driven by whoever complained most recently rather than by evidence of where the greatest inefficiency lies.