How do mid-size law firms decide which AI requests to fund?
Not by whoever asks loudest, at least not on paper. About 40% of mid-size firms now route AI requests through a committee or an approval process. Yet 44% still have no formal way to tell a senior partner no, and on plenty of weeks the loudest voice wins anyway, process or no process. The firms making real progress share one habit: they tie tool access to proven use before the first pilot, not after the unused licenses pile up. The problem underneath all of it is shelfware, tools bought under pressure and touched by almost no one.
| What mid-size firms report | Share |
|---|---|
| Have no formal mechanism to turn down a senior request | 44% |
| Route AI requests through a formal committee or approval process | 40% |
| Deliberately reward early adopters with first access | 16% |
How we know this
Sidebar puts one question a week to legal management professionals at firms of 10 to 200 attorneys. Members are verified by title, employer, and firm size before they are admitted, and every reply is private. This page draws on every Sidebar cycle that has touched this question, and it is updated as new replies come in. We publish patterns across the group, never individual firms, and only once at least five members have replied to that question. Full methodology at gosidebar.ai/methodology.
Committees are the default answer, and authority is their weak point
Roughly 40% of firms built some version of an AI committee, a technology committee, or an approval process to route requests through. On paper that is governance. In practice it only works if the group can actually reject a request, including one from a senior partner. A committee that meets on schedule but cannot say no does not improve decisions. It just adds a step, and slows everything down while the real decision defaults to whoever pushes hardest. The tell is easy to spot from the outside: ask a committee member to describe the last request they turned down. Firms with real authority behind the process have an answer ready. Firms with a committee that exists mostly to be seen existing do not, and the silence is the more honest answer of the two.
Shelfware is the crisis nobody names in the meeting
The same story shows up across firm after firm: a tool bought under pressure from a senior voice, used by a handful of people, ignored by everyone else. Firms are blunt about it: the honest answer is they are trying to stop buying before they have built the habit. Money spent on licenses no one opens is not a technology problem. It is a governance failure, and it is the direct cost of letting the loudest request set the budget. It also compounds in a specific way that a single bad purchase does not: the next senior voice who wants a tool points to the unused license sitting on the books as proof that AI spend does not pay off, when the actual lesson was narrower and had nothing to do with the technology itself.
A formal committee does not stop adoption happening around it
The governance structures firms describe are real, but they are not the whole picture. Regardless of what a committee approves on paper, attorneys and staff are reaching for AI tools on their own, sometimes with a manager's blessing and sometimes without one, well ahead of any formal review. That gap between the approved list and what people are actually opening is where the real compliance exposure sits, because a committee can only govern what it knows is happening. A firm with an airtight approval process and no visibility into ungoverned use has solved the easier half of the problem and left the harder half untouched. Committees rarely find out about this the easy way. It tends to surface during a security review, a client audit, or a conflict check that turns up a tool nobody on the approval list remembers authorizing, at which point the firm is explaining after the fact rather than governing in advance.
Pilot groups are becoming the standard
The model more and more firms are landing on looks the same: one representative per practice area, licenses handed out selectively against a real use case, and a deliberate path from pilot to firm-wide rollout. The firms doing this describe it as a way to generate proof, and managing the politics is a side effect. It gives the committee something better than an opinion to point to when the next request comes in, which is the actual adoption data from the last one. The structure also solves a problem that a straight yes-or-no approval process cannot: it gives the committee a graceful way to say maybe. A pilot is a smaller commitment than a firm-wide license, which makes it politically easier to approve even for a request the committee is genuinely unsure about, and the data it produces removes the guesswork from the larger decision that follows.
Access is starting to follow accountability, not volume
The clearest shift in this group is firms tying tool access to measurable outcomes, whether billable hours, case volume, or a documented efficiency gain. Some now counsel attorneys who request a tool and then never use it, with the license revoked as the consequence. That is a real move from who asked loudest to who delivered, and the firms that make it early, while it is still a process question and not yet a named partner's pet project, are the ones that avoid the shelfware pile in the first place. Revoking a license is a harder conversation than granting one, which is exactly why so few firms do it consistently. A policy that only ever adds access and never removes it is not really an accountability system. It is a one-way valve that looks like governance until someone finally audits the actual usage numbers behind it.
Shadow tools are turning shelfware into a wider problem than a bad purchase
The shelfware story above gets more complicated once shadow tool use enters the picture. Several firms now report buying a licensed assistant for one group, attorneys most often, and finding that administrative staff turned out to be the heavier users, while the people the license was bought for kept reaching for a personal AI subscription instead. That means a firm can run its funding process correctly, get a committee sign-off, buy the seat count it modeled, and still end up with the exact shelfware problem the process was built to prevent, because the approval covered the tool rather than the actual behavior of the people it was meant for. Firms naming this are starting to ask for something governance has not built yet: a single environment that consolidates the several tools already in informal use, instead of one more formally approved license sitting alongside them.
Legal operations already expects this to get worse before it gets better
The Blickstein Group's 18th annual legal operations survey found that nearly three-quarters of legal operations professionals surveyed expect ungoverned, unsanctioned AI use inside their organizations to become a growing problem, not a shrinking one. That expectation lines up with what our members describe: formal committees and approval processes exist, and people are still reaching for tools outside them. The honest reading of both data points together is that governance built only around the request-and-approve cycle is solving for the visible half of the problem while the harder half, use nobody asked permission for, keeps growing underneath it.
Source: Law360, In-House Legal Operations' Stubborn 'Shadow AI' Problem
What to do with this
This week, pull usage data on every AI tool the firm is currently paying for. Before approving any new request, show the requestor the actual adoption rate on the last tool their group asked for. Several members in this group discovered the shelfware problem only after the next purchase was already on the table, which is the worst possible time to learn it. Make the data part of the approval conversation, not an afterthought, and while pulling it, ask the harder question the data will not answer on its own: what tools are people already using that never went through the request process at all. The committee can only govern what it can see. Pair the usage pull with one more question and the exercise stops being an audit and starts being a policy. Ask whoever has the highest and lowest adoption numbers what is different about how they use the tool day to day. That answer, more than any dashboard, is what tells the committee whether the next purchase is likely to become shelfware or a genuine habit.
Members get more. The breakdown by firm size, practice area, and tech stack. A new question every Tuesday, the outlier answers that cut against the pattern, and most weeks an expert take with one clear next step.
Frequently asked questions
- How do law firms decide which AI tool requests to approve?
- About 40% now route requests through a committee or a formal approval process, but 44% still have no formal way to turn down a senior partner. On plenty of weeks, whoever asks the loudest still wins regardless of the process written down on paper.
- What exactly counts as AI shelfware at a mid-size law firm?
- A tool purchased under pressure, often from a senior partner, that a handful of people use and everyone else ignores. Members describe it as a governance failure rather than a technology problem, and the direct cost of letting the loudest request set the budget without checking prior adoption first.
- Does a formal AI approval committee stop unauthorized tool use?
- Not on its own. A committee can only govern the requests that come through it. Legal operations professionals broadly expect ungoverned AI use to keep growing, which tracks with what mid-size firms report: formal process and unsanctioned use are both rising in these firms at the same time.
- How can a law firm reduce AI shelfware?
- Tie license access to measured outcomes rather than who asked for the tool. Firms doing this well pull usage data before approving the next request, pilot with one representative per practice area before a firm-wide rollout, and revoke access when a requested tool goes unused for months.