Workflow Automation: Conflicts

The Matter

The team's process map separates conflicts and new matter intake into 9 recurring steps, and decomposition divides them into language, source-backed retrieval, legal judgment, and deterministic configuration, with different evidence, failure states, controls, and accountable performers attached to each one. Each row below carries the condition that decides when a step clears on its own and when it routes to a person.

A model that drafts well is an asset at summarizing and a hazard at family expansion for the same reason: fluency is the deliverable in one place and a mask in the other. Sorting the nine steps by the kind of work each performs, with the evidence, failure states, and accountable performer that kind carries, settles where a model belongs before model quality enters the conversation.

The Proposal

Steps 6 and 7 stay with the attorney: the conflict rules place adversity and clearance with a licensed person, and ABA Formal Opinion 512 holds supervision there when AI enters the workflow.¹ An attorney at those two steps lets everything upstream clear on evidence rather than wait on review: the volume lives at steps 1 through 5, 8, and 9, where source-linked work moves without a person in the path; the consequence lives at 6 and 7, where one person decides with the assembled record in front of them.

The Matter

The process map shows a request moving from intake to search, review, and decision. The case moves differently. A misspelled party name enters the form, the database returns nothing, the reviewer leaves the system for an old matter note, the note leads to an attorney who remembers why the relationship received different treatment 4 years ago, and the answer settles in an email thread because the form never learned how to ask the question that resolves it.

The official workflow still shows one clean path, while the exception queue, where requests land when the clean path cannot finish them, preserves what actually happened. A missing field sends someone back to the requester, ambiguous authority leaves the matter waiting for a partner, and local knowledge appears only when the person carrying it happens to be at their desk.

The Proposal

Following one retyped field through the people who chased it is the build specification: each detour names what has to survive the handoff, where the exception travels, when the system stops, and whose authority resolves the question. The answer then needs a route back into the record, so the next matter does not depend on the same accidental memory. Iteration and recurrence prevention sit at the nucleus of innovation-team work.

The Matter

Once the working process becomes visible, most of the engineering weight moves toward retrieval. A model can extract, compare, summarize, and draft while remaining unable to make a stale record current, prove an unrecorded relationship, or resolve a contradiction the sources leave open.

The case requires one query to cross public registries, commercial hierarchy data, confidential matter records, screened files, and stale caches before the reviewer sees one name.

Before the candidate set is assembled (the list of names the search will run), the records stay in separate piles: client and prospective-client information, attorney notes, screened matters, declined matters, public registry data, and commercial hierarchy data. The task decides which piles the model may read, because retrieval from one undifferentiated pool turns a search into a disclosure.

The sources themselves move at different cadences and support different claims. SEC EDGAR (Electronic Data Gathering, Analysis, and Retrieval) updates in real time as filings disseminate, with delays under 1 second for submissions and under 1 minute for XBRL data (the structured data inside filings).²

GLEIF (Global Legal Entity Identifier Foundation) publishes its consolidated master files, the Golden Copy, 3 times daily, Delaware presents entity information current as of the search and prohibits automated tools on its public search, and OpenCorporates directs users to correct the official record first and allow 30 days for its own to update.^{3 4 5}

The Proposal

When every source in the defined search plan answers, the system reports a covered zero. If a source fails, times out, or cannot be queried under its terms, the same blank screen represents a degraded zero. Neither condition proves the absence of a conflict unless the reviewer can see what ran, what answered, and what remains unresolved.

Conceptual interface states: what a reviewer can verify behind each kind of empty report.

The Matter

Retrieval runs on an index: one searchable body of records the system reads before it answers. In this case the index reaches beyond public company data into client names, prospective-client disclosures, matter descriptions, attorney impressions, waiver histories, screened relationships, and the fragments from which the firm will later reconstruct what it knew. Three controls follow: what must disappear before the search runs, what the search may read, and what may leave the firm only with counsel's approval.

The Proposal

Screened matters must disappear before retrieval assembles the candidate set because hiding a summary after generation leaves the protected content inside the model's context, whereas a properly timed access decision excludes the record and still leaves audit evidence that the control ran.

While a generated letter remains in a review queue, a lawyer can decide whether it should communicate acceptance, request a waiver, impose conditions, or decline the representation. Once the system sends it, the recipient may reasonably read the same language as the firm's decision, and the workflow has exercised the authority it was supposed to preserve for counsel.

The Matter

An associate's draft arrives carrying a maker: a name, a role, and a history of edits someone can ask about. An AI generated answer can enter the file with none of that, polished enough to read as finished before anyone reconstructs what produced it, and that missing trail is the gap supervision has to close. In the expanded family list, a name and a jurisdiction with nothing beside them and no provenance flag the whole set as weak.

The Proposal

The name list must carry, per name: source class, source date, relationship type, retrieval status, unresolved discrepancy, and any model contribution, visible with the name rather than resting in an audit log no one consults during clearance.

A matter summary needs the same route back: the authorized text it condenses, the person who accepts or corrects it, and the matter file that carries the conclusion into the next request.

The same strip follows the name; nothing about it is recomputed or lost between stations.

The Matter

A known risk in AI adoption is the separate interface: it adds effort before the work can begin and gives reviewers another reason to return to the path they already know. The client/conflict request already arrives through email, a form, and an existing matter system.

In the ABA's 2024 technology survey, 67 percent of lawyers report technology training available at their firm, under half agree the training they received was adequate, and 47 percent say technology problems sometimes cut into their productivity.⁶ A tool that opens as one more destination bills its learning curve against hours the reviewer already owes to matters.

The Proposal

A conflicts design that places extracted parties, proposed variants, source-backed relationships, and coverage failures inside the same systems asks the reviewer to learn nothing before the work begins. The vendors made the same bet: Microsoft's Copilot sits inside Word and Outlook, Anthropic ships a Claude add-in for Word in beta with native tracked changes, and OpenAI's models run inside Microsoft 365 Copilot rather than in a separate window.^{7 8}

The team still needs training on the tool's error profile, data handling, stopping points, and escalation paths; the product team carries the cost of reaching the work.

The Matter

Because aggregate activity cannot say whether a tool works, the measurement plan follows repeat usage per eligible reviewer, counting each person whose queue put the tool in their path, and the matters on which reviewers return.

Following each matter from request to decision brings the useful measures into view: the share of extracted parties that survive without correction, the distribution of clearance time, the volume and age of the exception queue, the share of reports with complete source coverage, and the rate at which reviewers override proposed relationships or summaries.

The Proposal

Instrumentation (usage logging) that stops at the click can show that someone opened the tool while leaving the firm unable to tell whether the candidate set is complete, the exception reached the right lawyer, or the final decision improved.

In its 2026 AI in Professional Services Report, Thomson Reuters Institute reported:

Use is context, not proof of workflow adoption.

More than 1,500 respondents across 27 countries. Source: Thomson Reuters Institute, 2026 AI in Professional Services Report, February 9, 2026.⁹

Those figures describe the industry; the case still has to measure its own matters, and the matters show where the method stops.

This workflow has three limits:

1) The records limit:

Matter descriptions that say "general corporate advice," affiliate records that omit relationships, waiver documents that remain untagged, and outside counsel requirements stored in disconnected systems bring the model to the edge of the available record. No model can retrieve what no one recorded. The build here starts with ownership of each source, repaired mapping between records, and a narrow statement of what the system may claim.

2) The one person bottleneck:

The request that survives intake can still wait on one partner, and when one person's availability controls the queue, faster extraction and cleaner summaries deliver more matters to the same unavailable decision maker while the product improves the waiting room. The build here is authority: who else can decide, who covers an absence, and when the question escalates.

3) The exceptions limit:

When most requests leave the clean path, routing beats generation: a layer that records the gap, finds the person who can resolve it, and returns the answer to the governed record is worth more than another drafting feature: it turns this matter's detour into the next matter's answer.

Commonly, a team already knows its evidentiary steps, traces what enters and leaves them, and builds the source, access, review, and escalation structure around the decision. The innovation team sits with them, atomizes the workflow, and with that knowledge automates the high-volume mechanical steps and optimizes the rest.

Corporate family expansion carries the sharpest retrieval risk in this case because fluent output can alter the factual set before the search begins.

Other legal workflows conceal the same problem inside a step that resembles writing even though the answer depends on evidence the model cannot create.