Stop Calling a Technology Budget an AI Strategy

AI-generated editorial illustration. People depicted are fictional.

A technology budget is an amount of money. Calling it an AI strategy does not make it one.

The 2026 US legal market report gives buyers a useful spending baseline. It does not give any individual firm a receipt for better judgment. My reaction to Anirudh Menon’s paired analysis is that the first procurement question should be painfully concrete: show one completed matter whose result became more useful to the client.

Use the AI contract review evaluation guide to define the buying task and acceptance checks for the wider workflow.

Ask for a result someone was willing to rely on, with the cost of checking it included.

That standard is harder to satisfy than a product demonstration because it asks who benefited. A lawyer who produces an early draft faster may have created room for deeper analysis. A department may have shortened a queue. A client may have received a lower bill. Those are all plausible gains. They are different claims, and each needs its own evidence.

Use the CLM business-case worksheet to separate capacity gains from realizable cash savings.

I would add one requirement to a pilot: write the rejection memo before the purchase. State the finding that would make the team decline expansion. Examples to examine include:

• Reviewers spending too long reconstructing the reasoning.
• The product losing the link between an agreement and its amendment.
• Savings that depend on additional work the team does not have.

Turn a proposed rejection rule into evidence with the paired AI contract-review tests rather than relying on a demonstration.

If nothing could make the pilot fail, it is a launch campaign with test documents.

There is also a limit to financial creativity. ABA Opinion 512 says hourly bills must reflect time worked. This is Model Rules guidance; applicable professional rules still control. Dressing an efficiency gain in a familiar billing narrative is not a business model.

This is an argument for a better purchase, not a prediction that legal AI will disappoint. I would rather see a modest deployment with a clear owner, a measurable benefit and an honest stopping rule than an ambitious rollout whose success is defined by how many people opened the application.

Ask the sponsor to record the finding that would stop expansion, who will assess it and when the team will decide.

This is opinion responding to a January 2026 report on earlier market data. It offers no new market survey or product test.