Prevent competing workflows from making conflicting changes
Prepared with AI assistance by Alactic Editorial. Examples identified as hypothetical are illustrative, not customer results.
A payment reminder is queued while a payment-confirmation workflow marks the invoice paid. A reactivation sequence resumes while a support workflow is handling an unresolved complaint. Each workflow follows its own rules, yet the customer receives contradictory treatment.
To prevent conflicting automation workflows, assign authority over each shared business state, then define what every competing workflow must do when that authority changes the state. The choices should be explicit: wait for a condition, yield to another owner, stop the current action, or request human review. Recheck those rules before consequential actions, not only when a workflow starts.
This is a business-coordination problem: who is entitled to decide what happens next? The guidance below covers those decisions, rather than database locking, integration concurrency, or shared-inbox collisions.
1. Find the decisions that more than one workflow can make
Start with shared decisions rather than a list of automations. A workflow inventory tells you what exists; a decision inventory reveals where workflows can disagree.
Look for states that determine whether someone gets contacted, who owns the next step, or whether work proceeds. Typical candidates include payment status, opportunity status, service readiness, outreach eligibility, and an active human-review hold.
For each state, ask:
- Which workflows can change it directly?
- Which workflows act as though they have changed it, even without updating a field?
- Which queued actions depend on an earlier version of it?
- Which person can override it, and for how long?
The second question is easy to miss. A retention workflow may leave an account marked as closing but still send a renewal offer. It has contradicted the closure decision through its behaviour, even if it never touched the account status.
Also look for unrelated meanings compressed into one field. A customer can have a paid invoice, an unresolved complaint, and no permission for promotional outreach at the same time. A single active/inactive status cannot represent those distinctions reliably. Separate the decisions before assigning precedence.
2. Give each shared state an authority, not just a priority number
A priority number says which workflow wins. Authority explains why it wins, what evidence it needs, and where its power ends.
For every shared state, record an accountable business owner, the evidence accepted for a transition, the workflows permitted to make that transition, and any authorised override. A billing workflow might own reminder eligibility after a confirmed payment event. A sales workflow should not reverse that decision merely because an opportunity remains open.
Do not give one workflow universal precedence. Authority should follow the decision. A support workflow can place a promotional-outreach hold during complaint handling without gaining authority to mark an invoice paid. A payment workflow can stop reminders without closing a complaint.
Use the same restraint for human overrides. Define the affected state, the reason, the owner, and the condition that ends the override. An indefinite manual hold with no accountable owner can strand work as effectively as a broken automation.
If authority cannot be established, the safe response depends on the action. A nonessential offer can stop. A disputed payment classification may need review. Silence should not silently become permission to make a consequential change.
3. Choose between wait, yield, stop, and review
These responses are not interchangeable. Naming the difference prevents a paused workflow from unexpectedly returning as the owner.
- Wait: Keep the intention pending because a defined prerequisite is unresolved. Specify a resume condition, a review deadline, and what happens if neither arrives.
- Yield: Withdraw the competing intention because another workflow now owns the next step. Any later action requires a fresh eligibility decision, not continuation from an old queue position.
- Stop: End the current action or workflow instance because its purpose is no longer valid. Record the reason so retries do not recreate it.
- Request review: Suspend the disputed action and send a named decision-maker the conflicting evidence. State which unrelated actions may continue.
Waiting is appropriate when the original action may still be useful. Yielding is appropriate when responsibility has moved elsewhere. Stopping is appropriate when the underlying need has ended. Review is appropriate when evidence or authority is ambiguous.
Each choice has a cost. Waiting can delay service; yielding can lose a commercial opportunity; stopping can require a new workflow later; review consumes staff attention. Choose deliberately rather than defaulting every conflict into an ever-growing exception queue.
4. Use this filled hypothetical workflow-conflict matrix
Fictional worked example: A business coordinates sales follow-up, billing, onboarding, and support. The following matrix is a recommended operating design, not a description of controls available in a particular product. Each entry is a matrix row, including the observable result that proves the losing workflow actually yielded.
Shared state: reminder eligibility for an invoice
- Competing intentions: Collections wants to send a scheduled reminder; payment confirmation wants the invoice treated as paid.
- Authority and precedence: Billing owns payment classification. Confirmed payment for the same invoice outranks an earlier reminder schedule.
- Response: Stop the pending reminder for that invoice. A reversal or correction requires a new billing decision.
- Observable outcome: The reminder is recorded as stopped because the invoice is paid; it does not return on the next retry. Other unpaid invoices remain separately eligible.
Shared state: promotional-outreach hold
- Competing intentions: Reactivation wants to send an offer; support wants promotional contact paused while a complaint is handled.
- Authority and precedence: The support owner controls the complaint-related hold. The hold outranks campaign eligibility, but does not itself resolve recipient permission.
- Response: Reactivation yields. Support owns the next contact decision until the hold is explicitly resolved.
- Observable outcome: The queued offer is withdrawn with a support-hold reason. Closing the complaint triggers a fresh outreach assessment rather than releasing the old offer.
Shared state: onboarding readiness
- Competing intentions: A sales handoff wants onboarding to begin; onboarding sees an unconfirmed prerequisite.
- Authority and precedence: The onboarding owner determines readiness using the agreed prerequisite checklist. A sales-stage change alone is insufficient.
- Response: Wait for confirmation. In this hypothetical design, unresolved cases go to the onboarding owner after one business day.
- Observable outcome: The customer receives no start instructions prematurely. The record shows the missing prerequisite, the review deadline, and the person responsible.
Shared state: next commercial contact owner
- Competing intentions: A nurture sequence wants to send a general follow-up; an account owner has an active, recorded negotiation.
- Authority and precedence: The account owner controls commercial contact while the negotiation hold is active and has a documented next step.
- Response: Nurture yields. An expired hold goes to review rather than silently restoring the old sequence.
- Observable outcome: The generic follow-up is withdrawn; the account owner's next step remains visible, with a review date.
Shared state: outreach eligibility after an opt-out
- Competing intentions: A campaign wants to contact an eligible audience; a recipient has opted out of the relevant outreach.
- Authority and precedence: The recorded opt-out takes precedence over campaign membership. A new import or campaign selection cannot erase it.
- Response: Stop the affected outreach. If the scope or identity is unclear, request review without sending the disputed message.
- Observable outcome: The recipient is excluded with a visible reason, and rerunning audience selection does not restore the blocked action.
Shared state: account closure decision
- Competing intentions: Retention wants to resume offers; the account owner has accepted a closure request.
- Authority and precedence: The closure decision outranks retention eligibility. Conflicting evidence about whether closure was withdrawn goes to the account owner.
- Response: Stop retention outreach; request review before reversing the closure decision.
- Observable outcome: No offer is released from the old queue. Any reopening records the authorising decision and starts a fresh assessment.
The matrix is useful only if its outcomes can be inspected. A label saying paused is insufficient when the queued message remains eligible for release.
5. Recheck authority at the point of action
Eligibility at entry is not continuing authority. A workflow may enter a queue before a complaint, payment, opt-out, or manual hold exists.
Use a business-level action gate immediately before a consequential step. It should answer whether the intention remains valid, whether another owner has taken precedence, whether the evidence is current enough for the decision, and whether the action remains authorised. This describes a required operating rule; the implementation depends on the tools available.
Include queued work and retries in that rule. A retry should reconsider the present state rather than reproduce yesterday's decision. Where the tooling cannot reliably withdraw queued work, avoid scheduling far beyond the point at which eligibility can be checked, or route the action through an authorised review step.
For WhatsApp, keep campaign preparation, template review, and actual delivery separate. Recipient permission and opt-outs matter independently of internal workflow precedence. Approved templates are used to initiate Platform conversations and outside the 24-hour customer-service window, which opens or resets with a user message. Template approval does not guarantee delivery. Check the current WhatsApp Business Messaging Policy for precise requirements, and maintain a clear human escalation path.
6. Make review actionable and prevent endless handoffs
A conflict notification should contain enough information to decide, not merely announce that something failed. Include the shared state, competing intentions, evidence supporting each, actions already taken, pending external actions, proposed disposition, and accountable reviewer.
Define what continues during review. A disagreement about promotional outreach need not block internal support investigation. Conversely, sending the disputed offer while review is pending defeats the purpose of escalation.
Watch for circular waiting. If onboarding waits for sales approval while sales waits for onboarding readiness, neither workflow can progress. Assign one owner to resolve the prerequisite or specify a permitted exception. Do not solve the deadlock by letting whichever workflow wakes up first take control.
Review also needs an exit decision: resume under current conditions, stop, transfer ownership, or correct the evidence. Closing the review task without settling the pending intention leaves the conflict intact.
7. Recover from conflicts that have already escaped
When contradictory actions have already happened, first stop the affected pending actions within the authorised scope. Do not disable unrelated workflows unless the conflict reaches them too.
- Establish the current state. Identify the authoritative evidence and distinguish intended actions from actions actually completed.
- Resolve ownership. Assign a person to decide the next business step and any customer communication.
- Correct the queue. Withdraw stale intentions, resolve holds, and prevent retries from rebuilding the same action.
- Repair the rule. Update the matrix where authority, scope, or resume conditions were missing.
- Replay the scenario. Confirm that the losing workflow now produces the expected observable outcome.
An external action may not be reversible. Changing a record cannot unsend a reminder. Keep customer-facing recovery separate from internal state correction, and require appropriate authorisation for any follow-up.
8. Test one shared state before expanding
Choose one frequent or consequential conflict and complete its matrix row with the people who own both workflows. Check whether your workspace and current plan support the required holds, cancellations, evidence records, and review routing. Where they do not, establish a manual gate rather than assuming automatic enforcement exists.
Test both arrival orders: the authoritative event before the competing workflow starts, and after it has queued an action. Also test a retry, an expired hold, ambiguous evidence, and a human override.
Record whether the expected action was blocked, whether the reason was visible, whether a reviewer was assigned when needed, and whether work resumed only after fresh authorisation. Track contradictory actions that escaped, pending waits without owners, and repeated reviews of the same unresolved state. Fewer messages sent is not, by itself, evidence of better coordination.
Your next action is concrete: take one pending workflow conflict, fill its matrix row, and verify that the yielding workflow leaves the exact evidence you specified. Expand the design only after that outcome is observable.
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