The first group’s real job is not to benefit. It is to produce evidence and a story the second group will believe.
Start a conversation with the AI Adoption Concierge, already scoped to practice group rollout. Pick a starting point, or describe your situation directly.
Firm-wide rollouts of anything tend to fail, and AI is worse than most because the value varies so much by practice area. The sequence that works starts with a group chosen for its likelihood of succeeding rather than its importance — repetitive, well-bounded work, a leader who is genuinely interested, and enough volume that a change is visible in the numbers. That group produces the evidence, the internal case studies and the credible advocates that make the second rollout possible. Underneath all of it sits a question most firms would rather not open: whether people are still measured and paid on the metric the firm is asking them to reduce.
Choose the first group on these, not on revenue or prestige.
Tasks with a clear shape and an obvious quality test. Bespoke advisory work is a poor first target.
Genuine interest from the group head, not compliance with a firm initiative.
Sufficient throughput that a change shows in the numbers rather than being lost in variance.
The group's existing tools and workflow accommodate the new one without a re-platforming project.
Where hours are the measure, efficiency is a personal cost. This is the usual hidden blocker.
A group mid-trial or mid-transaction has none, however willing they are in principle.
How the sequence runs.
Mostly whether there is a second rollout at all.
A firm asking lawyers to become more efficient while measuring and rewarding them on hours has set the two in direct opposition. Every rollout that stalls for unclear reasons should be checked against this first — it is almost always the answer nobody wanted.
One with repetitive, well-defined work, a group head who actively wants it, and enough volume for a change to be visible — high-volume transactional support, document-heavy litigation practice, or a knowledge function are common choices. The instinct to start with the most prestigious or highest-billing group is understandable and usually wrong: their work is bespoke, hard to score, and their partners have the least patience for an imperfect tool. The first group's job is to produce evidence, not to capture the largest benefit.
Name it explicitly, because leaving it unspoken does not make it stop operating. Some firms adjust targets for groups adopting AI on specific work; some move those matter types to alternative pricing so hours stop being the measure; some change how contribution is assessed toward matter profitability and client outcomes. What does not work is asking people to be more efficient while measuring them on volume and hoping goodwill covers the gap — that produces polite adoption in public and none in practice.
Mandating use tends to produce compliance rather than adoption, and compliance is hard to distinguish from success in the reporting. What is reasonably mandatory is the safety layer — the verification step, the confidentiality rules and the training — which apply to anyone using these tools at all. Beyond that, most firms find that a genuinely useful tool with visible internal advocates spreads faster than one people were instructed to use, and spreads to people who then actually rely on it.
Longer than the plan, and the honest expectation is measured in years rather than quarters for anything approaching universal use — which very few firms have reached. The data consistently shows most adoption remaining partial and uneven, concentrated in larger firms and in particular groups. A realistic sequence is a first group demonstrating value over a couple of quarters, then two or three more, with the practice areas whose work suits it least coming last or not at all. Planning for eighteen months to a broad rollout is more defensible than planning for six.
Describe your practice groups and how people are measured. The Institute will help you sequence it.