Back to all articlesCampaign delivery operations

Keep Sending or Pause? Triage a Campaign With Partial Failures

Alactic Editorial9 min read

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

Some recipients have delivery confirmations, others have failed, and the campaign summary says “failed.” Stopping everything may interrupt unaffected work. Continuing everything may expand a problem you have not contained.

The useful question is which remaining sends have enough evidence to proceed. Continue a separate group when comparable outcomes support that choice. Hold a subset when failures follow an observable boundary. Pause the campaign when failures cross boundaries, the evidence is unreliable, or a narrower hold cannot be enforced.

This workflow covers initial containment, not technical root-cause investigation or template review procedures. It ends with a filled decision record that another operator can use without reconstructing the incident.

1. Establish what has actually happened

Take a timestamped snapshot before interpreting the campaign label. Map the available message statuses into four operational buckets, retaining the original status values:

  • Confirmed delivered: a delivery confirmation is present.
  • Reported failed: an explicit failure is present; preserve its reason and timestamp where available.
  • Unresolved: processing has begun, but the delivery outcome is not established.
  • Not yet submitted: the message has not entered sending processing.

A successful submission is not a delivery confirmation. A queued message may not be equivalent to an unsubmitted message. If the available records cannot distinguish those states, document the uncertainty instead of forcing a classification.

Record whether dispatch is still active and what the available pause control affects. A pause request does not establish that queued or in-flight messages have stopped. Capture its request time and account for subsequent status changes in the next snapshot.

For WhatsApp, campaign preparation, template approval, and delivery are separate states. Approval does not guarantee delivery. Recipient permission still matters, and opt-outs must be honoured throughout the incident. Use the current WhatsApp Business Messaging Policy for precise requirements.

2. Compare groups that could explain the pattern

Begin with observable dimensions: exact content version, send period, sender, reported failure reason, and any processing group already present in the records. These are comparison fields, not diagnoses. Failures clustered around one version do not, by themselves, prove that its content caused the problem.

For each candidate group, compare delivered, failed, unresolved, and unsubmitted counts at a consistent observation time. A recently attempted group may have more unresolved records simply because its outcomes have had less time to arrive.

  • Do explicit recipient-specific reasons appear on a few records while comparable peers receive delivery confirmations?
  • Are failures concentrated in one version across different recipient groups?
  • Do different versions begin failing during the same period?
  • Is an unresolved backlog obscuring the apparent comparison?

Counts need denominators. Eight failures among a small attempted group mean something different from eight among a much larger group. Keep unresolved outcomes visible rather than counting them as successes.

An untouched queue is not an unaffected comparison group: it has no delivery evidence yet. Likewise, an earlier successful group may be a poor comparison if conditions changed before the current failures began.

3. Use this filled hypothetical triage matrix

Fictional worked examples: the following entries are independent hypothetical snapshots, not stages of one campaign or Alactic customer results. Their numbers illustrate decisions, not acceptable failure thresholds. Reuse the four fields for each pattern you investigate.

Entry A: isolated recipient failures

  • Observation: of 300 attempted messages, 292 are delivered, five have explicit recipient-specific failure reasons, and three remain unresolved. The failures span different content versions; comparable peers in each version have delivery confirmations.
  • Evidence to check: preserve each reported reason and check for another shared factor, such as a recent send period. Establish whether the unresolved records could conceal a new cluster.
  • Initial decision: hold the five failed records for individual review. Further unsubmitted work is a candidate for continuation, subject to a recorded scope and stop condition. The three unresolved attempts are not retry candidates.
  • What would overturn it: failures spread among peers, reasons no longer support a recipient-specific distinction, or unresolved outcomes expand across groups.

Entry B: shared-content pattern

  • Observation: version B has 120 attempts: 84 delivered, 28 failed, and eight unresolved. Other versions have 360 attempts: 336 delivered, eight failed, and 16 unresolved. Version B failures span multiple recipient groups.
  • Evidence to check: verify exact version identifiers, compare similar send periods, inspect reported reasons, and look for another common factor.
  • Initial decision: hold unsubmitted version B work. Other versions are candidates for continuation, not automatically cleared. Failed and unresolved attempts remain outside that continuation scope.
  • What would overturn it: comparable failures appear outside version B, version mapping proves unreliable, or the available control cannot isolate the subset.

Entry C: unexplained cross-group failures

  • Observation: of 240 attempts, 150 are delivered, 54 failed, and 36 unresolved. Failures span three versions and several processing groups during the same recent period.
  • Evidence to check: reconcile individual records with the summary, compare timestamps, and investigate whether delayed reporting explains the apparent spread. A shared error label is evidence to inspect, not proof of a cause.
  • Initial decision: pause further campaign dispatch using an authorised control. No reliable unaffected boundary currently supports continued sending.
  • What would overturn it: reconciled evidence establishes a narrower boundary and supports a specifically authorised restart.

4. Work through the numbers before choosing a scope

Extend hypothetical entry B into a campaign of 1,200 intended messages. At the snapshot, 480 have been attempted and 720 have not been submitted. The following breakdown accounts for every intended message.

GroupDeliveredFailedUnresolvedUnsubmittedTotal
Version B84288180300
Other versions336816540900
Campaign42036247201,200

Version B has 28 failures among 120 attempts, approximately 23.3%. Other versions have eight failures among 360 attempts, approximately 2.2%. Those hypothetical percentages describe the snapshot; they do not establish a universal pause threshold or prove causation.

The unresolved counts also matter: eight for B and 16 elsewhere. Their eventual outcomes could strengthen or weaken the proposed boundary. Check comparable send periods before treating the difference as meaningful.

The possible continuation scope is therefore 540 unsubmitted messages, not all 900 records belonging to other versions. The latter total includes deliveries, failures, and unresolved attempts. Confusing those populations is how a containment decision can accidentally become a resend instruction.

5. Record the decision and handoff in one place

The following filled artifact continues the fictional example. It is a recommended operating record, not evidence that a particular workspace provides these controls. Its authorisation applies only after the named checks are recorded as complete.

Filled decision record: version B hold

  • Snapshot: hypothetical review at 10:00. Campaign total: 1,200; delivered: 420; failed: 36; unresolved: 24; unsubmitted: 720.
  • Working interpretation: failures are concentrated in version B. Cause remains unconfirmed.
  • Evidence supporting separation: exact version mapping and comparable send-period checks are complete in this fictional scenario. Other versions have 336 delivery confirmations among 360 attempts. Reported failure details remain attached to individual records.
  • Held scope: 180 unsubmitted version B messages. Its 28 failed records require separate review; its eight unresolved attempts await outcomes.
  • Permitted continuation: only the 540 unsubmitted messages in other versions. All 24 unresolved attempts remain excluded from retries. The other eight failed records also require separate review; none is authorised for retry by this record.
  • Control check: the authorised operator has verified that the subset hold excludes version B from further dispatch. If that separation cannot be maintained, use a campaign-wide pause rather than a replacement campaign with overlapping work.
  • Owner and checkpoint: campaign operations lead owns the decision. Review outcomes after the next authorised batch, before authorising another. Record the batch identifier and review time.
  • Stop condition: pause further dispatch if the same failure pattern appears outside version B, version mapping becomes unreliable, or unresolved outcomes prevent a meaningful comparison.
  • Release status: no release of held work and no retries are authorised by this record.

This record trades some disruption for a bounded continuation. It is not a claim that the continuing group is risk-free. The next operator should be able to identify the permitted population directly, rather than infer it from a campaign-level status.

6. Avoid shortcuts that obscure the next decision

Do not resend everyone labelled unsuccessful. That label may combine explicit failures with unresolved attempts. A retry decision requires attempt-level review, including whether delivery has since been confirmed and whether another attempt is justified and permitted.

Do not change several variables at once. Replacing content, changing routing, and rebuilding the send together makes later outcomes harder to interpret. Keep initial containment distinct from corrective work, and preserve the original version-to-attempt mapping.

Do not use administrative progress as delivery evidence. A prepared campaign, completed content edit, or approved template tells you something about readiness, not recipient delivery.

Do not let an automated suggestion authorise itself. A rule or assistant may flag a pattern or propose a hold, but consequential sending decisions require an authorised owner and a clear human escalation path. Check available controls and permissions in the workspace and current plan; this workflow does not assert that Alactic CRM, Communications, ACID, or the Android app share a particular capability.

7. Require evidence before releasing held work

A hold needs a release condition that addresses why it was imposed. “Content updated” or “review complete” is insufficient on its own.

For a shared-content hold, the authorised owner should record what uncertainty has been resolved, what evidence supports a bounded restart, its observation checkpoint, and its stop condition. Identify the exact version and records eligible for that restart. If the suspected boundary remains unexplained, say so and explain why the proposed scope is still supportable—or retain the hold.

Any necessary content or template work must follow the appropriate process, but completing that work is not evidence of delivery. For WhatsApp Platform conversations, approved templates are used to initiate conversations and outside the 24-hour customer-service window, which opens or resets with a user message. Check current policy requirements before authorisation.

For recipient-specific failures, assess each record rather than releasing it because peers succeeded. For a cross-group pause, obtain reconciled evidence supporting a narrower restart; otherwise escalate while maintaining the pause.

Where a limited restart is justified, define the observation checkpoint before sending. Do not expand merely because submission succeeded while relevant delivery outcomes remain unresolved.

8. Judge the next review by observable outcomes

The next review should answer three questions: Is every intended record accounted for? Did actual dispatch stay within the authorised scope? Do new outcomes support or overturn the working boundary?

Compare the new snapshot with the preserved original, including messages that progressed after a pause request. Record changes without erasing what was known when the earlier decision was made.

The immediate success measure is a controlled, explainable decision—not a clean-looking campaign label. A useful handoff shows exactly what may send, what remains held, who owns the next review, and which observation will change the plan.

Campaign operationsDelivery triageWhatsApp business messaging

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