Start with the friction
Ask teams where work waits, gets copied or has to be checked twice. Look beyond the number of clicks. A short task repeated across many orders may cost more attention than a long task done once a month. Map the trigger, inputs, decision points, systems touched and expected result. That map reveals whether the process is stable enough to automate.
Separate rules from judgment
Routine validation, routing and data movement are good candidates. Ambiguous requests, unusual commercial decisions and sensitive approvals often still need a person. A useful workflow makes the handoff visible: the system handles clear cases and presents exceptions with the context needed to decide. Hiding exceptions inside an opaque flow usually creates more support work later.
Check the data first
Automation depends on reliable identifiers and clear ownership. If customer records differ between CRM and ERP, automatically moving them faster will not solve the disagreement. Decide which system owns each field, what happens when a value is missing, and how failed steps are retried. The design should make a failed order or document easy to inspect and recover.
Pilot one meaningful flow
Choose a bounded process such as incoming order triage or onboarding document checks. Record current effort and failure patterns before changing it. Release in stages, keep a human review path, and compare the result with the original problem. Expand only when the workflow is dependable and teams can explain how it behaves.