Collect candidates from repeated handoffs.
Ask where people re-enter the same information, watch an inbox, rename documents, prepare a standard first draft or chase a predictable approval. Capture the trigger, frequency, average effort, systems involved and typical output.
Include the people doing the work. A task that appears simple from a management view may contain judgement, relationship context or informal exception handling that is not written down.
Score suitability before estimated savings.
Use a short comparison rather than choosing the task with the biggest headline cost. Evidence quality matters: measure a sample period or review recent cases instead of relying only on memory.
- Frequency and manual effort are high enough to observe.
- Inputs are available and reasonably consistent.
- The correct output can be reviewed against clear criteria.
- Exceptions are identifiable and can reach a responsible person.
- A process owner has time to test and approve changes.
Avoid a first pilot with hidden consequences.
Delay tasks involving irreversible decisions, sensitive data, unclear permissions or many unsupported integrations until the operating controls are mature. Customer-facing messages and financial actions need particularly clear approval and recovery rules.
A low-risk internal preparation step can still be valuable. Drafting, classification or structured extraction may reduce effort while leaving the accountable decision with a person.
Write a one-page pilot brief.
State the current process, included cases, excluded cases, sample data, human review, success measures and stop conditions. Name the owner and the users who will provide feedback. Record what remains manual.
The free assessment can turn this material into a bounded prototype plan. Production access, integrations and live processing are separately scoped after the team reviews feasibility and the evidence needed from a pilot.
Questions before you start
Questions before you start
Should we automate the task that takes the most time?
Not automatically. High effort matters, but unstable inputs, rare cases or serious consequences can make a task a poor first pilot. Balance value with testability and control.
How many processes should the first pilot include?
Usually one bounded journey. Multiple tightly related steps can be included when they share the same trigger, owner and acceptance criteria.
What if our process is not documented?
Use recent examples to map it. The mapping may reveal a simpler rule, form or responsibility change that should happen before automation.