All resources

A practical resource

One workflow: purpose, evidence and next decision

A practical worksheet for agreeing one AI workflow trial, its owner, boundaries, baseline, stop conditions and next decision.

Alongside People · 7 min read ·

Alongside People · One-workflow worksheet · 3 October 2026

Use this worksheet with the people doing a recurring job and the person relying on its result. Complete one copy per workflow. Record what you know, where the evidence is and what remains unknown. There is no maturity score or certification. Human approval can remain part of every consequential step; choosing simpler automation or less AI is a valid outcome.

This Alongside People discussion aid accompanies From AI experiments to dependable work. It does not provide permission, policy approval, legal assurance or procurement eligibility. Keep sensitive data, personal details and credentials out of this sheet; use appropriately controlled references to evidence instead. Fictional practice examples must be marked as fictional.

Begin with one decision

For a first conversation, answer these eight questions together. Use the full sections below when you are ready to prepare a trial. Unknowns are useful; do not fill them with confident guesses.

QuestionYour record
What recurring job needs to improve?
Who owns the outcome and can decide?
Which information and actions are permitted?
What makes the current work difficult?
What would a useful result look like for its recipient?
Which failure would make the trial unacceptable?
Which current examples will provide a fair comparison?
What is the next decision, and what evidence will support it?

Fictional worked example: A delivery coordinator prepares a weekly operations brief. The manager owns its acceptance; the tool may read an approved non-personal update collection and create a draft, but cannot send or change commitments. The team wants fewer missing details without more total checking work. An invented commitment stops the trial. They compare recent briefs of similar scope and keep every source and correction. They will decide whether to repeat the test after examining the agreed sample, including a conflicting update. No benefit has yet been observed.

Use an evidence reference such as “TRIAL-01 / example-03 / source-check / 3 October 2026” to locate the observation in an approved record system. Keep personal details and source contents in that system rather than copying them into this worksheet.

1. Describe the job

QuestionYour record
Workflow and version/date
Outcome owner and authority to decide
Person doing the work; person relying on it
What triggers the work? What is the finished result?
What decision or action does the result support?
What happens now, including handoffs and waiting?
What should be easier? What must remain dependable?
Which decisions and actions remain human-owned?
Could clearer information, removing a step or ordinary automation help?

Attach or reference one representative process map. Ask the recipient what they still have to find out. Record the answer here: __________.

2. Identify information and permission

QuestionRecord or evidence reference
Information sources, dates and person responsible for currency
Information allowed in the proposed tool; tool/edition/environment
Source of that permission and approver
Actions allowed; prohibited actions; approval before action
Who can grant access, change scope or pause use?
Run time/cost limits; who receives held work or an alert?
Applicable organisation/sector rules; responsible reviewers
Unresolved information, privacy, security, records or procurement questions

Permission decision: ready for the agreed bounded trial / revise / defer. Decision-maker, reason and date: __________. If approval is unresolved, do not submit information to a tool to complete the exercise. For government work, identify the actual organisation and its rules rather than assuming a common policy for agencies, councils and Commonwealth bodies.

3. Set a testable question and baseline

Question: If we change __________, will __________ improve for __________ while __________ remains dependable?

Primary outcome and exact definition: __________.

Example set, period and reason they are representative: __________. Include incomplete, conflicting or difficult inputs. Known sample limits: __________.

MeasureCurrent process: observation and sourceProposed change: observation and source
Primary outcome
Quality: errors, omissions, recipient usefulness
End-to-end elapsed time, including waiting
Active preparation effort
Human review, correction and handoff effort
Setup, training, operating and software cost
Exceptions, rework and recovery burden
Experience of people doing and receiving work

Use comparable units and conditions. Separate observed figures from estimates; name currencies where relevant. Do not assume time saved becomes reduced expenditure. Preserve the baseline and criteria. Explain changed conditions rather than quietly replacing the comparison.

Acceptance criteria agreed before testing: __________.

Stop conditions and who may stop: __________.

Manual route to finish the job after stopping: __________.

4. Review the evidence for each stage

These are discussion states, not a compulsory progression. Mark each item evidence available, unknown, or needs work, and record the evidence reference. Do not total the marks.

StageEntry questionEvidence for the next decisionYour observations and evidence
UnderstandIs there a recurring job or proposed use?Process map, owner, users, baseline, constraints, testable question
TryAre question, permissions, comparison and stop conditions agreed?Representative results, difficult cases, errors, checking effort, owner decision
RepeatDoes the trial justify further work, with an operational owner?Consistent instructions, checks, permissions, exception records, fallback, baseline comparison with sample limits
OperateHas routine use in the agreed scope been authorised?Owner/backup, access and change arrangements, visible failures, escalation, exercised recovery, reviewed use
ImproveIs there enough routine-use evidence for review?Recorded retain/revise/expand/retire decision, supporting observations and next trigger

What remains human-owned at the current scope? __________.

Who checks the result, what do they check, and when do they step in? __________.

Failure signals observed or anticipated: __________.

Pause/revert route tested, date and outcome: __________.

What would invalidate the current evidence or permission, including a model, tool, source or scope change? __________.

How will an error already distributed be corrected, and who owns that action? __________.

5. Make a decision

Choose one: continue in current scope / revise and retest / seek missing approval / use a simpler approach / pause / retire. Record the reason and evidence, including disappointing findings: __________.

ActionOwnerDue date or review triggerRequired evidence or permission
Next useful change
Missing evidence to collect
Operating responsibility and backup
Next outcome review

Outcome owner's decision, name/role and date: __________.

Where is the operating guidance kept, and can another person use it? __________.

Broader delegation needs its own purpose, permissions, checks and evidence. An earlier trial does not authorise it. A technically working result does not establish operating readiness if people cannot recover the job or maintain it.

If you want help

The Alongside People team provides practical AI advice and hands-on implementation. Paul Volpato is the consultancy’s founder and CEO. A conversation can begin with this worksheet, one representative example and the people responsible for the work. Possible next steps are an opportunity/process map, a bounded experiment, an agreed working improvement with handover, or operating practices for related work. Scope, delivery and ongoing responsibilities are agreed separately. Tell us what you're working through.

How we can help put an idea to work

Let’s talk it through

What have you tried?
What’s next?

Tell us what looks promising, what’s getting in the way, or what you’re still trying to work out. A few sentences is plenty to begin.

Let’s talk