Back to all articlesAutomation design

Define stop paths that cancel future work cleanly

Alactic Editorial10 min read

Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.

A workflow reaches its stop condition, but a reminder is still waiting to run. A reviewer still has an open approval request. A delayed response arrives and moves the record forward again. The main branch has ended; the work has not.

To design an automation stop path, define the scope of the stop, prevent new work from being authorised within that scope, and give every pending component a recorded disposition. Account separately for actions already in progress and actions already completed. Stopping future work does not reverse the past.

The practical test is simple: after a stop, can an operator explain what will no longer happen, what may still finish, what remains unresolved, and why? The filled checklist below turns those questions into an operating specification rather than a vague instruction to cancel everything.

1. Define exactly what the stop applies to

Start with the unit being stopped. It might be one workflow run, one branch, or all future runs of a particular process for a record. Those are different decisions. Stopping a single follow-up sequence should not silently cancel an unrelated service task. Conversely, ending the current run is insufficient if the same trigger can immediately start another one.

Write the scope in terms an operator can verify:

  • Target: the specific run, branch, record, or recurring process affected.
  • Trigger: the event or authorised decision that requests the stop.
  • Boundary: the work that must not receive further authorisation once the stop is accepted.
  • Exclusions: related work that remains active and why.
  • Re-entry rule: whether a later event can start a new run, and what authorisation it requires.

Also distinguish stopping from pausing. A pause preserves an intended route back into the same unfinished work. A stop should make that run terminal. If the business later wants to proceed, a newly authorised run is usually easier to explain than silently reviving cancelled steps.

This is a design recommendation, not a statement that any particular product provides these controls. Map it to the capabilities available in your workspace and current plan before relying on it.

2. Classify work before deciding what cancellation means

A single pending label hides important differences. An action waiting for permission is not equivalent to an action that has already been handed off. Give each component a lifecycle state that determines its stop treatment.

  • Planned: defined in the workflow but not yet scheduled. Do not create it after the stop.
  • Queued: waiting to begin. Mark it as cancelled or otherwise ineligible to proceed, with a reason.
  • Waiting: dependent on time, an event, or a response. Close the wait and define how a late event will be handled.
  • Awaiting review: waiting for a human decision. Close or supersede the request so a later approval cannot authorise the stopped work.
  • In progress: already started or handed off. Record what is known and reconcile the eventual outcome.
  • Completed: already performed. Preserve the result; consider any corrective action separately.

Include retries and secondary branches in this inventory. A retry is future work, even when it belongs to an action that previously failed. An escalation timer is still pending work, even when the main task appears closed.

The design goal is not to claim that every action can be interrupted. It is to ensure that nothing remains unaccounted for. Where interruption is unavailable or uncertain, the stop record should say so explicitly.

3. Use a filled stop-path checklist

Fictional worked example: A business uses a workflow to follow up on a requested quotation. The run includes an internal ownership task, a timed follow-up, and a revised-quote review. An operator receives confirmation that the customer no longer wants the quotation and requests a stop. All names, timings, identifiers, and outcomes below are hypothetical.

This checklist describes the required business behaviour. It does not prescribe technical queue-cancellation mechanisms or assert that a product implements them.

Filled hypothetical checklist: quotation follow-up run QF-204

  1. Stop trigger and authority: Operator Maya records the customer’s request to end quotation follow-up. Her role permits stopping this workflow. The request is accepted at 14:10 on the hypothetical business day.
  2. Scope: Stop run QF-204 and prevent automatic re-entry into the same quotation-follow-up process for this quotation. Leave the separate open service case unchanged.
  3. Pending actions: Cancel the queued follow-up task and the unstarted revised-quote preparation step. Record each as not performed because the quotation follow-up was stopped. Neither becomes eligible through an ordinary retry.
  4. Timers and waits: Close the hypothetical next-day reminder timer, the wait for a quotation response, and the review-escalation timer. A late response may be recorded, but it must not resume this run.
  5. Review requests: Close review R-83 as superseded by the stop. Show the reviewer that no decision is needed. A previously opened approval link must not authorise the cancelled revision.
  6. In-progress work: An internal ownership update started before 14:10 has no confirmed outcome. Record it as awaiting reconciliation, assign Maya as owner, and prevent its result from launching follow-up steps.
  7. Completed work: Preserve the already-created quotation record and the completed internal task history. Do not delete them or describe them as undone.
  8. Corrective action: No reversal is authorised by this stop. If the ownership update completes and needs correction, create a separate reviewed task with its own reason.
  9. Human handoff: Maya owns the unresolved ownership-update outcome. The quotation team sees that follow-up has ended; the service team retains its separate case.
  10. Final recorded reason: Use reason category customer declined quotation follow-up. Record the operator, acceptance time, scope, cancelled components, preserved completed work, unresolved action, and re-entry restriction.
  11. Completion test: Mark the stop fully reconciled only when every affected component has a disposition and the in-progress ownership update has a confirmed outcome. Until then, display the run as stopped with reconciliation outstanding.

The checklist deliberately avoids a single cancelled status for everything. It distinguishes work that never happened, work that happened, and work whose outcome is not yet known.

4. Separate the stop decision from cleanup completion

A stop has two useful milestones: acceptance and reconciliation. Acceptance establishes that no further work should be authorised within the defined scope. Reconciliation accounts for the components that existed when that decision was made.

If you wait for every component to finish cleanup before recognising the stop, more work may become eligible while the operator believes cancellation is underway. If you immediately label everything cancelled, you may conceal actions that already started.

Use a small set of clearly explained business states:

  • Stop requested: the request has arrived but its authority and scope have not yet been accepted.
  • Stopped, reconciliation outstanding: future work is no longer authorised, but one or more existing components need an outcome.
  • Stopped, reconciled: every affected component has a known disposition.

These labels are recommendations, not required interface wording. The important distinction is that outstanding cleanup must not act as permission for the workflow to continue.

For work near the boundary, record whether it had already started when the stop was accepted. If that cannot be established, preserve the uncertainty and assign an investigation. Do not replace an unknown outcome with a reassuring cancellation label.

5. Close waits and reviews without creating new triggers

Waits and reviews deserve explicit treatment because they often produce events after the workflow has ended. A timer can expire, a reviewer can approve from an old screen, or a response can arrive after the relevant branch has closed.

For each late event, define both its information value and its authority. A response may still be useful history without having permission to restart work. A reviewer’s decision may be recorded as late without being applied.

In the hypothetical quotation example, closing review R-83 should have these observable outcomes:

  • The reviewer can see that the request no longer requires action.
  • An attempted late approval does not release the revised quotation step.
  • The attempt remains distinguishable from an approval applied while the review was active.
  • If the reviewer believes work should resume, there is a named human route to request a new authorised run.

Apply the same reasoning to escalation. Closing a review but leaving its overdue reminder active can create unnecessary work for another person. Inventory the reminders, reassignment rules, and secondary waits attached to each review, not just the review itself.

A repeated stop request should also be harmless: it may add relevant context, but it should not recreate cleanup tasks or reopen closed reviews.

6. Keep reversal separate from cancellation

A stop changes what may happen next. A reversal attempts to change something that already happened. Combining them creates ambiguity about authority, risk, and history.

In the hypothetical example, the quotation record exists before the stop. Preserving it accurately reflects completed work. Deleting it as part of cancellation would be a separate destructive action, not proof that the workflow stopped cleanly.

Where correction is needed, specify it as a new action:

  1. Identify the completed action and its confirmed effect.
  2. Explain why correction is needed.
  3. Identify who can authorise the correction.
  4. Check whether other work now depends on the existing result.
  5. Record the corrective action and its outcome separately from the stop.

The trade-off is additional review effort. The benefit is a clearer operational record: the original action occurred, the workflow stopped, and an authorised correction followed. Avoid describing this as rollback unless the actual behaviour supports that description.

Some effects cannot be meaningfully reversed. This article does not cover message withdrawal rules or delivery guarantees; neither should be inferred from a workflow’s stopped status.

7. Plan recovery for incomplete cleanup

Cleanup itself can fail. A review may remain visible, a timer may lack a confirmed disposition, or an in-progress action may finish without a recorded result. Treat these as owned exceptions rather than reasons to restart the original workflow.

Define a recovery record with the affected component, last known state, required next check, responsible person, and escalation route. A retry of reconciliation should inspect or complete the missing disposition, not repeat the original business action.

For the hypothetical ownership update, Maya would first establish whether it completed. Repeating the update simply because the outcome is unknown could create an additional effect without resolving the first one.

Be especially careful with partial cleanup. If most components are closed but one remains unresolved, retain the accepted stop and show the outstanding exception. Reopening the workflow to fix the exception would restore permissions that the stop was meant to remove.

Manual intervention also needs boundaries. An operator may be allowed to reconcile a component without being allowed to restart the process. Make those decisions distinct in the operating procedure.

8. Test the boundary and measure what remains

Reviewing a diagram is not enough. Test the moments when state changes overlap. Use fictional records in an appropriate test environment and verify observable outcomes rather than relying only on a final status label.

  • Stop before a queued step starts: the step does not proceed and has a recorded disposition.
  • Stop while a timer expires: the expiry does not authorise new work after the stop is accepted.
  • Approve after stopping: the old review cannot release its dependent action.
  • Stop during in-progress work: the outcome is reconciled without launching downstream steps.
  • Repeat the stop: no duplicate cleanup or corrective work is created.
  • Interrupt cleanup: reconciliation can resume without reopening the stopped run.
  • Attempt re-entry: the defined restriction holds, while unrelated work remains unaffected.

For each test, capture the stop acceptance time, component dispositions, any action authorised afterward, and the unresolved owner. Distinguish a previously started action finishing later from a new action receiving authorisation after the boundary.

Next action: choose one live workflow design and fill the checklist for every pending component before changing its stop logic. Then run the boundary tests and count three things: components without a disposition, late events that restart work, and unresolved items without an owner. Use the findings to revise the design; do not treat a clean-looking final status as sufficient evidence that the stop worked.

Automation designWorkflow cancellationOperational controlsHuman review

Put the next conversation to work.

Explore Alactic Communications for WhatsApp conversations, campaigns and AI-assisted workflows. Choose the product and plan that fit your team.

Explore Communications