Most AI adoption advice for professional services firms starts with Everett Rogers’s diffusion of innovations curve. Innovators, early adopters, early majority, late majority, laggards. A 2.5% / 13.5% / 34% / 34% / 16% split. The framework is 60 years old and holds up remarkably well.
While the framework is useful, it’s primarily descriptive. It tells you what categories of people exist, but it doesn’t tell you what to do with them.
A managing partner trying to figure out how to roll AI out across the firm next quarter doesn’t need to know which Rogers’s category each partner falls into. What matters is this: Are they going to figure out where AI helps on their own, or are they waiting for someone to tell them?
In practice, that’s a two-bucket question: Are they experimenters or are they taskers?
Define the two
An experimenter picks up the tool and starts trying things. They don’t need a use case sold to them. In fact, they would rather discover the use cases themselves. In Rogers’s terms, they look a lot like innovators and early adopters, roughly the top 16% of any firm. In legal, that often means a mid-level partner who gravitated toward technology early, or an associate who started using ChatGPT in law school and never stopped. They generate the firm’s actual use cases by trying things and seeing what works.
A tasker is not opposed to AI. They are not a laggard. They are perfectly willing to use the tool when someone hands them a clear pattern. What they do not do is invent the pattern themselves. Using Rogers’s categories, that is roughly the early and late majority, about 68% of the firm. This includes most senior partners, most middle-tier associates, and most operations staff. They are not the problem. They are most of the firm.
If you treat the whole firm like experimenters by giving them an AI tool and telling them to go figure it out, the experimenters will love it, and the taskers will quietly stop using the tool within three weeks.
If you treat the whole firm like taskers by waiting until you have a perfect prompt library before turning anything on, the experimenters will get frustrated and use ChatGPT on their phones anyway. You will lose the use cases they would have discovered for you.
The right play is sequenced: experimenters discover the patterns, taskers scale them, and a structure connects the two.
The data that should worry every firm
McKinsey’s 2025 State of AI survey found that 88% of organizations use AI in at least one function. Only about 6% are high performers, generating a measurable impact on EBIT from AI. The other 82% have the tool and not the outcome.
The high performers share one habit. They redesign workflows. Deloitte’s 2026 enterprise survey found that 84% of organizations have not redesigned roles or workflows around AI. If workflow redesign is where value gets uncovered, that gap may help explain why so many organizations struggle to move beyond incremental gains.
BCG’s number is the one I keep coming back to. The leaders in AI adoption allocate 70% of their budget to people and processes, 20% to tech and data, and 10% to algorithms. Most firms invert that ratio. They buy the tool, plug it in, and hope.
The whole experimenter and tasker framing exists because that 70% is where the work actually lives. The bottleneck is not the tool. The bottleneck is moving successful patterns from a handful of experimenters to the majority of the firm.
What the experimenters need from you
Three things.
A safe sandbox. Pick a few approved AI tools, run them through security and confidentiality review, then leave the experimenters alone. Get the data leakage and ethical wall questions answered upfront so partners don’t browse AI on their personal devices to avoid IT.
Second, permission to use non-billable time. Most experimenters’ wins appear first in non-billable work, whether that is drafting an internal memo, summarizing a meeting, or performing another administrative task. The advantage is incentive design. When the first use cases live on non-billable activity, no one has to argue with the comp committee about realization rates while they are still figuring out what works.
Third, a way to surface what they find. That is the Center of Excellence. More on it in a minute.
Most firms get the first two right and skip the third. That is how you end up with three partners who are AI experts and a hundred people who never got the benefit.
What the taskers need is a Center of Excellence
The phrase Center of Excellence (CoE) carries some baggage. It can mean nothing. It can also mean a real operational function. The distinction matters.
A real AI CoE in a law firm does five things.
- Use case prioritization. Of the things the experimenters found, which ones are worth standardizing? Macfarlanes didn’t try to scale everything at once. The firm rolled out Harvey practice area by practice area until 80% of their lawyers were on Harvey regularly. That is what disciplined prioritization looks like.
- Prompt and workflow standardization. Once a use case is chosen, build the prompts, templates, and integration points so a tasker can use them without ever having to be creative with AI. If Copilot in Word proves effective for redlining indemnity clauses, the CoE can turn it into a prompt template that any associate can run.
- Training in the flow of work. Not a one-hour seminar. Specific 5-minute trainings tied to specific workflows the tasker already does. “Here is how to use Copilot when you are drafting your engagement letter.”
- Governance. Confidentiality, hallucination checks, ethical walls, audit trails. About half of organizations with AI in use have reported an AI-related incident in the last year, per McKinsey. The CoE owns the rules and the human-in-the-loop checkpoints.
- Vendor evaluation. What to buy, what to build, what to wait out. The vendor landscape changes every six months. Someone needs to own that decision instead of letting every practice group buy its own toy.
The most common CoE failure mode is treating it as an innovation lab. Innovation labs ship demos. A CoE ships adoption. The output of an innovation lab is a slide deck. The output of a CoE is a junior associate in their second week using AI for clause comparison without ever having been formally trained on it.
The embedded AI shortcut
One of the things changing adoption is AI moving into the software that lawyers already use. Microsoft Copilot is inside Outlook, Word, Teams, and Excel. NetDocuments includes ndMAX within its document management system. Curo 365 is built on Microsoft Dynamics and Filevine has its own AI layer. Filevine’s LOIS sits on both their database and on Word.
I have seen firms try to roll out a brand-new AI app to lawyers as if it were a separate skill to learn. Adoption almost always stalls. I’ve seen the same firms turn on Copilot inside Outlook, and within a week, most of the firm is using it for email replies because it’s already there.
The practical implication is that some of the tasker enablement work is already getting done by your software vendors. The CoE’s job is to know which embedded AI features are good enough to standardize on, and where you still need to layer on prompt design.
Most mid-market firms should start with the AI capabilities already embedded in their existing stack before adding new tools. For example, start with Microsoft Copilot in Dynamics 365, expand into the AI capabilities built into the practice management platform, and add specialized tools where additional depth is needed.
Why this is harder than it sounds
Change management is the real work and it’s unglamorous.
A few patterns I have seen firms wrestle with.
Partner skepticism is real, and you cannot train your way through it. The fastest way to convert a skeptical partner is to put them next to an experimenter partner they respect and let them watch. Social proof beats slide decks.
The billable hour creates a separate challenge. If a partner uses AI to do in 30 minutes what used to take 5 billable hours, the firm has to figure out how that partner gets paid. If the answer is “less,” you will see less of that behavior. This belongs in a broader conversation about legal pricing and incentives, which I explore in more detail here. For now, keep experimentation on non-billable time.
Junior associate identity is the hardest people problem. The first-pass research and drafting that defines a young lawyer’s value is exactly what AI is best at. Chris Tart-Roberts at Macfarlanes has framed the honest reframe well: the next generation of lawyers will see AI as a core part of their practice from day one. Think of it the way an associate in 1995 thought about Westlaw. New tool, fast learning curve, same job at a different layer.
What firm leaders should actually decide
The question is not “how do we adopt AI?” The question is “which experimenters in our firm are already finding the wins, and what is the structure that connects them to the rest of the firm?”
If you cannot name the experimenters, find them this week. They are already there. They are using AI on their own time.
If you do not have a CoE or something that functions like one, stand it up. It does not need a building or a budget line. It needs an owner with time, governance authority, and a mandate to standardize. One senior person with two or three smart operators is enough to start.
If you are buying new AI tools before you have ridden the embedded ones for everything they can do, slow down. The taskers do not need more apps. They need more value out of the ones they already use.
The compressed version
Most firms are losing AI adoption because they treat everyone in the firm like an early adopter. They are not. About 16% are. The other 68% are taskers who will use AI when someone hands them the pattern and will not use it otherwise.
The play is sequenced. Let the experimenters discover the use cases on non-billable work. Build a Center of Excellence to standardize what they find. Use the embedded AI in your existing software as the delivery layer. Fix the incentives so the wins do not get hoarded.
In sum, AI adoption in a law firm is not a training problem. It is a sequencing problem. Get the order right, and the rest is logistics.
Need help implementing AI in your law firm?
Reach out to our legal advisory experts at any time to assist with your firm’s AI adoption efforts!
This publication contains general information only and Sikich is not, by means of this publication, rendering accounting, business, financial, investment, legal, tax, or any other professional advice or services. This publication is not a substitute for such professional advice or services, nor should you use it as a basis for any decision, action or omission that may affect you or your business. Before making any decision, taking any action or omitting an action that may affect you or your business, you should consult a qualified professional advisor. In addition, this publication may contain certain content generated by an artificial intelligence (AI) language model. You acknowledge that Sikich shall not be responsible for any loss sustained by you or any person who relies on this publication.
