A first AI test needs a defined task, suitable information, a person to check the output, appropriate data protection and a way to compare the result with today's process. Readiness belongs to the task you are testing; the whole business does not receive a single pass mark.
Use these five checks with the people who do the work. Each gap points to preparation you can tackle before starting. This is a planning aid, not a certification or a prediction of financial return.
1. Can you describe the workflow?
Choose one recurring job, such as drafting routine enquiry replies or extracting invoice details. Write down what starts it, what information it uses, who handles each step and what a finished result looks like. Ask the person doing the work to check your map. Make the task specific enough to test.
- Which part is repetitive, and which part needs human judgement?
- Who owns the process today and can approve a change?
- What happens when an input is missing or an answer is uncertain?
Next step if unclear: document one real example from start to finish before choosing a tool. The readiness guide shows how to shortlist pilot candidates.
2. Is the information usable and permitted?
Check whether the pilot's inputs are available, accurate enough for the task and lawfully usable in the proposed tool. Digital records can help, but a large historical dataset is not automatically required. A simple drafting assistant may need a current approved template; a forecasting tool may need reliable history. Confirm who may access the data and where it will be stored.
- Can the team find a small set of representative examples, including difficult cases?
- Do those examples contain personal, confidential or sensitive information?
- Have you checked the vendor's data use, retention, access and deletion terms?
The Office of the Australian Information Commissioner guidance on commercial AI products recommends due diligence and particular care with personal information. The Australian Cyber Security Centre's small business AI guidance covers tool policies, supplier checks and unreliable outputs. Review the guidance against your own obligations before uploading live records.
3. Who reviews the output and handles exceptions?
Nominate a human owner and define what they must check before an AI-assisted output reaches a customer or changes a record. Specify which decisions remain with a qualified person. Give staff a way to flag wrong, biased or unsafe results, and a way to pause the workflow. Consultation with the team can reveal exceptions that a process diagram misses.
Next step if unclear: try the task on non-sensitive examples with human review before connecting it to live systems.
4. Can you fund and support a small test?
Ask for a written proposal covering setup, software, connections, training, checking time, ongoing charges and how to stop. Allow the team time to test and make corrections. Agree a spending limit and a date to decide what happens next.
Next step if unclear: collect a scoped proposal and compare its full costs with the existing process. A trial can still be worthwhile if it improves quality or response time rather than reducing labour.
5. How will you tell whether the pilot helped?
Record the current staff effort, corrections, turnaround time, complaints or exceptions that matter for the task. Test comparable cases and count any extra checking AI creates. Agree in advance what would justify keeping, changing or stopping the project.
For example, if the pilot drafts replies, compare similar enquiry types and measure both drafting time and corrections before sending. A faster first draft is not a saving if staff spend longer fixing it. Keep the reviewer and escalation path in place while testing.
How to interpret your answers
Choose the next action from the gaps you found
- No defined task: map one workflow and its exceptions.
- Data or access concerns: use safe sample data and resolve permissions before live use.
- No reviewer: assign an owner and approval step.
- No baseline or budget: record current effort and request a bounded proposal.
- All checks addressed: run a limited pilot and review the evidence before scaling.
These are planning prompts, not a numerical readiness score. A high-stakes task may need much more testing and oversight than a low-risk internal draft, even when the same boxes appear to be checked.
Turn the checklist into a first project
- Select one low-risk, recurring task with a clear owner.
- Document the current steps, exceptions and baseline.
- Choose a tool only after checking data handling and access.
- Test representative cases with human review and record corrections.
- Compare the trial with the baseline, including costs and reviewer effort. Continue, revise or stop based on the result.
Need help shaping the scope? Read the readiness guide and common questions, or speak with Flowtivity about a possible first workflow. A conversation can identify questions to investigate; results and costs depend on your systems and trial.