Back to all articlesCampaign delivery operations

Restart a Paused Campaign Without Replaying Completed Work

Alactic Editorial9 min read

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

A campaign pauses after a delivery problem. The fault is corrected, but the original audience still contains people who received the message, attempts with unknown outcomes, and recipients who should no longer be contacted. Restarting that audience can turn a recovery into a second incident.

The safer approach is to authorise a reconciled remainder, not replay the original audience. Establish that old execution cannot overlap with the restart, exclude completed work, hold uncertain attempts, refresh recipient eligibility, and approve a specific release manifest.

The practical control is a restart packet: one reviewable record connecting the correction, recipient-level evidence, and release decision. The workflow below is a recommended operating design, not a claim that a platform automatically preserves restart state or enforces these controls.

1. Establish the old execution’s submission boundary

A pause button is not evidence that every sending path has stopped. A worker may already have handed off requests, scheduled retries may remain active, and delayed status updates may still arrive.

Record the campaign reference, audience snapshot, pause time, and every path that could submit work. Include workers, retry queues, scheduled tasks, and manual actions. Then establish a boundary after which the old execution cannot hand off another request.

This requires more than stopping a worker. Account for queued items, items already removed from the queue, dispatched requests, and potentially in-flight requests. Record how the stop was enforced and which evidence demonstrates that no old path can submit against the proposed remainder.

A durable dispatch ledger can help if entries are recorded before handoff and every submission path uses it. A log written only after acknowledgement is insufficient when acknowledgements were lost. Equally, an empty queue does not prove that its former contents were never submitted.

If the boundary cannot be established, hold the overlap rather than launch a replacement. Preserve original records and give the restart a separate reference linked to the interrupted campaign.

2. Separate correction evidence from restart approval

“The issue is fixed” addresses the suspected cause. It does not establish which recipients still need work or whether sending remains appropriate.

Ask the technical owner for a correction record stating what changed, how it was checked, and what remains unproven. For lost acknowledgements, useful verification would show both successful acknowledgement recording and safe handling when an acknowledgement is missing. Testing only the successful path leaves the original failure case unanswered.

Keep verification limited and authorised. A test should not quietly become a restart of the audience. Its result establishes only what was checked, not delivery across the paused campaign.

If HTTP 429 caused the pause, it indicates rate limiting; a Retry-After header may indicate when to retry. Providers apply limits differently, so consult the response and relevant provider guidance rather than choosing a universal waiting period. MDN explains HTTP 429.

The technical owner can confirm that the correction is ready for review. An authorised campaign owner separately approves the reconciled release scope. Neither decision substitutes for the other.

3. Classify every intended unit using affirmative evidence

Define the unit of work before counting it. For a single-message campaign, use one intended message to one recipient. For a sequence, include the step: completing one step does not authorise replaying it when a later step resumes.

Reconcile the paused audience against queue, dispatch, attempt, and outcome records. Resolve duplicate recipient identities rather than treating separate CRM rows as separate permission to send.

  • Completed—exclude: evidence meets the campaign’s defined completion criterion.
  • Uncertain or in flight—hold: a handoff may have occurred, an acknowledgement is missing, or evidence conflicts.
  • Suppressed—exclude: a current opt-out or another exclusion prevents release.
  • Verified not submitted—candidate: affirmative evidence establishes that the unit never crossed the submission boundary.
  • Confirmed unsuccessful—review: the recorded outcome conclusively supports considering another attempt.

Missing submission evidence does not establish non-submission. A candidate might instead be supported by a verified unsent queue reconciled against a complete, durable pre-handoff ledger. If ledger coverage or queue provenance is uncertain, hold the affected units.

An accepted request without a final outcome is also a hold, not a fresh-send candidate. Conversely, not every known failure is suitable for retry: a destination problem may need correction, or another attempt may no longer be appropriate.

Assign each unit to one decision bucket, while retaining all underlying status and suppression evidence. The equation is: paused audience = completed exclusions + uncertain holds + suppression exclusions + approved remainder + other explicit holds. Balanced counts are necessary, but recipient-level evidence and non-overlapping membership make them meaningful.

4. Refresh the facts that changed during the pause

The remainder consists of work still outstanding and still appropriate to attempt. Refresh opt-outs and suppression records near release. Confirm that the content remains relevant; a time-sensitive update may become misleading while delivery is interrupted.

If content, destination, or scope must change materially, seek a new decision rather than hiding the change inside a restart. The same applies when a previously valid recipient record requires correction.

For WhatsApp, recipient permission matters and opt-outs must be honoured. Approved templates are used to initiate Platform conversations and outside the 24-hour customer-service window, which opens or resets with a user message. A pause can change the applicable messaging context. Template approval does not guarantee delivery; consult the current WhatsApp Business Messaging Policy for precise requirements. Where automation is involved, retain a clear human escalation path.

Keep campaign preparation, template review, and actual delivery as separate states. A readiness check or template approval is not evidence that an interrupted attempt failed or that a replacement message should be sent.

5. Filled hypothetical restart packet

Fictional worked example: all references, people, times, counts, controls, and verification results below are hypothetical. This filled record is reusable as a structure; it is not an Alactic customer result or a description of built-in product capabilities.

Incident and correction

  • Campaign: CAMP-HYP-042, service update; snapshot of 1,000 unique recipient-message units.
  • Pause: 12 May, 10:20 UTC, after acknowledgement responses were lost. Some requests might have been accepted despite missing local acknowledgements.
  • Correction: FIX-HYP-017 repairs acknowledgement recording and routes missing acknowledgements to a hold instead of immediate resend.
  • Verification: CHECK-HYP-006 records authorised test requests with acknowledgements and a simulated missing-acknowledgement case that entered a hold. It does not establish delivery for the paused audience.
  • Technical review: Priya accepted the correction evidence for restart review at 13:10 UTC.

Containment and handoff reconciliation

  • Submission boundary: at 10:22 UTC, the old execution’s dispatch authorisation was withdrawn and its automatic retries disabled. The dispatch control checks execution authorisation before every handoff; its recorded rejection check confirms that the old execution cannot submit again. Manual submission routes were also restricted pending approval.
  • Ledger coverage: DISPATCH-HYP-042 contains durable intent entries written before every handoff through the campaign’s only dispatch route. The reviewer verified route coverage and reconciled queue removals and worker-held items against those entries.
  • Dispatched set: 685 units crossed the boundary: 610 with recorded delivery outcomes, 30 with conclusive unsuccessful outcomes, and 45 with unresolved outcomes.
  • In-flight accounting: worker shutdown records and outbound-request records were reconciled with all 685 dispatch entries. Every request open at the pause belongs to the 45 unresolved units. No other worker-held or unrecorded request remains unaccounted for; downstream completion of those 45 remains possible.
  • Verified unsent set: 315 units remain in the quarantined unsent queue. Snapshot membership, enqueue records, and queue-removal history reconcile with the pre-handoff ledger: none crossed the boundary. The old execution cannot release this queue.
  • Boundary balance: 685 dispatched + 315 verified unsent = 1,000. Evidence package BOUNDARY-HYP-042 was reviewed before any replacement was authorised.

Recipient decision manifests

  • Completed exclusions: 610. COMPLETE-HYP-042 links each unit to its recorded delivery outcome.
  • Uncertain holds: 45. HOLD-HYP-042 contains every unresolved dispatched unit, including all potentially in-flight work. Sam owns investigation; none is approved for this restart.
  • Suppression exclusions: 15. These are members of the verified unsent set with newly recorded opt-outs. Suppression evidence was refreshed at 13:20 UTC.
  • Proposed remainder: 330. This contains 300 affirmatively verified unsent units and 30 conclusively unsuccessful attempts reviewed as suitable for another attempt. The latter records establish unsuccessful outcomes, not merely missing acknowledgements.
  • Decision balance: 610 completed + 45 uncertain + 15 suppressed + 330 proposed = 1,000. Manifest membership checks found no overlap. If the unsent evidence had been unavailable, the 300 would have remained held, leaving only 30 candidates for approval.

Approval and release

  • Restart: RESTART-HYP-042-A, linked to the original campaign.
  • Fixed scope: REMAINDER-HYP-042-A contains exactly 330 recipient-message units, with no dynamic audience expansion. The campaign owner confirmed content relevance and current permission and messaging-context checks.
  • Staging: release 20 approved units, then observe. The remaining 310 require Sam’s recorded release after review; any later suppression removes the affected unit rather than replacing it.
  • Stop conditions: any repeat submission for the same intended unit, renewed acknowledgement loss, or evidence of old-execution activity triggers another pause.
  • Ownership: Sam monitors attempts and delayed outcomes, maintains holds, and escalates technical issues to Priya and scope decisions to campaign owner Lena.
  • Approval: Lena approved this manifest at 13:35 UTC subject to the staged conditions. HOLD-HYP-042 is explicitly outside that approval.

6. Monitor the release against the approved manifest

A controlled first slice limits exposure while the team observes the correction. Choose its size and observation period according to the failure, available signals, and ability to intervene. The hypothetical quantities above are not universal thresholds.

Compare actual handoffs with the approved units. Check that exclusions remained excluded, holds remained held, and each released unit has an understandable state. A healthy summary indicator is not enough to release the next stage.

Record authorised units released, actual submissions, known outcomes, unresolved outcomes, and any submission outside scope. Distinguish accepted requests from confirmed deliveries. Track recurrence of the original fault and whether delayed outcomes change the hold set.

There is a trade-off: waiting for clearer outcomes delays the campaign, while releasing faster can create more ambiguous work. Record that choice. Early acknowledgement success does not prove that later messages will arrive.

Check the current workspace and plan for available controls. If a fixed manifest or staged release requires a manual equivalent, establish it before sending. If old execution cannot be isolated, keep the campaign paused.

7. Preserve the ledger if recovery pauses again

If another interruption occurs, stop the new execution and preserve its evidence. Carry forward existing exclusions and add newly completed work; do not rebuild the remainder from memory or restart the original audience.

New ambiguous attempts join the hold set. A held unit moves to completed only with supporting outcome evidence. A conclusively unsuccessful unit may be considered for a separately approved recovery after current eligibility checks.

If a duplicate appears, investigate the affected execution path and possible wider overlap. Keep the duplicate record as incident evidence rather than deleting it to make totals look correct.

Some outcomes may remain unresolved. Leaving those recipients out can be a deliberate decision, with the campaign owner accepting incomplete coverage rather than an unsupported retry. The next useful action is to reconcile one paused campaign without sending anything: where evidence stops, release authority should stop too.

Campaign operationsDelivery recoveryWhatsAppOperational controls

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