Who actually owns the rollout?
A platform gets bought by one group, deployed by a second, and used by a third — which is why so many legal AI programmes look excellent on paper and thin in practice. This is for whoever has to make it real.
01 The handoff problem
Watch how a legal AI platform usually enters a firm. A committee evaluates vendors. IT and security run diligence. Innovation or knowledge management configures it. Then it is announced, and ownership quietly evaporates — because the people who bought it have finished their job, and the people who need to change their working habits never had one.
The result is a deployment with no owner. Everyone can point to the platform. Nobody is accountable for whether a mid-level associate in a regional office is any good at it.
Deployment is a project. Adoption is a programme. Projects end. Programmes have an owner, a budget line, and a number they are judged on. Most firms fund the first and hope for the second.
02 Three failure patterns
The enthusiast ceiling
Usage climbs fast for six weeks, then flattens. What happened is that everyone predisposed to try new tools has tried it, and the curve has hit the edge of that population. Further growth requires reaching people who are not curious about AI — a completely different exercise, and one that a launch email cannot do.
The permission fog
Ask five people what they are allowed to put into the firm's AI platform and you get five answers, most of them more conservative than the actual policy. Uncertainty reads as risk, and sensible professionals respond to risk by not doing the thing. This is usually a communication failure rather than a policy failure, and it is cheap to fix.
The demo-matter gap
Training is delivered on a clean, invented matter where the tool performs beautifully. People return to their own work, which is messier, and the transfer does not happen. The fix is unglamorous: teach on live work, accept that the demo will be less impressive, and let people watch the tool struggle a bit. Watching a recovery is more instructive than watching a success.
03 Who should own it
There is no single right answer, but there is a reliable test: the owner should be someone whose performance is judged on whether people use it well, not on whether it was delivered.
| Candidate owner | Honest assessment |
|---|---|
| IT / CIO | Right for security, identity, integration. Wrong as the adoption owner — lawyers do not change professional habits because IT asked. |
| Innovation / KM | Usually the best home. Has the process knowledge and the credibility. Often lacks authority over practice groups, which has to be granted explicitly. |
| Practice-group leaders | Where the actual influence is. Cannot run the programme, but nothing works without at least a few of them visibly using it. |
| L&D / professional development | Owns the delivery muscle and the calendar. Frequently left out of AI entirely, which is odd given the problem is a training problem. |
| The vendor | Will train on their product, capably. Will not tell you when the answer is to not use their product. |
A firm wants to lift usage among mid-level associates. It can fund exactly one of: more licences, a better platform configuration, or eight hours of practice-specific training per associate. Which moves the number most?
Commit to an answer first.
Training, and it is usually not close — provided the associates already have licences. Unused licences are not a capacity constraint, and configuration mostly improves the experience of people already using the thing. The binding constraint at that stage is almost always skill and confidence, not access.
The exception worth naming: if licences are genuinely rationed and your target population does not have one, nothing else matters until they do.
04 The group nobody plans for
Firmwide AI programmes are, in practice, attorney programmes. The licences, the training, the policy examples and the success metrics are all built around fee earners, because that is where the billable-hour arithmetic is legible.
Meanwhile a large fraction of any firm — paralegals, marketing and business development, finance, HR, records, secretarial and administrative staff — does work that these tools handle extremely well, and receives no training at all. Nobody decides this. It falls between budgets.
We give that group its own lesson later in this track, because it is the most reliably overlooked source of return in a legal AI deployment.
A rollout health check
- There is one named person accountable for adoption, distinct from deployment
- Every user can state, without checking, what they may and may not put into the tool
- Training has been delivered on live matters, not demo matters
- At least a few visible senior practitioners use it in front of others
- Someone looks at the usage data monthly and acts on it
- Non-fee-earning staff appear somewhere in the plan
This week
Ask three people in different roles — one fee earner, one paralegal, one business-services colleague — the same question: what are you allowed to put into the firm's AI tool? If the three answers differ, you have found the cheapest available improvement to your programme.