A twenty-person firm can adopt AI faster than a two-thousand-person firm, and usually does not. The advantages are structural: no committee, no legacy platform, no procurement cycle, and a principal who can approve a decision on Tuesday and have it running on Wednesday. The obstacles are real but smaller than most small-firm leaders think, and the budget is smaller than they fear.
I spend time on both sides of this. The national firms have resources and governance. The small firms have speed, and speed is worth more right now than either of them realizes.
Your actual advantages
You can decide. A large firm needs IT, legal, risk, a practice-area committee, and a pilot governance structure before anyone opens a tool. You need one conversation. That difference is measured in quarters.
You have no platform inertia. No enterprise agreement you signed in 2019 that everything must integrate with, no internal system somebody built that would have to be retired first.
Your workflows are visible. In a twenty-person firm, one principal genuinely knows where the hours go. In a large firm that information is spread across offices and nobody has the whole picture. Selecting the right pilot is largely an information problem, and you have the information.
Adoption is a conversation, not a program. Twenty people is a room. You can train everyone in a morning, notice within a week who is actually using it, and fix problems by talking to a person rather than issuing guidance.
Your actual constraints, and what they mean
No IT function. Real, and it means you should categorically avoid anything requiring integration work. Choose tools that work out of the box with the systems you already have. This one constraint should drive your entire selection.
Nobody has spare hours. Also real. Everyone is billable, so the internal time to run this is genuinely expensive. It argues for a narrow pilot rather than a broad one, and for choosing a workflow whose owner is already suffering.
You cannot absorb a bad quarter. A national firm can write off a failed pilot. You cannot, which is why the sequencing below is deliberately cheap-to-fail rather than ambitious.
Notice that none of these is about capability or budget. The tools cost the same for you as for anyone, and the per-seat price is genuinely modest at your headcount.
A plan that fits
Month one: use what you already pay for, and watch
Your document suite or project platform almost certainly includes AI features now. Turn them on. Cost: nothing incremental.
Then pay attention for four weeks to what people actually reach for when the barrier is zero. This is the single most useful piece of pilot-selection data available to you, it is free, and it usually contradicts what leadership assumed. In small firms the answer is frequently proposals or report narrative rather than anything technical.
Month two: baseline two workflows
Before you buy anything, measure. Pick the two workflows that showed up in month one and capture hours per instance from timesheets, calendar days from request to delivered, and how often the work comes back needing substantive correction.
Two weeks of light effort, and it converts every later claim from an opinion into a number. Skipping it is the most common reason a pilot cannot be defended afterward, and it is unrecoverable once the workflow changes. The full version is the four numbers that convince a partner group, and it applies at twenty people as much as at two thousand.
Month three: buy one thing, for everyone
One general-purpose assistant, enterprise terms, whole firm. At your headcount the annual cost is roughly one modest software line item, and breadth beats depth at this stage because you do not yet know where the value concentrates.
Enterprise terms are not optional if client material will ever touch it. Consumer tiers typically permit the provider to use your inputs, which conflicts with the confidentiality language in your agreements. This is the one place not to economize.
Month four: train properly, then leave it alone
A morning. Show three real examples on your own closed-out projects, then run an error hunt where people find planted mistakes in AI-generated passages from your own document types. People who catch a fabricated citation once never trust one again, and that single exercise does more than any amount of guidance. The patterns to plant are in what AI gets wrong in construction documents.
Add four lines to whatever review checklist you already use: citations verified, numbers traced to source, project-specific deviations confirmed present, firm voice checked. Then stop adding process.
Month five and six: measure again, decide once
Re-measure the same two workflows. If the numbers moved, you have a result and can consider one domain tool for whichever workflow is now clearly worth it. If they did not, you have learned something cheap and you stop, which is a legitimate outcome.
What small firms should skip entirely
Building anything. You have no software function and no bench. An internal tool becomes one person's permanent obligation and a dependency you cannot staff. The full argument is in build or buy, and it applies with extra force at your size.
Multiple domain tools. Each one is a separate subscription, training burden, and vendor relationship. One at most, and only after it is earned.
A six-page policy. You need one page: approved tools, what information may go in, what must be verified, who decides. See the four clauses that matter. A twenty-person firm that produces a formal governance document has spent effort on the wrong thing.
Waiting for the market to settle. It will not settle on a timeline useful to you, and the compounding advantage is in your people learning to work this way rather than in picking the eventual winner.
The competitive reality
The firms you compete against for work in the ten to fifty person range are mostly not doing any of this yet. That will not last, and the gap that matters is not who bought which tool. It is which firms spent a year building the habit of verifying, measuring, and adjusting.
There is also a recruiting dimension worth naming. Early-career engineers ask about this now, and a small firm that can describe how it works this way has an argument against a larger firm's salary. A small firm that has banned the tools has handed those candidates a reason to leave.
Your advantage is that you can decide on Tuesday. Most of the firms your size are still waiting for someone to tell them it is safe.
If you only do one thing
Turn on what you already pay for, and watch for a month. It costs nothing, it requires no decision, and it will tell you more about where AI helps your firm than any article, including this one.
Then spend two weeks measuring the workflow it points you at, before you buy anything at all.
If you want a compressed version of this run with you rather than by you, our readiness engagement is scoped to fit smaller practices and ends with a phased plan and a budget rather than a recommendation to buy something. Start a conversation.
← Back to insights