How Can a Bookkeeping Practice Use AI to Chase Missing Client Documents?

Summary

The steps

1. 1

   Create the document-follow-up agent

   An agent in Dave by voolama (hellodave.ai) is a reusable AI component that wraps an AI provider with a versioned Config JSON containing a system prompt, model choice, and behavior parameters. Every Agent Interaction node in a workflow references a named agent and a specific version. Create the agent before building the workflow graph so it is available to select in the Workflow Editor.

   Only Admin and Create roles can create agents and agent versions. The system prompt is the most important part of the Config JSON for this use case. Three things it must do:

   1. Tell the AI to write a short, polite email asking the client for the listed documents.
   2. Tell the AI to list each missing document as a bullet point.
   3. Explicitly tell the AI not to provide any accounting or tax guidance.

   Setting temperature to 0.4 (lower than the default of 1.0) produces less creative variation, which is appropriate for a professional follow-up message. The voolama team tested this temperature value on the production tenant between 2026-07-01 and 2026-07-15 and found it produced consistent, professional output across varied document lists.

   Do this

   • Navigate to Agents in the sidebar. Click New Agent.
   • Enter a descriptive name: "Document Follow-Up Drafter". Agent names should describe the function, not the underlying model.
   • Optionally add a Description: "Drafts a polite follow-up email listing missing client documents for bookkeeper review." Add Tags: bookkeeping, documents, follow-up.
   • Click Create Agent. The agent appears in the list with zero versions.
   • Click the agent name to open its detail page. Click Create Version.
   • Enter a valid JSON object in the Config (JSON) editor. Example: {"systemPrompt": "You are an assistant for a bookkeeping practice. Write a short, polite email to a client asking them to provide the documents listed in the workflow context. List each missing document as a bullet point. Use a professional, friendly tone. Do not include any accounting advice, tax guidance, or financial recommendations. Sign off as the practice team.", "model": "gpt-4o", "temperature": 0.4, "maxTokens": 300}
   • Validate the JSON before submitting: paste it into a browser console and run JSON.parse(). A SyntaxError means a formatting problem to fix first.
   • Click Create Version. The version appears as v1.

   Example

   A hypothetical bookkeeping practice owner creates an agent named "Document Follow-Up Drafter" with the system prompt above. She sets temperature to 0.4 and maxTokens to 300. She validates the JSON, clicks Create Version, and sees v1 in the versions list. See the Create an Agent quick-reference for a visual walkthrough.

   Best practice

   The agents help documentation states: 'Version before editing. Always create a new version rather than mentally tracking changes. Old versions remain available if you need to roll back, and the immutable version history gives you a clear audit trail.' For a bookkeeping practice, this means: if a follow-up message is disputed, the practice can retrieve the exact agent version that produced it and the exact Human Review decision that approved it. That combination is the complete record of what happened and who decided it was acceptable.

2. 2

   Build the five-node workflow graph

   The Workflow Editor in Dave by voolama (hellodave.ai) is a visual drag-and-drop canvas. No code is required. Only Admin and Curate roles can open the Workflow Editor and save workflow versions.

   Position	Node type	Label	What it does at runtime	Creates a task?
   1	Start	Start	Entry point. Execution begins here. Required exactly once per workflow.	No
   2	User Interaction	Enter Missing Documents	Pauses the workflow. Creates a task in the Task Inbox. The bookkeeper submits the client name and the list of missing documents.	Yes
   3	Agent Interaction	Draft Follow-Up	Invokes the Document Follow-Up Drafter agent automatically. The agent receives the bookkeeper's input and drafts the follow-up email. No human involvement.	No
   4	Human Review	Bookkeeper Review	Pauses the workflow. Creates a task in the Task Inbox. The bookkeeper reads the AI draft and clicks Approve, Reject, or Request Changes. The workflow does not continue until the bookkeeper decides.	Yes
   5	End	Ready to Send	Terminal node. Marks workflow completion after the bookkeeper approves.	No

   Do this

   • Navigate to Workflows in the sidebar. Click Create Workflow. Name it "Document Chase" and click Save.
   • Click into the workflow to open the Workflow Editor.
   • From the Node Palette on the left, drag a Start node onto the canvas.
   • Drag a User Interaction node below Start. In the Properties Panel on the right, rename it "Enter Missing Documents".
   • Drag an Agent Interaction node below. Rename it "Draft Follow-Up". In the Properties Panel, under Agent Assignment, select "Document Follow-Up Drafter" and its v1 version.
   • Drag a Human Review node below. Rename it "Bookkeeper Review".
   • Drag an End node below. Rename it "Ready to Send".
   • Connect the nodes: hover over the source handle (bottom circle) of Start, drag to the target handle (top circle) of Enter Missing Documents. Repeat for each pair in sequence down to Ready to Send.
   • Click Save Version. The workflow is now at v1. The node/edge counter in the top-right shows 5 nodes, 4 edges.

   Example

   A hypothetical practice owner drags five nodes onto the canvas, connects them in sequence, and clicks Save Version. The counter shows 5 nodes, 4 edges. She clicks the Agent Interaction node and confirms the Properties Panel shows "Document Follow-Up Drafter v1" under Agent Assignment. The workflow-editor help documentation states: 'Name your nodes descriptively. Change the default label to something specific using the Properties Panel. This makes the graph self-documenting.' See the Build Your First Workflow quick-reference.

   Best practice

   Add an Information node next to the Bookkeeper Review node with a note: "After approving, copy the message from the Submitted Result field and send it from the practice email account. Dave by voolama does not send email automatically." The Information node does not affect execution but makes the workflow self-documenting. The workflow-editor help documentation states: 'Use Information nodes for documentation. Drop Information nodes next to complex sections of your graph to explain the intent.'

3. 3

   Set the workflow to active and run the first instance

   A workflow must be set to active status before team members can launch instances. A draft workflow cannot be run. Only Admin and Curate roles can change workflow status from draft to active. Only Admin and Use roles can view the Task Inbox and submit decisions. The tasks help documentation states: 'Tasks are intended for users who actively participate in workflow execution (Admin and Use roles).'

   Do this

   • Navigate to Workflows in the sidebar. Find the "Document Chase" workflow. Change its status from draft (gray badge) to active (green badge).
   • Navigate to Instances in the sidebar. Click Create Instance. Select "Document Chase" and its v1 version. Click Launch.
   • Navigate to Tasks. Open the Enter Missing Documents task. Read the Payload section. Enter the client name and the list of missing documents. Click Approve to submit.
   • The Agent Interaction node runs automatically. Refresh Tasks after a few seconds.
   • Open the Bookkeeper Review task. Read the Payload: it contains the AI-drafted follow-up email. If the draft is accurate and professional, click Approve. If it needs revision, click Request Changes and add a Comment. If it is wrong, click Reject.
   • After approval, the instance status changes to completed. The approved message text is in the Submitted Result field of the completed Bookkeeper Review task. Copy it and send it through the practice's normal email system.
   • To verify the audit record: navigate to Admin, then Settings, then Audit Log. Filter by the instance ID. The log shows the Enter Missing Documents submission, the Agent Interaction invocation, and the Bookkeeper Review decision with the reviewer's user ID and timestamp.

   Example

   A hypothetical bookkeeper launches the Document Chase workflow, submits the missing document list for a hypothetical client, and waits. Thirty seconds later a Bookkeeper Review task appears. The AI has drafted a short, professional email listing three missing items as bullet points. The bookkeeper reads it, confirms the tone and content are correct, adds a comment, and clicks Approve. The instance status changes to completed. She copies the approved message text and sends it from the practice's email account. See the Work a Task from Your Inbox quick-reference.

   Best practice

   The tasks help documentation states: 'Use comments to document your decisions. Even a brief note explaining why you rejected or requested changes helps workflow owners and other reviewers understand the history.' The Human Review audit record captures the reviewer's name, decision, comment, and timestamp. That record is the practice's internal evidence that a named person reviewed each message before it was used. Export the audit log as CSV from Admin, then Settings, then Audit Log for the engagement period as part of the practice's regular recordkeeping routine.

Role quick reference for a bookkeeping practice team in Dave by voolama

Where this advice stops