Back to all articlesAutomation design

Choose the event that should start a workflow

Alactic Editorial10 min read

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

A record owner fixes a spelling mistake, and the workflow creates another handoff task. Someone adds an internal note, and it starts again. The problem is not that the record changed. It is that the workflow treats every change as evidence that new work is required.

Choose a creation event when the work belongs to the arrival of a new record. Choose a meaningful state transition when the work becomes appropriate only after a business condition changes. Choose a scheduled eligibility check when time passing, rather than an edit, makes the work due. In every case, separate the event that prompts evaluation from the evidence that permits a start.

The practical test is simple: can you explain why this event represents new work, why the record is eligible now, and why that work has not already been started? If not, the trigger definition is incomplete.

1. Define the unit of work before choosing its trigger

Start with a sentence describing the intended action, not the available event options. “Run when the record changes” describes system activity. “Create one onboarding handoff for each approved implementation cycle” describes business work.

A useful trigger contract has five parts:

  • Purpose: the specific work the workflow should initiate.
  • Candidate event: the occurrence that asks whether the work should begin.
  • Start evidence: the facts that must be true before starting.
  • Work identity: what makes this a distinct piece of work rather than a repeat.
  • Hold or rejection reason: what happens when the evidence is missing or contradictory.

Work identity matters because a record and a task are not necessarily one-to-one. A customer might have several approved projects. Each project may justify a new handoff, while repeated edits to one project do not. Define the identity at the level of the business obligation: for example, project plus approval cycle, rather than customer alone.

Then ask what actually makes that obligation exist. Is it arrival, approval, or reaching a due time? That answer should drive trigger selection.

2. Use creation only when arrival is the business event

A creation trigger is a good fit when every new record requires initial work and the record already contains enough evidence to do it. Initial triage, assignment for review, or a data-quality inspection can fit this pattern. The work concerns the arrival itself, not a later decision.

Creation is a poor substitute for readiness. A newly created record may be a draft, an incomplete import, or a placeholder awaiting confirmation. Starting a handoff at that point moves uncertainty downstream: the receiving team must discover which details are absent and whether approval ever happened.

Decide what should happen to an incomplete new record. If the creation workflow only performs an intake check, it can create an exception for a person to resolve. If its purpose is to start approved work, it should not proceed without approval evidence.

There is also a recovery question: if a record fails the creation-time check and becomes eligible later, what will reconsider it? Creation happens only once. Without a separate readiness transition or deliberate review path, that record can remain stranded.

3. Use a state transition when readiness changes

A transition trigger represents movement from a non-eligible state into an eligible one. It is not simply a test that the record currently has an eligible status.

For a hypothetical handoff process, the meaningful transition might be from “Awaiting approval” to “Ready for handoff.” An edit to the address while the record remains “Ready for handoff” is not another transition. Neither is saving the same status again.

Define readiness as a business condition, not merely a label. A record might need an assigned owner, a confirmed scope, and recorded approval as well as the status. If the label changes before those facts exist, the start should be held.

This creates an important design choice. You can make “Ready” a controlled state that is entered only after all prerequisites are complete, or you can define entry into eligibility as the moment the complete set of conditions becomes true. Either approach can work. What fails is watching only the status change while allowing missing prerequisites to be added later without any reconsideration path.

A reversal also needs a rule. Moving out of readiness and back into it does not necessarily create a new obligation. Require evidence of a genuinely new cycle before treating re-entry as new work.

4. Use a scheduled check when time makes work due

Some work becomes necessary without anyone editing the record. A review date arrives. An unresolved item reaches its follow-up date. A temporary hold expires. A creation or edit trigger cannot reliably represent those moments because no edit is required.

A scheduled eligibility check asks which records are due now and still qualify. Its start evidence should include the due time, the unresolved condition, any hold or cancellation state, and whether the work for that due occurrence has already started.

The trade-off is timing. A scheduled check introduces a gap between becoming due and being considered. Choose that interval from the business tolerance, not from a preference for frequent automation. Also define the relevant time zone and whether overdue work should be picked up after a missed check.

Repeated discovery is normal here. The same unresolved record may appear in several checks. That should not create several tasks for the same due occurrence. A new start needs a new obligation, such as a separately defined follow-up cycle, rather than another scan of the same facts.

5. Worked example: a filled trigger decision table

Fictional worked example: A service business wants to create one onboarding handoff for each approved project cycle. The handoff requires an assigned implementation owner, confirmed scope, and recorded approval. Separately, unresolved handoffs need a review when their review date arrives. All details below are hypothetical recommended design choices, not product capabilities or customer results.

The decision table is presented as filled rows so each option includes the accepted event, rejected events, and the reason for the choice.

Row A — Record creation

  • Proposed work: inspect a new project record for intake completeness.
  • Accepted event: a new project record is created and is not marked as a duplicate.
  • Required evidence: an identifiable project and a responsible intake reviewer.
  • Rejected events: an existing record receives a new note; an owner edits contact details; a duplicate record is imported.
  • Reason: arrival justifies an intake inspection, but unrelated edits and duplicate arrivals do not create new intake work.
  • Choice for onboarding handoff: reject creation as the handoff trigger because creation does not establish approval.
  • Recovery: missing intake ownership goes to an exception review rather than silently starting a handoff.

Row B — Entry into an eligible state

  • Proposed work: create the onboarding handoff for the approved project cycle.
  • Accepted event: the project enters readiness with approval recorded, scope confirmed, and an implementation owner assigned.
  • Required evidence: a previously non-eligible condition, current readiness, the approval-cycle identity, and no existing handoff start for that cycle.
  • Rejected events: a spelling correction while ready; another save of the ready status; a readiness label without approval; leaving and re-entering readiness within an already-started cycle.
  • Reason: the readiness transition establishes the obligation. Cosmetic changes and status toggles do not establish another one.
  • Choice for onboarding handoff: use this as the primary trigger.
  • Recovery: hold inconsistent records for review; reconsider them when the missing evidence is supplied through the defined readiness path.

Row C — Scheduled eligibility check

  • Proposed work: create a review task for an unresolved handoff whose review date has arrived.
  • Accepted event: a scheduled check finds a due, unresolved handoff with no review task already started for that review occurrence.
  • Required evidence: a valid review date and time zone, unresolved status, no active hold, and an unused review-occurrence identity.
  • Rejected events: the due date is still in the future; the handoff is complete; a hold is active; an earlier check already started the same review.
  • Reason: time creates the review obligation, but repeat checks do not create additional obligations.
  • Choice for onboarding handoff: use for due reviews, not as a replacement for the approval transition.
  • Recovery: after a missed check, consider overdue unresolved items rather than only items due at the exact next check time.

The result is not one broad “record updated” workflow. It is three distinct responsibilities: intake inspection, approved handoff, and due review. Keeping those purposes separate makes both rejection and recovery easier to explain.

6. Make the start decision inspectable

For the hypothetical handoff, the start decision should be explainable from a compact evidence record:

  • Candidate: this project entered readiness for this approval cycle.
  • Readiness: approval, scope confirmation, and ownership are present.
  • Authority: the workflow is authorised to initiate this action for this record.
  • Exclusions: the project is not cancelled, held, or identified as a duplicate.
  • Prior work: no handoff has already started for the same cycle.
  • Decision: start, hold for missing evidence, or reject as ineligible or already started.

Treat unknown evidence differently from confirmed eligibility. If nobody can establish whether approval exists, the workflow should not infer approval from a convenient status label. Hold the start and identify the person responsible for resolving the uncertainty.

Eligibility also needs to remain true when the action begins. A record may be cancelled after it was first considered. The recommended business rule is to recheck the relevant conditions before committing the work, rather than treating an earlier evaluation as permanent permission.

Record a reason specific enough to support correction. “Missing implementation owner” tells an operator what to fix. “Not eligible” alone does not.

7. Define recovery without turning every edit into a retry

A narrow trigger can miss work if the process has no recovery route. A broad trigger can create repeated work if every edit becomes a retry. Resolve that tension explicitly.

Use different handling for different outcomes:

  • Missing prerequisite: assign a correction owner and define how the completed evidence will be reconsidered.
  • Already started: refer to the existing work; do not create another task merely because someone requested a retry.
  • Uncertain outcome: check whether the work exists before authorising another attempt.
  • Cancelled or held: preserve the stop decision until an authorised business change removes it.
  • Missed eligibility: use a deliberate reconciliation review to find eligible obligations without matching work.

Reconciliation is a safety net, not permission to leave the main trigger vague. It should apply the same start evidence and work-identity rules. Otherwise the recovery process can create the duplicates the primary workflow was designed to prevent.

Historical records deserve a separate decision too. Enabling a new rule should not silently imply that every previously eligible record requires action. Define the intended historical scope, review it, and authorise any catch-up work deliberately.

8. Test rejected events before enabling starts

Validate the decision rule without initiating downstream work first. Use a labelled hypothetical test set or a reviewed set of real records with appropriate access. Include ordinary edits, incomplete readiness, repeat readiness, cancellations, overdue items, and records whose work already exists.

For each case, write the expected decision before evaluating the rule. Then compare the result with the expectation. Do not test only the happy path: the original problem lives in events that should do nothing.

Your next action is to take one workflow currently triggered by any edit and produce four observable outputs:

  1. A one-sentence obligation and a defined work identity.
  2. A filled decision row naming accepted and rejected events.
  3. A start-evidence checklist with a recovery owner for missing evidence.
  4. A reviewed test log showing expected versus actual starts, holds, and rejections.

Measure unrelated edits that would have started work, valid obligations that were missed, and repeated starts for the same obligation. Keep these separate: reducing starts is not an improvement if legitimate work disappears with them. The rule is ready for an authorised rollout when the reviewed cases behave as intended and unresolved exceptions have an explicit owner.

Automation designWorkflow triggersCRM operationsProcess 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