Artificial intelligence can produce a legal draft in minutes, but the work needed to make that draft reliable may be obscuring the technology’s true financial impact on law firms and their clients.

Lawyers must still check facts and citations, assess legal reasoning, correct errors and take responsibility for the finished product. That review can consume enough time—and require enough senior expertise—to narrow the savings suggested by a near-instant first draft.

The issue has taken on urgency amid broader questions about AI’s economic return. Torsten Slok has argued that, outside big technology companies, AI has yet to deliver a financial payoff to corporate America. In legal services, one possible explanation is that firms are not fully accounting for the human work required after AI generates an answer.

Harshit Garg, founder and CEO of a company that builds agentic AI for lawyers, raised the measurement problem in an article for Legal Insider. AI providers can track tokens, generation speed and the cost of producing output, he noted. Firms, however, may not connect that usage to the time lawyers subsequently spend reviewing and revising it.

The distinction matters for both profitability and pricing. In a hypothetical comparison, a document that traditionally takes four hours to draft and one hour to review requires five hours of work. If AI reduces drafting to 0.1 hours but doubles review time to two hours, the total falls to 2.1 hours—a saving of about 58%, rather than the near-total reduction implied by drafting speed alone.

The dollar savings can be smaller still when a senior lawyer performs more of the work. At an illustrative $250 an hour for an associate and $800 for a partner, the traditional approach costs $1,800. The AI-assisted version costs $1,625, a reduction of less than 10%, before accounting for the cost of the AI tool.

That gap can complicate conversations with clients who have heard sweeping claims about AI-driven efficiency. Firms may have a legitimate reason for substantial review charges, but explaining them becomes harder if they cannot show how the review relates to the AI-generated work. Clio has found that 57% of consumers believe faster work should result in lower costs.

For firms assessing AI, the more useful comparison is therefore not drafting time against generation time, but the total cost of delivering an acceptable final product. Measuring that work could help firms evaluate their investments more accurately while giving clients a clearer picture of where savings—and professional judgment—fit into the bill.