Automation Concepts and Techniques
What you must understand
RPA delivery runs through a lifecycle: process discovery and prioritisation, a PDD (business view, written by the business analyst), an SDD (technical design), development, testing, deployment and monitoring in production. Know the delivery models: attended (triggered by a human, runs on that user's desktop, good for assisted tasks), unattended (triggered from Orchestrator by time or queue, no human present) and hybrid. Know that attended robots can still consume queues, and that human-in-the-loop steps (approvals, Action Center tasks) are a legitimate design choice, not a failure.
Good candidates are rule-based, high volume, stable-UI, digital-input processes with low exception rates. Poor candidates involve subjective judgement, unstructured paper input, or UIs that change weekly.
Traps that appear in questions
- Confusing PDD with SDD, or attributing both to the developer.
- Assuming a partially automatable process should be fully automated; the standard answer is to automate the deterministic portion and route the rest to a human.
- Ignoring maintenance cost or exception handling when judging ROI.
- Believing attended robots cannot use Orchestrator queues or assets (they can).
Rule of thumb
Score candidates with a pipeline metric: (volume x handling time x error rate) divided by (complexity x expected change frequency). Highest score goes first.
Worked example: an invoice team receives 5,000 invoices monthly; 80% follow three fixed templates, 20% are free-form scans. Automate the 80% deterministically with an unattended robot and route the remainder to a validation queue rather than attempting one all-encompassing automation.