Your Conflict Check Searches a Database That Is Missing Half the Names

A firm I work with lost a case worth roughly $400,000 in fees to a conflict check. Not to a conflict. To the check.

The prospective client called on a Thursday afternoon. Intake was clean, the facts were good, and the managing partner wanted the file. But the person who ran conflicts was out that week, and clearing the name required manually sweeping three systems plus a folder of scanned intake sheets. The run did not finish until the following Tuesday. By then the client had signed with a firm down the road that had answered him on Friday morning.

There was no conflict. There was never going to be a conflict. The firm paid full acquisition cost for a case it lost to its own internal process.

Every partner reading this has some version of that story. Almost every one of them files it under "we should put another person on intake." That is the wrong diagnosis, and the right one is uncomfortable.

The conflict check is not a compliance task. It is an intake speed problem wearing a compliance costume, and it runs on top of a database that is structurally incomplete.

Two separate failures that firms treat as one

When conflicts go wrong, they go wrong in two completely different directions, and firms almost always conflate them.

The first failure is speed. The check takes too long, intake stalls, and a prospective client who is shopping signs somewhere else. This failure is visible. It shows up as a lost case, and it costs money.

The second failure is coverage. The check runs fast and comes back clean, but it searched a name universe that never contained the name that mattered. This failure is invisible until it is catastrophic. It shows up eighteen months later as a disqualification motion, a bar grievance, or a malpractice carrier asking questions.

Here is why treating these as one problem is dangerous: every fix that addresses speed alone makes coverage worse. Rush the office manager, cut the manual sweep, search fewer places, sign faster. That is exactly how a firm gets both fast and wrong.

You cannot solve one without solving the other. Any process, human or automated, that gets faster by looking at less is not a conflicts process. It is a formality with a timestamp on it.

The database you search is a record of what somebody typed

Ask a firm where their conflicts data lives and you will get a confident answer: it is in the CMS/CRM. Litify, Filevine, Clio, MyCase, Smart Advocate, Salesforce. There is a party table, it has names in it, you search the names.

Now list the categories of names that exist inside your firm but were never entered into that party table.

Adverse parties mentioned only in correspondence. An opposing carrier's adjuster emails your associate and copies a third party you have never heard of. That name is now in your firm's knowledge. It is not in your party table.

Corporate families, DBAs, and subsidiaries. You searched "Riverside Medical Group" and got nothing. The adverse party is "RVS Health Partners LLC," which is the same ownership under a different registration. Your search was technically correct and substantively useless.

Declined and dead matters. Someone called in 2023, sat through a forty minute intake call, disclosed confidential facts, and never signed. Under the rules in most states, that prospective client conversation can still disqualify you. Was it entered as a party record? In most firms, no. It exists as a call recording, an intake note, and a memory.

Name variants. Married and former names, transliterations, Anglicized first names, nicknames written on an intake sheet, and the four ways a client's own paperwork spells their surname.

Co-counsel and referral relationships. Fee-sharing arrangements and joint representations that lived their entire life inside an email thread between two partners.

Imputed conflicts from lateral hires. A new associate arrives carrying every matter she touched at her prior firm. That list, if it was ever collected, is a Word document from onboarding sitting in a shared drive.

Put it together and you get the rule that should worry you:

Your conflicts database is not a record of your firm's relationships. It is a record of the subset of relationships that somebody had time to type into a structured field.

The gap between those two things is your actual exposure. And nobody knows how big it is, because measuring it requires searching the places you are not searching.

The speed cost is larger than firms measure

The coverage problem is the one that ends careers. The speed problem is the one bleeding money right now, every month, quietly.

Run the numbers for a plaintiff firm with paid intake. Depending on practice area and market, fully loaded cost per signed case runs somewhere between $1,800 and $4,000 once you count media spend, intake labor, and the leads that never converted. In personal injury specifically, a prospective client is typically calling three or more firms before deciding. Speed to signature is the single biggest lever in that competition.

Say the firm runs 300 intakes a month. Say 6 percent of them stall in conflicts for more than 24 hours, which is a conservative number for any firm doing manual clearance. That is 18 files sitting in limbo. Lose a third of those to a faster competitor and you have surrendered 6 signed cases a month that you already paid to acquire.

At $1,800 acquisition cost, that is nearly $11,000 a month in spend you burned for nothing. That number does not include the fee value of the cases themselves, which is the part that actually hurts.

Defense and business firms have the same problem in a different currency. A new matter that sits three days in conflicts is three days of unbilled time on work you already won. The carrier assigned it, the file is open, and your associate cannot touch it. Panel counsel work is thin enough on margin without donating the first 72 hours.

Neither of these losses appears on any report. There is no line item called "revenue lost to conflicts latency," so it never gets managed.

What a complete conflict check actually requires

If you were designing the process from scratch, the name universe you search has five layers. Most firms search the first one and call it done.

Layer one: structured party records in the CMS/CRM. Clients, adverse parties, insurers, and related entities that someone entered into a field. This is what everybody searches.

Layer two: prospective and declined contacts in intake systems. Lead Docket, Go High Level, Send It By Text, or whatever sits in front of the CMS/CRM. These are the people who talked to you and did not sign, which is precisely the category that creates prospective client conflicts.

Layer three: unstructured communication. Email bodies, SMS threads, voicemail transcripts, intake call recordings, and the contents of uploaded PDFs. This is where most of the names actually live, and it is essentially unsearched at most firms because searching it manually is not realistic.

Layer four: entity resolution. Turning "Bob Danner," "Robert J. Danner," and "R. Danner" into one entity. Turning a company name into its corporate family. Matching a DOB and an address when the spelling is wrong. Without this layer, layers one through three return false negatives all day and you never know.

Layer five: imputed conflicts. Prior firm matters for every attorney and staff member, family relationships, board seats, and ownership interests. Static data, rarely updated, almost never searched automatically.

A firm that searches layer one in under a minute has a fast process and no coverage. A firm that searches all five layers by hand has coverage and a five day intake cycle. The manual version of this job cannot be both complete and fast, which is why it is always one or the other.

Where AI Fusion actually fits, and where it does not

This is the part where I want to be precise, because the wrong version of this is genuinely dangerous.

The AI does not clear the conflict. It assembles the record a human uses to clear the conflict. That distinction is not marketing hedge. It is the entire design constraint. Conflicts clearance is an ethical judgment that belongs to a lawyer, and any vendor telling you their system returns a clean yes or no is selling you something you should not buy.

What the AI Super Agent build does instead:

It runs continuously instead of on request. Every inbound email, text message, call transcript, and uploaded document gets scanned for entity names as it arrives and indexed against the matter it belongs to. The name universe stays current in the background. Nobody is reconstructing it under deadline pressure on a Thursday afternoon.

It expands the search terms before it searches. A new intake gives you a name, a DOB, an address, adverse parties, an employer, and an insurer. Each of those gets expanded into variants and corporate relationships before anything is queried. This is layer four doing its job.

It produces a conflicts memo in draft. Not a verdict. A document listing every potential hit, the layer it came from, the matter it relates to, the date, and a direct link to the source message or file. A human reads it and decides.

It flags ambiguity rather than resolving it. "Possible match: Robert Danner, adverse party in Danner v. Meridian, closed 2021. Same date of birth. Different address. Source: intake call recording, March 2021." That goes to a person. The system does not get a vote on whether it counts.

It writes the entire run to an audit trail. What was searched, which layers were queried, what came back, who reviewed it, when they approved it, and what they wrote. If a disqualification motion lands two years later, you can produce the actual record instead of an affidavit about your general practice.

The practical result: a job that consumed a day and a half of an office manager's fragmented attention becomes a memo sitting in the queue in under ten minutes, with a human signing off in five. Coverage goes up and cycle time goes down at the same time, which is the combination the manual process cannot produce.

And like everything else we deploy, it starts in draft mode. Every memo goes to a human for weeks before any part of the routing gets automated. The AI earns scope on the firm's timeline.

One honest caveat. Better search does not help a firm that does not want to find anything. If your conflicts process exists to generate a checkbox rather than an answer, surfacing more potential hits will feel like the system is broken. It is not broken. It is working, and the firm has a leadership problem that no software fixes.

A twenty minute test you can run this week

You do not need a vendor to find out whether you have this problem.

Test one. Pull three closed matters from 2022 or 2023. For each, write down every adverse party, insurer, corporate affiliate, expert, and material witness you can find by reading the file. Now search only your conflicts database for those names. Count the hits. The names that do not come back are names your firm knows and your conflict check cannot see.

Test two. Pull five intakes from last month that did not sign. Search for them as prospective clients. If they do not appear anywhere searchable, you have prospective client exposure with no record of it, and you will not discover that until someone adverse to one of them walks in the door.

Test three. Time your last twenty conflict checks from request to clearance. Not the average, the distribution. The average will look fine. The tail is where you are losing cases.

Five questions for any vendor selling conflicts automation

1. Does it search unstructured communication, or only structured party fields? If it is only reading the CMS/CRM party table, it is a faster version of the search you already do badly.

2. How does it handle entity resolution and corporate families, and can I inspect the match logic? "Proprietary" is not an acceptable answer for something that determines whether you get disqualified.

3. Does it produce a memo a human signs, or does it return a yes or no? Only one of those is defensible.

4. Can I produce the complete record of a specific conflicts run from eighteen months ago, including what was searched and who approved it?

5. What happens to a name that appears in exactly one voicemail transcript and nowhere else? The answer to this question tells you everything about how the system was actually built.

The firms that get this right are not the ones that bought the fastest conflicts tool. They are the ones that stopped treating the conflict check as a gate to get through and started treating it as a search problem over everything the firm knows. That is a different job, and it is not one a person with three other responsibilities can do in an afternoon.

---

See what your conflicts picture actually looks like

Book a working session and we will map the five layers against your current stack, show you where the names in your firm are living outside your CMS/CRM, and build the conflicts memo workflow around the systems you already run.

Book a demo or see how the AI Super Agent works.