A blank worksheet · one workflow
AI workflow
decision record.
Map the work, compare the options and decide what to measure. This is a planning aid, not a project proposal, a promised deliverable or a prediction of results.
By Sajad & Hassan, Ampliflow
01
Name the outcome
- What should happen when the work is done?
- Who needs the result, and who owns the decision?
- What currently prevents that outcome?
02
Follow the work
- What starts the process, and what information does it need?
- Who handles each step and the handover?
- Where does work wait, repeat or lose context? What happens when information or a person is missing?
03
Record a starting point
- Record the measurement period and source.
- Record work received and completed, time spent, outstanding work and errors.
- What is not measured or remains uncertain? Keep units and periods comparable.
04
Compare realistic options
- Compare a process change, a feature in existing software, a fixed rule, AI assistance, a new build and leaving the process unchanged.
- For each option, record evidence, constraints, running costs and the smallest test that could change your mind.
05
Define a bounded test
- Check normal examples and missing, ambiguous or conflicting inputs.
- Which permissions and actions need human approval? What happens on a failed step, duplicate request or unavailable system?
- How will repeated outputs be checked? What evidence would make you continue, change direction or stop?
06
Decide who carries it forward
- Name the decision owner and people affected.
- Record the access, implementation, training and support still to agree, and who reviews the result and when.
- Compare the result with the starting point. What changed, what did not, and what remains uncertain?
Measure the business outcome separately from the quality of an AI answer. Record observations before claiming an improvement.