Find a recurring manual handoff
Look for places where people copy information between services, send routine reminders or reconcile the same data. Those handoffs are easier to describe and test than “automate the whole department”.
Ask an employee to show the actual workflow. Identify the trigger, the required information and where the output goes. Sometimes the obstacle is access or unclear operating rules rather than missing software.
Separate a rule from a judgement call
Write down the condition: when this event happens, perform that action. If the answer depends on negotiations or undocumented experience every time, clarify the rules first or leave the decision with a person.
Recording an agreed enquiry can be defined more precisely than automatically promising individual terms to a customer. A suitable first workflow has a clear boundary and an understood way to handle exceptions.
Check the data and access
Confirm that the required information is available through a reliable method. Review integration options, service limits, field formats and account permissions. A website or dashboard does not necessarily provide a suitable API.
Agree on the source of truth: which system owns the current status, and what happens when records disagree? Plan for repeated events as well. Two identical notifications should not silently create two unrelated tasks.
Agree how to assess the result
Describe an observable change: a record reaches the right system, an owner receives a task or a status survives the handoff. Compare this with the original workflow rather than treating an integration going live as the result.
Plan for failure in advance. Who notices an unavailable service, where can they find an unfinished action, and how can the work continue manually? The automation should remain understandable after the initial launch.
Choose a manageable first workflow
Compare candidates by how clear their rules are, whether data is accessible and what a mistake would cause. Start where the output can be checked and the change can be stopped if needed.
Before you begin
- A process owner and a recurring problem are identified.
- The trigger, action and expected output are defined.
- Data access and service limits are checked.
- Duplicate handling and failure recovery are understood.
- Acceptance criteria and a manual fallback are agreed.
You may not need to replace the tools. Connecting two existing services and removing one manual handoff can be a useful first step. Choose the next workflow from what you observe in operation, rather than from a list of technologies.