Compare candidate workflows
| Workflow | First useful version | Main control |
|---|---|---|
| Enquiry routing | Assign by selected service and area | A monitored queue for uncertain requests |
| Appointment preparation | Collect required details before a visit | Staff confirmation of missing information |
| Document intake | Extract candidate fields for review | Source references and exception handling |
| Overdue follow-up | Remind the responsible staff member | Stop when the enquiry status changes |
| Internal knowledge lookup | Retrieve approved guidance | Show sources and escalate uncertainty |
A simple rule can route a service dropdown without an AI model. Interpretation becomes relevant when the input is unstructured and the decision cannot be captured reliably by the existing rules.
Pick the first pilot deliberately
Score candidates using observed frequency, staff time, error cost, data readiness and recoverability. These are planning criteria, not a prediction of savings. Avoid selecting a workflow solely because a tool demonstrates it impressively.
For a hypothetical maintenance company, routing enquiries may be a better first pilot than an autonomous quoting agent. Routing has a narrower output and can fall back to a shared queue; a quote can create commercial commitments.
Use workflow automation and the automation basics guide to define the initial scope.
Write the acceptance criteria
Describe a normal input, an incomplete input, a duplicate and an unavailable integration. Specify what may be created or updated, which steps require approval and how staff can stop the workflow.
Record the original request and the resulting state. If an action's outcome is uncertain, check the destination before retrying. A notification about a failed connection does not prove that the destination received nothing.
Decide whether to expand
Compare completed work, review effort, errors and customer impact with the previous process. Expand only when the owner can explain the results and operate the fallback.
For CRM automation, use the CRM integration guide to plan ownership and handoffs between systems.
FAQ
Does every automation need an AI agent?
No. Use ordinary rules for clear, repeatable conditions. Add AI when interpretation is necessary and its output can be checked within the business's acceptable risk.
Can we calculate savings from a demonstration?
A demonstration can suggest potential, but savings require observed task time, review effort, exceptions and operating costs. Label any early calculation as a planning estimate.
What makes a pilot ready for real use?
It needs tested normal and failure paths, clear ownership, appropriate access and a working recovery method. The team must also know which outcomes still require review.
If your business wants to choose a first automation, get in touch with a recurring task and its current bottleneck.