Design, build and monitor one automation, end to end

The situation

Pick a real business and one repetitive task in it. You will score candidates, design the workflow including its failure branch before writing anything, transform twenty real messy records, add the monitoring that tells you when it silently stops, schedule it in the right timezone, and build it with a dry-run mode you read line by line before going live.

What you'll be able to show

• Score tasks on the four traits and choose one with a stated reason
• Write a workflow with trigger, steps, conditions, actions and a failure branch
• Add logging, alerting, dead-man detection and a holding place for inputs
• Build with a dry-run and run it against real data first

What you need

• Any automation tool you can use — Python, a no-code tool, a spreadsheet script
• Twenty real records from any source (anonymised)
• A scheduler, and a way to be alerted (email, WhatsApp to yourself)

Tasks

1. 1Score and choose
   List five repetitive tasks in the business. Score each on the four traits and estimate time saved per month. Choose one, and write one sentence per rejected task on why not.
   A correct result: A scored table, one choice, four reasons for not choosing the others.
2. 2The workflow, on paper
   Write the task as trigger, steps, conditions, actions, and what happens on failure. No code. If you cannot write the failure branch, stop and understand the process first.
   A correct result: A workflow document with an explicit failure branch.
3. 3Twenty records, transformed
   Take twenty real records. Write the transform that maps them to a new set of field names, normalising phone numbers and dates. Count how many needed cleaning.
   A correct result: Twenty transformed records and the count that needed cleaning.
4. 4Monitoring
   Add a log line per run, an alert on failure, a way to detect that it has NOT run at all (a dead-man check), and a holding place for inputs when the destination is unavailable. Unplug the destination and confirm nothing is lost.
   A correct result: Four monitoring pieces present; the unplugged test shows inputs held, not lost.
5. 5Schedule, in the right timezone
   Schedule it daily. Confirm from its own log that it fired at the intended local time — not five hours out. Break it on purpose and confirm the alert reaches you.
   A correct result: A log entry at the right local time and an alert received.
6. 6Build with a dry run
   Build the automation you designed, with a dry-run mode. Run the dry run against real data and read every line of what it says it would do. Only then run it live.
   A correct result: The dry-run output, annotated with anything you corrected before going live, and the first live run's log.

What to hand in

The scored table, the workflow document, the transform with its count, the monitoring evidence, the schedule proof, and the dry-run output with the live log.

How it is marked