# Place checkpoints where the consequence changes

Asking for approval after every small step creates queues and weak attention. Put review around meaningful commitments, with enough detail to understand the combined effect.

By Cobnex editorial. Published 2026-09-10. Updated 2026-09-11.

## Separate preparation from commitment

An assistant may gather records, draft a comparison and calculate a proposed change before anything consequential happens. Requiring approval for each harmless preparation step can slow the task without improving the final decision.

Identify the point where information is released, a record is changed or a commitment becomes difficult to reverse. That is a natural place to consider a checkpoint. The exact boundary depends on the workflow and its authority requirements.

Do not use reversibility as the only criterion. A technically reversible change may still notify customers, trigger downstream jobs or expose information that cannot be recalled.

## Estimate the review workload

Measure expected proposals per day and the time needed to inspect a complete proposal. Include requests for clarification, re-review after changes and coverage outside normal reviewer availability.

For an illustrative calculation, sixty proposals requiring five minutes each represent five hours of active review before interruptions and rework. If one reviewer has only two hours available, the queue will grow even when the software responds instantly.

Use these estimates to choose staffing, batching or narrower automation scope. Do not hide the mismatch behind an optimistic response-time target.

## Batch only compatible decisions

A batch approval can reduce repetitive work when the actions share a clear rule and the reviewer can inspect the full effect. It becomes risky when important exceptions disappear inside a large total.

Show which items differ, which failed validation and what the batch will change. Bind the decision to the exact item set. Adding another item after approval should not inherit the earlier decision.

Consider partial approval if the business process supports it, but make the resulting state explicit. The executor needs to know which items are authorised rather than treating the batch as one vague yes.

## Watch for review fatigue

High approval rates and fast clicks may indicate routine, correct proposals or a checkpoint that people no longer examine. Use sampled review and controlled evaluation cases to understand the difference.

Improve proposal quality before adding more reviewers. Missing evidence, unclear target names and repeated harmless changes can consume attention that should go to consequential exceptions.

Keep mandatory authority requirements intact, then tune where human judgement adds value. The aim is a review process people can perform carefully at the actual workload, with clear commitments and manageable queues.

## Sources

- [OWASP: transaction authorisation](https://cheatsheetseries.owasp.org/cheatsheets/Transaction_Authorization_Cheat_Sheet.html)