AI & Automation Guide

What AI-assisted work means in Kraal

Kraal includes an AI assistant that you can interact with from any view in the application. The AI helps accelerate recurring accounting tasks — from triggering syncs to reconciling bank transactions to creating documents.

AI-assisted workflows are intended to:

• Accelerate repetitive operational steps.
• Handle ambiguous bank reconciliation matches.
• Surface analysis, explanations, and recommendations.
• Create accounting documents on your behalf.

What the AI assistant can do

Trigger syncs and check status

• "Sync my QuickBooks" — triggers a QuickBooks data sync.
• "Any sync errors?" — checks integration status and surfaces issues.
• "Sync bank transactions" — triggers a bank feed refresh.

Bank reconciliation

• "Help me reconcile the bank" — runs AI-assisted matching on unreconciled transactions.
• The AI handles ambiguous matches that smart matching alone can't resolve confidently.

Create documents

• "Create an invoice for Acme Corp" — creates accounting documents using AI skills.
• "Record a payment from Client X" — creates payment entries.

Analysis and reporting

• Get AI-powered explanations on financial reports.
• Ask questions about trends, anomalies, or specific accounts.

General assistance

• Ask about workflow status, next steps, or how to use a feature.
• The AI works within your current client context.

Core operating loop

1. Provide a clear instruction with client and period context.
2. The AI launches a task tracked in the Command Center.
3. Monitor status and the event feed as the task runs.
4. Review outputs and artifacts when complete.
5. Approve, refine, or rerun as needed.

Instruction quality best practices

Use prompts that are:

• Specific: name the objective and expected output.
• Scoped: include the time period and target context.
• Verifiable: ask for outputs that can be checked.

Example structure:

• Goal: what you need done.
• Scope: which client and period.
• Output: what format or result you expect.
• Guardrails: any constraints or assumptions.

Task lifecycle model

AI tasks move through these statuses:

• Queued — waiting to start.
• Running — actively executing.
• Completed — finished and ready for review.
• Failed — encountered an error (triage and rerun or escalate).
• Archived — historical record.

All tasks are tracked in the Command Center, where you can see timelines, event feeds, and output artifacts.

Operationally:

• Treat completed as "ready for review," not automatic sign-off.
• Treat failed as a triage event — investigate, then rerun with corrected context or escalate.

Artifact review checklist

Before accepting AI-assisted outputs:

• [ ] Output corresponds to the requested scope and client.
• [ ] Key numbers and assumptions are plausible.
• [ ] Any exceptions are documented.
• [ ] Follow-up actions are assigned when needed.

Workflow library usage

Use workflow library entries when repeatable patterns exist. This improves consistency versus free-form prompting and helps teams standardize execution across clients.

Safety and governance guardrails

• Keep human validation in the loop for critical financial outcomes.
• Use role-appropriate permissions for who can initiate and approve actions.
• Maintain an auditable trail through task event logs and artifacts.
• AI suggestions should always be reviewed before they affect financial records.