Separate workflow entry conditions from action-time checks
Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.
A record enters a follow-up workflow while its request is open. During the wait, someone resolves the request, the recipient opts out, or another team completes the follow-up. If the delayed action relies on the original eligibility result, it can act on a state that no longer exists.
Check entry conditions when deciding whether to start a workflow. Recheck any condition whose current value could make the pending action inappropriate immediately before that action is authorised to proceed. Some conditions belong in both places: checking them at entry avoids unnecessary workflow runs, while checking them again prevents stale eligibility from becoming permission to act.
The practical task is not to repeat every rule everywhere. It is to distinguish historical reasons for entry from facts that must still be true at execution. The filled hypothetical matrix below shows how to make that distinction, including what happens when a record changes during a wait.
1. Define the pending action before classifying conditions
Start with a specific action, not a broad workflow name such as “lead nurturing” or “customer follow-up.” Different actions can require different current-state checks even when they sit in the same workflow.
For the fictional worked example in this article, the pending action is: send an already authorised WhatsApp follow-up about an unresolved service request after a hypothetical two-day wait. The follow-up should not proceed if the request has been resolved, the recipient has opted out, responsibility is unresolved, or the same follow-up has already been completed.
That statement exposes the facts that matter at execution. By contrast, “the request originally arrived through the website” explains why this record entered this particular workflow. A later correction to the source field does not necessarily make the service follow-up inappropriate.
Write an action statement with these elements:
- Action: what will happen, to which record and recipient.
- Purpose: what unresolved need the action serves.
- Current prerequisites: what must still be true when it happens.
- Stop conditions: which changes make the action unnecessary or inappropriate.
This article covers where those conditions belong. It does not define your permission policy, design permission records, or explain API authorisation. Passing workflow eligibility checks must not be treated as a substitute for the authorisation required to perform the action.
2. Use three placements, not one universal filter
Assign each condition one of three placements.
- Entry-only: establishes why the record belongs in this workflow. It need not remain true for the defined action to remain appropriate.
- Action-time: applies to the actual execution context, selected content, or attempt. It is checked immediately before the action proceeds.
- Both: is useful for admission and must still hold when the action is due.
A useful test is: “If this value changed during the wait, would we still want this exact action to happen?” If the answer is no, the condition needs an action-time check. If the condition also helps exclude unsuitable records before the wait, place it in both.
Do not classify a field solely by whether it can change. A source field can change without affecting the pending action. Conversely, a permission status that rarely changes still matters at action time because a change could invalidate the action.
Action-time does not mean “somewhere after the delay.” Put the check at the last practical decision point before execution. If another wait, queue, review, or retry separates the check from the action, reassess whether the result is still current enough to use.
3. Filled hypothetical eligibility matrix
Fictional worked example: the matrix below describes a recommended service-follow-up design, not an existing product control or a customer result. All timing, routing rules, and record details in this example are hypothetical. Each row is a condition that must pass at its assigned checkpoint.
Because the matrix concerns one defined action, its classifications should not be copied unchanged into a different workflow.
Condition Placement A. Original intake event was a website request Entry-only B. Request initially matched service-follow-up Entry-only C. Request is still unresolved Both D. Recipient permission permits this contact Both E. No recipient opt-out is currently recorded Both F. Record is not marked duplicate or invalid Both G. Current team still owns this follow-up Both H. This follow-up has not already been completed Both I. Current WhatsApp message route is suitable Action-time J. Selected content matches the current request Action-time K. Required execution authorisation is valid Action-time
Why each row belongs there
- A — Original intake event: this is a historical admission fact. Preserve the reason the run started rather than interpreting an editable source field as fresh evidence.
- B — Initial routing match: this records the workflow selected at intake. Current appropriateness is checked through the request, ownership, and content conditions below.
- C — Unresolved request: resolved requests should not enter this example workflow, and a request resolved during the wait should not receive the pending follow-up.
- D and E — Permission and opt-out: exclude unsuitable records at entry and check current status again before contact. An old positive result must not override a later opt-out.
- F — Duplicate or invalid record: exclude known problems early. If a duplicate is discovered later, stop the pending action and resolve which record should carry the work.
- G — Current ownership: this example requires the responsible team to retain ownership. A transfer prompts reassignment or review rather than a message sent under outdated responsibility.
- H — Follow-up not completed: prevent an unnecessary run where possible, then check again because a person or another workflow may complete the same task during the wait.
- I — Message route: assess the route available at the time of sending, not the route that was available when the workflow began.
- J — Content relevance: check the selected message against the current request. A still-open request can change enough to make an earlier draft misleading.
- K — Execution authorisation: require the appropriate current authorisation for the action. This row is a checkpoint, not an implementation specification.
For WhatsApp, recipient permission and opt-outs matter. Approved templates are used to initiate Platform conversations and outside the 24-hour customer-service window; a user message opens or resets that window. Those facts make an entry-time decision insufficient for a delayed message. Check the current WhatsApp Business Messaging Policy for precise requirements rather than treating this hypothetical matrix as a complete messaging policy.
4. Follow a record whose state changes during the wait
In the fictional example, record SR-104 enters on Monday morning. It came from a website request, matches the service-follow-up route, is unresolved, and passes the entry checks. The workflow schedules its hypothetical two-day wait. That establishes why the run exists; it does not reserve the right to send later.
On Tuesday, a team member resolves the request. The record remains in the waiting workflow because, in this example, there is no immediate cancellation mechanism. That is not necessarily a design failure if a reliable action-time gate still prevents the obsolete action.
When the wait ends on Wednesday, the workflow reads current state. Condition C fails: the request is resolved. The correct path is to skip the follow-up, record “request resolved before action,” and close this workflow run without sending. It should not reopen the request merely to satisfy its original route.
Consider a different branch of the same hypothetical example. The request remains unresolved, but ownership has moved to another team. Condition G fails. Rather than silently sending or abandoning the customer need, the workflow holds the action and routes the case for an ownership decision. The reviewer needs to see the pending action, the failed condition, and the current owner.
A third branch illustrates uncertainty rather than ineligibility. If current permission status cannot be retrieved, the workflow does not have a passing result. Hold the action and use the defined recovery path. “Could not check” and “checked and passed” must remain different states.
5. Give failed checks different outcomes
A single “conditions not met” exit hides information that operators need. Use outcomes that distinguish work that is finished from work that is blocked.
- Skip and close: the action is no longer needed, as when the request is resolved or the same follow-up is already complete.
- Stop contact: current permission or opt-out status disallows the proposed contact. Do not automatically switch channels to work around the restriction.
- Hold for review: responsibility, content relevance, or a conflicting record state needs a human decision.
- Retry the check: a temporary retrieval problem prevents an eligibility decision. Recheck before any later attempt.
Make the human escalation path explicit: who receives the exception, what evidence they see, and what decision releases or closes the pending action. A review queue without an owner is merely another delay.
Keep preparation, template review, and actual delivery separate. A prepared message is not a sent message, and template approval is not a guarantee of delivery. Likewise, an eligible action that encounters an execution failure should not be recorded as an eligibility failure. Different failures require different recovery work.
6. Recheck after retries and meaningful changes
A retry is another attempt at the action, not an extension of the earlier eligibility result. If a sending attempt fails and is rescheduled, rerun the applicable action-time conditions before retrying. The recipient may have opted out or the request may have been resolved in the meantime.
Also distinguish retrying a read from retrying an action. When a state lookup fails before execution, a later lookup can resolve the uncertainty. When an action may have succeeded but its result is unclear, first reconcile what happened. Blindly repeating it risks duplicating work.
There is also a gap between checking state and acting on it. Reduce avoidable delay between the final gate and execution. If the action is placed into another queue, define where eligibility will be checked again. A check before a long queue is not equivalent to a check before the eventual action.
Checking more often has costs: additional reads, more exception handling, and potentially longer processing. Reserve repeated checks for facts that can affect the pending action. Do not repeatedly reconstruct historical intake criteria merely because it is convenient to reuse one large filter.
7. Make the decision observable without copying everything
Keep enough decision history to explain why a run started and why an action proceeded, stopped, or waited. A useful decision record includes:
- The workflow run and intended action.
- The entry reason and entry decision time.
- The final eligibility check time and condition version.
- Pass, fail, or unknown for each applicable action-time condition.
- The resulting action, hold, skip, or escalation reason.
- The execution result, recorded separately from the eligibility result.
Avoid copying entire record histories into every decision log. Capture the minimum evidence needed to explain the decision under your organisation's data-handling practices.
These distinctions make operational review useful. “Resolved before action” suggests the gate prevented obsolete work. “Permission lookup unavailable” suggests a dependency problem. “Ownership unresolved” suggests a handoff problem. Combining them as “automation failed” prevents the team from seeing what actually needs attention.
8. Test one delayed action before expanding the pattern
Choose one workflow with a meaningful wait and complete its condition matrix. Then test it with fictional records representing a small, explicit set of state changes:
- An unchanged record passes and reaches the authorised action.
- A resolved record skips the unnecessary action.
- An opted-out recipient does not receive the proposed contact.
- A transferred record reaches the designated reviewer.
- An unavailable current-state check holds rather than passing by default.
- A previously completed follow-up is not repeated.
- A retry evaluates changed state instead of reusing the previous pass.
For each test, write the expected outcome before running it, then compare the observed path and decision record. If the environment cannot safely exercise an action, review the decision path without dispatching it.
Your measurable next step is to account for every condition on that one pending action: entry-only, action-time, or both, with an explicit outcome for failure or uncertainty. Review how many action attempts have a recorded current-state decision, how many holds lack an owner, and whether any attempts reused a decision from before a wait. Those observations show whether the placement is working—not merely whether the workflow is running.
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