Conflict Checks at Law Office Intake: Structuring the AI Search and Attorney Decision Gate

Dave by voolama (hellodave.ai) is managed SaaS: voolama LLC operates and hosts the platform, and you install nothing and provision no servers. The conflict check workflow uses five node types from the ten available in the Workflow Editor. The article AI Workflows in Regulated Review: Can You Use AI in a Regulated Review Process? covers where AI fits in processes that require professional judgment. For conflict checks: AI lists possible name matches; a licensed attorney decides what those matches mean under ABA Model Rule 1.7 (current clients) and ABA Model Rule 1.9 (former clients).

The five nodes, in execution order:

1. Start (green, code: start): the entry point. Every workflow has exactly one Start node. Execution begins here and cannot begin anywhere else.
2. User Interaction (blue, code: user_interaction): pauses the workflow and creates a task in the Task Inbox. Intake staff enter party names and paste the firm's client list. Six fields cover the required intake data: prospective client full legal name; opposing party names; related entities such as businesses or trusts; matter type; and the firm's client list as plain text. Only Admin and Use roles can see and complete tasks in the Task Inbox.
3. Agent Interaction (purple, code: agent_interaction): calls the conflict-search agent automatically. The agent searches only the text supplied in the User Interaction task and lists possible name matches. No task is created; this node runs without human involvement. The agent does not decide whether a conflict exists.
4. Human Review (amber, code: human_review): creates a review task in the Task Inbox. The attorney reads the agent's match list in the Payload section and clicks Approve (no conflict, proceed), Reject (conflict found, decline), or Request Changes (more information needed). The workflow does not advance until the attorney submits a decision. The Comment field captures the attorney's reasoning and is stored in the audit log.
5. End (red, code: end): one for each outcome. Every path must end at an End node or the instance will hang at runtime. Three End nodes cover: Proceed with Intake (Approve), Matter Declined (Reject), and a return to the User Interaction node (Request Changes).

This workflow does not search court databases, bar records, or any external system. It does not perform waiver analysis under ABA Model Rule 1.7(b). It does not replace a purpose-built conflicts database. Those are separate tools and separate decisions.

Navigate to Workflows in the sidebar. You need the Admin or Curate role to see this section and open the Workflow Editor. Click New Workflow, name it "Client Intake Conflict Check", and save. The Workflow Editor opens with a three-panel layout: the Node Palette on the left, the canvas in the center, and the Properties Panel on the right.

Step 1: Start node. Drag a Start node from the Node Palette onto the canvas. Every workflow requires exactly one Start node. Execution begins here.

Step 2: User Interaction node. Drag a User Interaction node onto the canvas below the Start node. Label it "Collect Party Names and Client List". In the Properties Panel under Fields, configure: prospective client full legal name (text); opposing party names, one per line (text); related entities such as businesses or trusts (text); matter type such as litigation or transactional (text); and firm client list (text, large field for pasting the firm's current and former client records). Under User Assignment, assign this task to the intake staff member. Connect the Start node's source handle (bottom circle) to this node's target handle (top circle).

The firm's client list is supplied as plain text at the start of each run. If the firm's records are in a spreadsheet, the staff member copies the relevant columns (client name, matter name, opposing parties) and pastes them into this field before submitting the task. The agent searches only that text.

Step 3: Agent Interaction node. Drag an Agent Interaction node onto the canvas below the User Interaction node. Label it "Search Client List for Name Matches". In the Properties Panel under Agent Assignment, select the conflict-search agent. Connect the User Interaction node's source handle to this node's target handle. This node runs automatically: no task is created and no human action is required.

The conflict-search agent's system prompt should instruct it to: receive the prospective client name, opposing party names, related entities, and the firm's client list; search the client list for any name that is identical to or closely similar to any supplied party name; list every possible match with the exact text from the client list where the match appears; and write NO MATCHES FOUND if no possible matches are identified. Set temperature to 0.1 for consistent output. The agent does not decide whether a conflict exists.

Step 4: Human Review node. Drag a Human Review node onto the canvas below the Agent Interaction node. Label it "Attorney Reviews Conflict Report". In the Properties Panel under User Assignment, assign this task to the attorney responsible for conflict decisions. Connect the Agent Interaction node's source handle to this node's target handle.

When the workflow reaches this node, a task of type human_review appears in the attorney's Task Inbox. The attorney reads the Payload section (which contains the agent's match list alongside the original party names) and clicks one of three buttons: Approve (green), Reject (red), or Request Changes (outlined). The attorney should use the Comment field to record the reasoning behind any Reject or Request Changes decision. That comment is captured in the audit log.

Connect the Approve edge to an End node labeled "Proceed with Intake". Connect the Reject edge to an End node labeled "Matter Declined". Connect the Request Changes edge back to the User Interaction node so intake staff can supply additional information and re-run the search.

Step 5: Activate. Confirm every path leads to an End node using the node and edge counter in the top-right corner of the canvas. Click Save Version. In the Workflows list, set the workflow status to Active. A workflow in Draft status cannot be run. Run a test instance with hypothetical party names and a sample client list before using it for a real matter.

If a team member cannot see the Workflows or Tasks section in the sidebar, their role does not include the required permission. See Why Can't I See Agents, Workflows, or Other Sections in Dave by voolama? for how to diagnose and fix role assignments. For how Human Review tasks work in practice, see How to Review AI Output Without Becoming the Bottleneck.

Every task decision in Dave by voolama is recorded in the per-tenant audit log. For a law office running conflict checks, the audit log entry for each Human Review task captures: the event type (task completion), the actor (the attorney's email address and user ID), the entity (the task ID and the parent workflow instance ID), the timestamp, and the submitted result (the attorney's decision and any comment they added).

The audit log is at Admin, then Settings, then Audit Log. Only the Admin role can view it. The log supports filtering by event type, entity type, actor, and date range. The Admin can export filtered results as a CSV file for record-keeping or to support a bar association inquiry.

When SOC 2 or GDPR compliance mode is active in the workspace, task completion events are tagged for SOC 2 CC6.1 (logical access) and GDPR Article 30 (processing records). The default audit log retention is 365 days. The SOC 2 recommendation is at least 2,555 days (7 years). A warning is displayed in the audit log settings if retention is set below that threshold. The appropriate retention period for a law office depends on state bar rules and the firm's own record-keeping policy; this article does not advise on that question.

The article AI Agent Audit Trails and Approval Workflows: How DAVE Keeps Every Action Accountable covers the audit trail in more detail, including how every agent action is logged alongside human decisions in the same per-tenant record.

This article covers one specific use of Dave by voolama: the workflow build for structured conflict checks at client intake. It does not cover the following.

• Legal advice on conflicts rules. ABA Model Rule 1.7 and ABA Model Rule 1.9 are cited here as the source of the attorney's duty to perform conflict checks. This article does not interpret those rules, advise on their application to any specific matter, or address state-specific variations. A licensed attorney must make every conflict determination. State bar rules may impose additional or different requirements.
• Waiver analysis. ABA Model Rule 1.7(b) permits certain conflicts to be waived by informed client consent under specific conditions. This workflow does not address waiver analysis or consent documentation. That analysis requires attorney judgment and is outside the scope of this article.
• Searching external sources. The AI agent searches only the text supplied in the User Interaction task. It does not search court dockets, bar records, corporate registries, or any external system. If a firm's conflict check process requires searching external sources, those searches must be done separately.
• Dedicated conflicts database software. This workflow is a structured intake step for firms that currently run conflict checks manually. It does not replace a purpose-built conflicts database with automated matter ingestion.
• Connecting an AI provider. Before any Agent Interaction node can run, at least one AI provider must be configured in the workspace under Admin. Dave by voolama supports seven AI provider types: OpenAI, Anthropic, Azure OpenAI, Google, OpenAI-compatible, local (Ollama), and custom. This article assumes provider setup is already complete.

Disclosure: Dave by voolama is the product described in this article, published by its maker, voolama LLC. This is not legal advice. A licensed attorney must decide every conflict question.