Mid-sized commercial firm

The challenge

The firm had bought three AI tools across eighteen months. Usage after the first month was near zero on all three. Partners attributed this to the technology; the actual cause was that no tool had been pointed at a named action, and the associates who would have used them were compensated on billed hours the tools would have deleted. The managing partner's honest position was that the firm could not rewrite its compensation model, and wanted to know what was possible without doing so.

Primitives most affected

The transformation, phase by phase

1. 01

   Name

   We documented the action sequences for the firm's two highest-volume commercial matter types by reconstructing them from real closed files rather than by interviewing partners - the reconstructed sequences differed materially from the described ones, which is the usual finding. We also stated plainly, in week two, that a full AI-native rebuild was not available to this firm and why.

2. 02

   Systematise

   Two action libraries, an explicit doer-and-reviewer chart per named action, and an anchoring rule adopted firm-wide and applied to partners first. The libraries converged three materially different processes that three partners each believed was the firm's process - which delivered measurable quality improvement before any AI was involved.

3. 03

   Automate

   First-pass drafting deployed by default on both matter types, running inside the role chart with the anchoring rule enforced. Critically, the firm decided to hold the client price steady and bank the margin improvement rather than pass it on - which resolved the compensation objection without requiring a partnership vote, because nobody's realisation figure fell.

The shift to AI-native

This is the case that tests the thesis rather than illustrating it. The firm has a leverage pyramid, an equity partnership, and hourly billing - every structural feature the argument identifies as blocking a full rebuild. All of that held: the firm did not become AI-native and will not. What it did was name and systematise all six primitives and install AI as a systematic rather than incidental layer, which is precisely the AI Adoption track. The commercial move that unlocked it was holding price and banking margin, which converted automation from a threat to realisation into an improvement in it. The three previously failed tool purchases were not a technology problem, and the fourth deployment succeeded on the same underlying model - because this time there was a named action underneath it.

Outcome

First-draft time

−61% on top 2 types

Process variance

3 processes → 1

Realisation rate

+9pts