Automation
Choosing workflows to automate
Map the work, identify exceptions, and decide which actions need human review before connecting your tools.
← GuidesIn this piece
Repetition makes a task worth examining. It does not automatically make the task ready for automation. A repeated process can depend on undocumented judgment, missing information, or a person who quietly repairs errors before anyone else sees them.
For an owner or operations manager, the first decision is which workflow deserves a closer look. This guide provides a selection method and a small operating specification you can use before choosing tools.
Map a real instance from beginning to end
Follow one recent item through the process. Record what triggered it, which information arrived, who acted, what they changed, and how the next person knew to continue. Ask about the last awkward case as well as the ordinary one.
An illustrative home-service enquiry might move from a website form to a shared inbox, then into a scheduling tool. The hidden work could be deciding whether the request is in the service area, detecting a duplicate, and waiting for photographs. Those decisions belong on the map.
Mark each step as one of three kinds:
- A rule: A known condition determines an action, such as checking a required field.
- An interpretation: Someone reads unstructured information and proposes a classification or summary.
- A consequential decision: Someone accepts work, makes a commitment, or changes access.
Use straightforward rules where they fit. Consider AI for interpretation only when you can review the result and manage errors. Keep consequential decisions under an explicitly agreed authority.
Choose candidates with a reasoned comparison
| Candidate property | What to ask | A reason to pause |
|---|---|---|
| Frequency | Does it recur enough to justify setup and maintenance? | Rare tasks with high setup effort |
| Consistency | Are inputs and rules reasonably stable? | Every case needs a different process |
| Verifiability | Can a person or rule recognize a wrong result? | A plausible answer cannot be checked |
| Recoverability | Can we undo or repair an action? | An error would create an irreversible commitment |
| Ownership | Who notices failures and resumes the work? | The workflow belongs to nobody |
Avoid multiplying these into a precise-looking return score without real measurements. First measure current effort and rework on a representative sample. Later compare the complete workflow, including review and maintenance.
Make exceptions part of the scope
For the enquiry example, a missing phone number should produce a visible incomplete record or a request for clarification. An unavailable scheduling service should create a retryable failure, not a success message. A second delivery of the same event should not create a second customer.
Duplicate deliveries are a real integration concern. Stripe, for example, documents duplicate webhook events and recommends recording processed event identifiers. That is an example from one provider; verify the delivery behavior of each system you connect. Stripe webhook documentation.
A named human review step also needs a workable interface. Show the source, proposed action, reason it was flagged, and choices to approve, correct, or reject. A notification saying “check this” without the underlying information moves the burden rather than resolving it.
Write an operating card
Use the following fields for one candidate:
- Trigger and authoritative source record.
- Required inputs and validation rules.
- Automated actions and actions reserved for a person.
- Duplicate detection and retry behavior.
- Failure destination, responsible owner, and response expectation.
- Manual fallback and the way to reconcile work completed during an outage.
- Evidence retained to explain what happened.
Keep credentials and sensitive data out of the card. It is a shared operating description, not a place to store secrets.
Match the work to the right service
The ValleyMen project includes workflows connecting a public program experience with member and admin software. The published scope demonstrates connected delivery; it does not prove savings for another business or establish that the same tools suit your process.
Explore AI Automation for existing workflows and tools →Opportunity mapping and an ongoing AI partnership are available within this service path. Existing package terms remain tied to their published scope. Integrations, data access, review requirements, and maintenance needs determine a custom estimate.
If the work requires a new portal or product, use Application Development. If people can already do the work inside an approved assistant but need a repeatable method, start with AI Training.
The lead-handoff worked example shows how the operating card becomes a concrete acceptance checklist.
Put this into practice
Bring the decision you’re working through.
A working session is a chance to review your goal, the tools or website you have, and the help you need. We’ll identify a sensible next step and discuss scope before any project commitment.
AI Automation →Book a working session