Approve or reject AI tasks

What AI tasks are

AI tasks are prepared pieces of bookkeeping work.

They may suggest a category change, a rule, an alias, a transfer cleanup, a vendor cleanup, a payee update, a 1099 review item, or a month-end action.

The point is simple: the AI should prepare the work so you are reviewing a clear decision instead of starting from scratch.

Approve vs reject

Approve a task when the proposed change matches the real business activity.

Reject a task when the suggestion is wrong, too broad, unclear, or not needed.

Rejection is useful. It tells LeedBooks not to repeat that pattern blindly. A good AI employee should learn from both approvals and rejections.

What to check

Before approving, check the practical details:

• Does the amount match the transaction?
• Does the category make sense?
• Is this actually a transfer?
• Is the vendor or payee name clean?
• Would a rule be too broad?
• Does this affect taxes or contractor tracking?
• Is the explanation specific enough?

If the answer is unclear, ask the AI for a shorter explanation.

When to ask AI

Ask the AI when a task looks close but you are not sure.

Good questions:

• "Why did you suggest this?"
• "What would this change affect?"
• "Is this a transfer or income?"
• "Would this rule catch unrelated transactions?"
• "Show me similar transactions."
• "What happens if I reject this?"

The AI should help you decide, not pressure you into approving.

What approval does

Approval can commit a prepared change depending on the task type.

Some approvals affect one transaction. Some create a rule for future activity. Some clean up names or aliases. Some mark a task as reviewed.

If a task affects more than one item, read the scope before approving.

Good habits

Keep approvals boring and consistent.

Useful habits:

• Review small batches often.
• Reject unclear tasks instead of guessing.
• Ask for similar transactions before creating broad rules.
• Keep categories simple.
• Treat tax-sensitive tasks carefully.
• Use notes when the reason matters later.

The goal is not to make every decision yourself forever. The goal is to teach the AI enough that the routine decisions become easier.