A few years back, a solo attorney's first move on a new matter was to open a spreadsheet, type in the prospective client's name and every adverse party, and cross-check against a running list. Twenty minutes on a good day. An hour if the parties were entities buried inside entities. Now that same attorney is being pitched software that promises to run the check in seconds, pull structured party data straight from intake forms, and flag hits across ten years of matter history before the first phone call ends.

That sales pitch is where a lot of small firms are starting with AI. It's the wrong place to start. The conflict check is one of the few tasks in a small practice where the lawyer's judgment is the whole point, and it's also one of the least painful bottlenecks in the day.

If you're deciding where AI goes first, the honest question isn't which workflow looks most futuristic. It's which workflow is costing you the most time and carries the least risk to hand off.

Why Does Everyone Reach for the Conflict Check First?

The conflict check gets picked because it looks like a database problem. Names in, hits out. It feels like the kind of thing a computer should have solved a decade ago, and vendors have built polished demos around exactly that framing.

But a real conflict analysis isn't a lookup. It's a judgment call about whether representing this client will materially limit your work for another, and about whether informed written consent can cure the problem. The lookup is the easy part.

Smokeball's practitioner guidance on conflict checks walks through the two categories of concurrent conflict the ABA recognizes and the moments during representation when a check needs to run: before consultation, before engagement, and again when new parties surface. None of that is automation-ready in the way vendors imply. The list of names is the input. The decision is still yours.

There's a second reason firms reach here first. It feels low-stakes. If the software misses a hit, the lawyer will catch it. That reasoning is backwards. The workflows that benefit most from AI are the ones where a lawyer will still review every output anyway, but where the underlying task eats hours a lawyer doesn't want to bill.

So Where Should AI Actually Go First?

Start with the workflows that are pure friction: high-volume, low-judgment, and already reviewed by a human before anything goes out the door. In a small firm, that usually means a short list:

• Intake triage. Sorting inbound leads, pulling contact details into the matter management system, and drafting the first response email. A person still reviews the email; the software removes the typing.
• Document summarization. Condensing a 400-page production, a deposition transcript, or a client's stack of emails into a working brief the attorney reads before the meeting.
• Discovery review. Tagging responsiveness and privilege on large document sets, with the attorney spot-checking the model's calls and owning the final designations.
• First-draft correspondence. Status updates to clients, scheduling replies, engagement letter shells. The lawyer edits and signs; the machine writes the boilerplate.
• Deadline and docket entry. Reading a court order and proposing calendar entries with the citations attached, so the paralegal is confirming rather than transcribing.

None of these are glamorous. All of them are where the hours actually leak. IntelligentHQ's piece on what what AI changes for a small firm without replacing the lawyer changes for a small firm without replacing the lawyer makes the same point from a different angle: the tools earn their keep on the busywork that keeps you at your desk after six, not on the tasks that define your professional judgment.

What Should Stay With the Lawyer, on Purpose?

Some workflows should stay manual even after the firm is comfortable with AI everywhere else. Not because software can't touch them, but because the client relationship or the ethical duty requires a person at the center.

• The conflict judgment. Let software surface the hits. Do not let it clear the matter. The decision to proceed, decline, or seek consent is the lawyer's.
• The first substantive conversation. Clients hire a person. A chatbot that qualifies a lead is fine; a chatbot that gives legal advice on the intake page is a problem.
• Fee and scope discussions. Engagement terms, fee agreements, and any conversation about limiting the scope belong to the attorney and get memorialized in writing.
• The signature on every filing. The attorney owns every citation, every factual assertion, and every argument, whether the first draft came from an associate or a model.

The ethical duty of competence already tells lawyers to stay current on the benefits and risks of the technology they use. Reading that as "you have to adopt AI" misses the second half. You also have to understand where it fails, and design your workflow so those failures never reach the client or the court.

Start with the friction. Leave the judgment alone. The conflict check can wait.