A new matter looks clean on the intake sheet. A name, a phone number, a two-line description of the problem. You book the consult, spend forty minutes learning the situation, and it's only when the prospect names the other side that something goes cold. You know that name. It's already on a file down the hall.
Now you're stuck. You can't sit on both sides of the same dispute, and you've already heard things you were better off not hearing. The consult was free. The cleanup isn't. You write the careful decline, you wall off what you learned, and you hope it never becomes a story anyone tells.
Here's the part that stings. The collision was visible the second the inquiry arrived. The name sat in your own client list the whole time. Nobody caught it because checking every new inquiry by hand, against every client and every matter and every opposing party you've ever logged, is exactly the chore that slips when the phone won't stop.
What a conflict check is really up against
A conflict check (making sure a new client doesn't clash with someone you already represent) isn't a single lookup. One inquiry can touch a dozen names, and most of them never appear on the intake form.
Think about who actually needs checking when a matter walks in the door:
- The prospect, under every version of their name. People go by nicknames, maiden names, and initials, and they rarely fill out forms the way your files spell them.
- The opposing party, who's often the real conflict and almost never on the first form you see.
- Related businesses, the LLC behind the person and the person behind the LLC.
- Spouses and family, especially in anything touching a home, an estate, or a closely held company.
- Old matters, including former clients whose files you closed years ago but whose interests still count.
Do that by memory and you'll catch the obvious ones. It's the near-miss that gets you, the name that's close enough to matter and different enough to skate past a tired reader at 5 PM.
Run the check the moment intake lands
The fix isn't more diligence from people who are already stretched. It's moving the check to the front, before anyone spends billable time.
A light automation (a small workflow that runs on its own when something happens) can watch your intake, whether that's a web form, a logged call, or a referral email. The instant a new inquiry arrives, it pulls the names out, compares them against your client and matter list, and looks for anything close.
Close is the key word. Good matching doesn't just find exact spellings. It catches "Bob Ferreira" against "Robert Ferreira," "Smith Builders LLC" against "Smith Brothers Construction," a married name against a maiden one. When it finds a possible hit, it drafts a flag: here's the new inquiry, here's who it looks like, here's the file it resembles. That flag lands in front of a person before the consult ever gets booked.
Nothing gets declined by a machine. The point is timing. You want the question raised while it's still cheap to answer.
Good at look-alikes, bad at judgment
This is the center of the whole thing, so let's be plain about it. AI is genuinely good at one half of a conflict check and genuinely bad at the other, and knowing which is which keeps you out of trouble.
What it's good at is surface. Sifting thousands of names, spotting the ones that rhyme, never getting bored on the four hundredth inquiry of the month. It won't skip a check because it's Friday. It doesn't care that a name looked familiar but probably isn't.
The thing it can't do is decide. Whether a look-alike is an actual conflict, whether an old closed matter still bars a new one, whether a distant business tie crosses an ethical line, that's judgment, and judgment is yours. The tool that tells you two names match has no idea whether representing one harms the other. It shouldn't pretend to.
So it will over-flag, and that's fine. Expect it to raise names that turn out to be strangers who happen to share a common surname. A flag a person clears in ten seconds is cheap. A conflict that surfaces a week into the matter is not. You're tuning the tool to be a little paranoid on purpose, and then you, the person who understands the rules, make every real call.
Keep the check itself quiet
A conflict check runs on names and relationships, not on the substance of anyone's problem. Build it that way. The automation needs to compare "who is this" against "who do we know," and it doesn't need the confidential details to do that job.
Keep the whole thing inside systems you already trust with client data, and don't route names through some random free tool to save an afternoon of setup. The flag a partner reads should live where your files live. Handled right, this tightens confidentiality rather than straining it, because a quiet automated check leaks less than an email to the whole office asking "does anyone remember a Ferreira matter?"
Worth doing this week
You don't need a big system to start. You need the check to happen every time, early. A few first steps:
- Write down where inquiries come in. Web form, phone log, referral email, the contact page. You can't check intake you can't see, so list every door first.
- Get your client and matter list into one searchable place. Names, opposing parties, related businesses, closed files. The check is only as good as the list it reads.
- Pick the names that matter. Decide which fields a new inquiry must always capture: prospect, other side, any business names. That's what the automation cross-checks.
- Name the human who clears flags. One person (or a clear backup) reads every flag and makes the call. The tool drafts; a partner decides.
- Run it in parallel for two weeks before you trust it. Keep doing whatever you do now, and watch what the check catches and what it misses. You'll learn how paranoid to set it.
Set up well, this runs in the background and only speaks up when it should. If you'd like help wiring it into the way your practice already takes in work, that's the sort of quiet install we do at AI Agency Mass. Either way, run the check before you book the consult. It's a lot easier to ask the question early than to unwind the answer late.