Decide when a completed record may enter the same workflow again
Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.
A record finishes a workflow. Later, someone corrects its phone number, adds a note, or submits a new service request. If every edit can restart the workflow, routine maintenance becomes new work. If completion blocks the record forever, legitimate future requests disappear.
Allow re-entry only when there is evidence of a distinct new business cycle, the previous outcome permits it, current eligibility still holds, and the restart is authorised. An edit can prompt an eligibility check; it should not, by itself, prove that another run is needed.
The practical task is to write a restart rule for each terminal outcome—not just for the broad status “finished.” The fictional decision table below gives you a starting policy, including when to allow, hold, or deny a new run.
1. Define what counts as another cycle
Start with a sentence that describes the unit of work. “Follow up with this contact” is too broad: a contact can have several unrelated needs. “Coordinate an appointment for one submitted service request” is specific enough to finish and, later, repeat.
For a repeatable workflow, distinguish three things:
- The record: the continuing person, account, or other business entity.
- The cycle: the particular request or obligation being handled.
- The workflow instance: the execution that attempts to handle that cycle.
A new cycle should normally create a new instance linked to the previous history. It should not erase the old outcome or turn a completed instance back into an unfinished one. A replacement instance for interrupted work is different: it may still belong to the original cycle and should be labelled accordingly.
Define the boundary before choosing triggers. For example: “One cycle begins when a customer submits a distinct appointment request and ends when the request is fulfilled, withdrawn, or expires under the operating policy.” This makes a new request meaningful evidence. A changed postal address is not.
This article concerns intentional business re-entry. Repeated webhook delivery, resolving whether two records represent the same contact, and deciding whether to reopen an opportunity are separate design problems.
2. Require positive evidence, not just a changed record
Use an explicit new-cycle signal. Depending on the process, that might be a new request reference, a confirmed renewal period, or a staff-created request backed by a documented customer instruction. The signal needs a business meaning, not merely a different technical value.
A practical evidence record contains:
- A cycle reference and request type.
- The source and time of the request.
- The relevant service, period, or subject.
- A link or reference to the supporting instruction.
- The person or authorised process proposing entry.
These are recommended information requirements, not claims about available product fields. Adapt them to the workspace and tools you actually use.
Do not count a newer timestamp as sufficient evidence. Importing old information today does not make the underlying request new. Nor does changing a status from completed to open establish a new need. Require the underlying business event.
Evidence can also be contradictory. A new request reference paired with a note saying “correction to previous request” should be held for review. A record of a request that was already fulfilled should not qualify just because its reference was added later.
Time-based gaps can support the rule, but cannot replace it. A waiting period may reduce accidental repetition while also delaying legitimate work. Start with the distinct-cycle test; add a waiting period only when the process has a concrete reason for one.
3. Use a filled re-entry decision table
Fictional worked example: A service business uses an appointment-coordination workflow. It handles one request at a time for each customer within this workflow. Completion means the requested appointment took place. Cancellation reasons distinguish withdrawal, duplicate requests, and a standing request not to receive automated appointment follow-up.
The following is a hypothetical operating policy, not a description of an existing product control. The compact table gives the decision; the numbered rules immediately below supply the required evidence and action.
Prior outcome Restart circumstance Decision Rule Completed Distinct appointment request Allow A Completed Edit or old request only Deny B Cancelled Earlier request withdrawn Allow* C Cancelled Duplicate of another request Deny D Cancelled Standing automation restriction Hold E Expired Customer renews the request Allow* F Expired Time passes or record is edited Deny G Manually stopped Staff pause or unresolved issue Hold H Manually stopped Issue cleared; restart approved Allow* I Any outcome Missing or conflicting evidence Hold J Any outcome Another instance is active Hold K Any outcome Current eligibility fails Deny L * Only after all common eligibility gates pass.
- A — Completed, distinct new request: Allow a new cycle when the customer requests another appointment and the request has its own reference, subject, and supporting instruction. Preserve the completed instance unchanged.
- B — Completed, incidental activity: Deny entry for a corrected phone number, internal note, owner change, or the same fulfilled request. Record that no distinct cycle was established.
- C — Cancelled because the earlier request was withdrawn: Allow a fresh request if the customer now wants a new appointment. Do not interpret withdrawal of one request as a permanent ban, or interpret a later edit as renewed intent.
- D — Cancelled as a duplicate: Deny a restart of that duplicate cycle. Direct staff to the surviving request. A genuinely unrelated new request must be assessed separately, not presented as a restart of the duplicate.
- E — Cancelled with a standing restriction: Hold automated re-entry. A designated reviewer must confirm whether the restriction has validly changed. If the restriction remains, deny; if cleared, reassess the new request against every gate.
- F — Expired, request renewed: Allow a new cycle only with fresh confirmation that the customer still wants an appointment and refreshed details where relevant. Link it to the expired cycle so staff can understand the history.
- G — Expired, no renewed intent: Deny entry triggered only by elapsed time, a batch update, or a status edit. Expiry ends this attempt; it does not authorise repeated attempts indefinitely.
- H — Manually stopped, issue unresolved: Hold. The stop reason and its scope remain effective until an authorised reviewer resolves them. A newly submitted request does not silently clear the stop.
- I — Manually stopped, issue resolved: Allow only after an authorised person records the resolution and restart approval. State whether the new instance replaces interrupted work in the same cycle or handles a genuinely new cycle.
- J — Incomplete evidence: Hold with a named owner and a specific missing item, such as confirmation of whether the request is new. Do not guess from the terminal status.
- K — Active instance exists: Hold under this example's one-active-instance rule. Decide whether the new request belongs in the active cycle or should wait as a separate cycle.
- L — Eligibility fails: Deny entry while the disqualifying condition applies, even if the new request is genuine and the previous outcome would otherwise permit re-entry.
The distinction between hold and deny matters. Hold means a named decision or piece of evidence is outstanding. Deny means the proposed restart does not meet the rule. Neither should create an active workflow instance.
4. Apply common gates after checking the prior outcome
A permissive row is not blanket approval. Apply a common set of gates to every proposed entry:
- Purpose: Is this request within the workflow's scope?
- Evidence: Is there a supported new cycle, or an explicitly approved replacement of interrupted work?
- Prior restrictions: Are any stop conditions or standing exclusions still active?
- Current eligibility: Are the necessary details and business conditions valid now?
- Concurrency: Would another active instance conflict with this one?
- Authority: Is the person or process permitted to initiate this run?
Set precedence explicitly: an unresolved stop or failed eligibility gate overrides an otherwise allowable restart. Staff should not be able to clear a standing restriction merely by selecting a more permissive outcome reason.
The fictional example serialises appointment coordination per customer. That is a deliberate trade-off: it simplifies conflicting instructions but can delay unrelated requests. If your business permits parallel cycles, define the scope more narrowly—for example, per request rather than per customer—and specify how shared information is handled.
Check authority at the point of restart, not only when someone configures the workflow. OWASP's authorization guidance recommends least privilege, deny-by-default behaviour, and permission checks on every request. This supports restricting restart and override permissions; it does not establish that any particular platform implements those controls.
5. Make the decision explainable to an operator
A useful decision record answers “Why did this run start?” without requiring someone to reconstruct a timeline from scattered notes. Keep the previous instance reference, terminal outcome and reason, proposed cycle evidence, rule applied, decision, decision time, and responsible actor together.
For an allowed entry, include the new instance reference and its relationship to earlier work. For a hold, include the reviewer and unresolved question. For a denial, state which condition failed and what evidence, if any, could justify a future proposal.
Hypothetical decision: “Held under rule H. Previous instance was manually stopped because appointment details were disputed. A new request is recorded, but the dispute remains unresolved. Service supervisor to confirm the correct details before any restart.”
This is more useful than “re-entry disabled.” It identifies the obstacle without treating a new request as permission to ignore earlier concerns.
Where a workflow includes messaging, assess communication eligibility separately from workflow eligibility. A valid new service request does not automatically permit every outbound message. For WhatsApp, honour recipient permission and opt-outs; approved templates are used to initiate Platform conversations and outside the 24-hour customer-service window, which opens or resets with a user message. Check the current WhatsApp Business Messaging Policy for precise requirements. Template approval is not a delivery guarantee, and automation needs a clear human escalation path.
6. Test boundary cases before enabling re-entry
Review decisions against realistic record histories, not just clean examples. Use the fictional table to test these cases:
- A completed request receives a spelling correction: deny under B.
- A customer submits a distinct request after completion: allow under A if all gates pass.
- An expired request receives an internal note saying “try again”: deny under G unless renewed customer intent is established.
- A new request arrives while a manual stop remains unresolved: hold under H, even though the request itself is genuine.
- A cancelled duplicate is resubmitted without a new underlying need: deny under D.
- A legitimate request arrives while another instance is active: hold under K in this example, then assign someone to resolve the overlap.
Also test missing reason codes, unavailable supporting instructions, and changes between the initial check and actual entry. Recheck material conditions when entry is authorised. An earlier eligibility decision should not outlive a newly recorded restriction.
If the workspace cannot enforce a needed distinction, do not replace it with an unrestricted “allow repeat” setting. Use a controlled review queue or manual entry process until the required conditions can be evaluated reliably. Check the workspace and current plan for relevant capabilities rather than assuming a proposed design is available.
7. Recover from a wrong restart without losing history
When an unintended instance starts, first prevent further authorised actions from that instance while the owner investigates. Record which rule should have blocked it and whether any work or communication has already occurred.
Keep the original evidence and decision history. Mark the unintended instance with a clear stop reason, then correct the rule, evidence requirement, or permission that failed. Deleting history may hide the symptom while leaving the same restart path open.
For a wrongly denied request, recover the legitimate business work through review. Do not solve one missed request by enabling re-entry for every completed record. For a held request, resolve the specific question and reassess eligibility; releasing a hold should not bypass the remaining gates.
If external communication occurred, assign a human to assess the appropriate follow-up. Do not automatically send a corrective message simply because an instance was stopped.
8. Review one workflow with observable decisions
Choose one repeatable workflow and classify a recent set of terminal instances by both status and reason. Write a restart rule for every reason you find, using the hypothetical table as a starting artifact rather than copying its appointment-specific assumptions.
Before enabling the policy, ask operators to classify the same sample independently. Disagreement reveals ambiguous cycle definitions, unclear stop scope, or missing evidence. Resolve those differences in the rules, not through informal exceptions.
During the first review period, track allowed entries with complete evidence, denied entries later found legitimate, unintended restarts, unresolved holds, and attempts blocked by an active instance. Examine the underlying cases rather than setting an arbitrary target rate.
Your next action: select one workflow, name its cycle, and fill in the completed, cancelled, expired, and manually stopped rules with the process owner. Finish only when each possible restart has an allow, hold, or deny decision—and an operator can explain what evidence would change it.
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