Conflict checks at the intake stage: the workflow most law firms get wrong

By the time most law firms run a conflict check on a new intake, the AI voice agent has already captured four minutes of substantive intake on a matter where the firm may already represent the adverse party. The conflict-check problem at intake is not whether to do it. It is when, and against what, and how fast. Most firms have a workflow that answers all three questions wrong by default.

This post is the practical workflow piece. Where conflict checks should fire in the intake call, how the AI talks to the firm's conflict database, what to do about corporate parent and subsidiary relationships, and why fuzzy matching is the field that decides whether the check is useful or theater.

Why conflict checks are the malpractice claim that does not make headlines

The ABA Standing Committee on Lawyers' Professional Liability and the major legal-malpractice carriers consistently report that conflict-related claims account for somewhere in the range of 12% to 18% of malpractice claims, depending on firm size and practice mix. They are rarely the highest-severity claim category by total payout, but they are reliably in the top five frequency categories. The headline-grabbing claims are big-firm transactional misses. The claims that actually drive carrier loss ratios are smaller firms that took on a matter without realizing they had a conflict, learned the conflict three weeks into the engagement, and either had to withdraw or settle malpractice exposure for the disruption.

Conflict misses at intake are nearly always preventable. They get missed because the intake workflow runs faster than the firm's check process. AI voice agents accelerate this gap unless they are integrated into the check process directly.

The current state at most firms

The standard intake workflow at most US law firms in 2026 looks like:

1. Caller dials the firm.
2. Front desk or receptionist (human or AI) takes basic information.
3. Caller's information is logged into the firm's CRM or intake system.
4. The next morning, an intake coordinator or paralegal reviews the previous day's calls and runs each name through the firm's conflict database.
5. If a conflict is identified, the firm calls back the prospective client to decline the matter.

This workflow has three specific problems for AI voice intake:

• It runs the conflict check after the firm has already received potentially significantly-harmful information, triggering Rule 1.18(c) exposure (covered in detail here).
• It runs the check after the prospective client has formed an expectation of representation. Calling back to decline 14 hours later is a customer-service failure that the prospective client remembers.
• It loses information value. The intake coordinator running the check has access to the call summary but not to the texture of the call. Marginal conflict calls (where the matter is "substantially related" to a prior representation) cannot be evaluated cleanly from a transcript alone.

Most firms accept this workflow because the alternative seems technically impossible. It is not technically impossible. It is technically nontrivial. The difference matters.

What a real-time conflict-check workflow actually looks like

The cleanest pattern, in our experience, is a four-stage intake sequence:

Stage 1: Identification (15 to 30 seconds). The AI captures the caller's name, contact information, the type of legal matter, and a brief description in the caller's words. No factual disclosure about specific allegations, settlement positions, or strategy. This stage is structured to gather just enough to identify the parties.

Stage 2: Opposing-party disclosure (10 to 15 seconds). The AI explicitly asks: "Before we go further, can you tell me the names of any other parties involved, including individuals, companies, or insurance carriers?" The caller responds. The AI logs the structured field.

Stage 3: Real-time conflict check (1 to 5 seconds). The AI sends the caller name and opposing-party names to the firm's conflict-check system via an API call. The conflict system runs fuzzy matching against the firm's prior-client and current-client list, and corporate-relationship tables (parent, subsidiary, related entities). Returns: clean, conflict, or ambiguous.

Stage 4: Branched continuation. If clean, the AI proceeds with substantive intake. If conflict, the AI politely concludes the call with a script that does not acknowledge the conflict directly ("we are not able to take this matter at this time and we recommend [referral]"). If ambiguous, the AI captures the rest of the matter description but flags the call for partner review before any commitment to representation.

The whole conflict-check sequence adds 25 to 50 seconds to the intake call. Most callers do not notice. The ones who do notice appreciate that the firm runs a structured process.

The hard part is not the AI

Building the AI voice agent side of this workflow is straightforward. It is a function call inserted into the intake state machine, with branching logic on the response. The vendor-side engineering is roughly a one-week effort.

The hard part is the conflict database itself. Three persistent problems:

• Fuzzy matching. "ABC Corp" vs "ABC Corporation" vs "A B C Corp." vs "ABC Inc." vs "ABC Holdings" all need to match against the same underlying company in the firm's prior-client list. Most firm conflict systems use exact-string match or prefix-match, which catches ABC Corp but misses ABC Holdings as a parent. Real fuzzy matching requires either commercial conflict-check software (Intapp, Aderant, LegalKEY) or a custom embedding-based match layer. Most small and mid-sized firms have neither.
• Corporate hierarchy. The firm represented Acme Corp on a real-estate deal three years ago. The caller wants to sue Acme Logistics, which is a wholly-owned subsidiary of Acme Holdings, which acquired Acme Corp last year. Whether this is a conflict depends on facts the firm's conflict database does not currently model. The big-firm conflict systems handle this. Most small-firm systems do not.
• Practice-area-specific normalization. In family law, opposing-party normalization often hinges on personal-name matching with nicknames, maiden names, and prior marriages. In corporate work, it hinges on corporate-genealogy normalization. In personal-injury, the conflict check often runs against the insurance carrier as much as the named defendant. A general-purpose conflict system rarely handles all three well.

The right way to think about AI voice intake conflict checks is that the AI is the trigger layer, the conflict database is the determination layer, and the integration between them is where almost all the engineering work lives. Vendor demos that show "we run a conflict check" without showing what they are checking against are not credible.

Practice-area exposure ranking

Not all practices have the same conflict-miss exposure. From highest to lowest, by combined frequency and severity:

1. M&A and complex commercial litigation. Highest exposure. Corporate hierarchy makes the check dependent on entity-relationship modeling that most firm systems do not have.
2. Family law. High exposure on personal-name fuzzy matching and on conflicts arising from prior representation of a former spouse.
3. Insurance defense. Specific exposure to carrier-side conflicts. Firm represents a carrier in unrelated matters; the caller is seeking to sue an insured of that same carrier.
4. Personal-injury plaintiff. Moderate exposure. Defendant identification is usually clean, but third-party defendants and insurance carriers are often unidentified at the intake stage.
5. Criminal defense. Lower frequency, but high severity on the specific case of co-defendant conflicts that emerge after intake.
6. Estate planning and basic transactional. Lowest exposure for solo practitioners. Higher for boutiques with corporate clients.

The ranking matters for vendor evaluation. A solo PI plaintiff firm has different conflict-system requirements than an M&A boutique. A vendor that pitches "we run a conflict check" without asking what practice area is selling theater.

Five questions to ask before your next vendor demo

The diligence pass for conflict-check workflow takes 20 minutes. The five questions:

1. At what point in the intake call does your AI run the conflict check? Anything later than first opposing-party disclosure is a Rule 1.18(c) exposure.
2. What does the check run against? "Our internal database" is not an answer. The acceptable answer names the firm-side system (CRM, conflict system, custom database) and the fields the check inspects.
3. How is fuzzy matching handled? Does the check handle corporate name variations, parent-subsidiary relationships, personal-name aliases? If the answer is "exact match," the system is missing somewhere between 30% and 50% of real conflicts in corporate-heavy practices.
4. What happens when the check returns ambiguous? The right answer involves capturing the rest of the call but flagging for human review before any representation commitment. The wrong answer is "we let the call complete and the firm decides in the morning."
5. What is the latency budget for the check? A real-time conflict check that takes 7 seconds is functionally impossible inside a voice conversation. Sub-2-second latency is required. Sub-1-second is the working target.

If a vendor cannot answer all five in writing, the intake workflow they are selling is unsafe for legal use. This is the diligence pass that separates serious vendors from theater.

The cost of getting this wrong

A single missed conflict that drives a malpractice claim in a contested matter routinely costs the firm $50,000 to $250,000 in defense costs and settlement, before reputational impact. The carriers report that the modal claim is not the catastrophic miss. It is the moderate miss that leads to firm withdrawal mid-engagement, billing reversal, and a malpractice claim from the client whose matter was disrupted.

The intake-stage conflict check is the cheapest failure prevention available in legal practice. It is also the one most often outsourced to a workflow that runs hours after the failure window has closed.

For the regulatory framework around conflict checks at intake, see ABA Model Rule 1.18 and AI voice intake. For the inbound-call regulatory companion, see TCPA and AI voice intake. To see how a real-time conflict-integration workflow runs in a working demo, book a call.