Define the task you are improving.
“Use more AI” does not tell you what should improve. Name a task or decision: prepare a response, complete a document review, answer a staff question, or forecast demand for a planning process.
Define the start and finish. Include the people who provide inputs, review outputs, resolve exceptions, and maintain the system. These boundaries determine what you measure.
Record a useful baseline.
Collect representative examples of the current workflow before changing it. Record hands-on time, waiting time, correction work, and the quality of the finished result where those measures are relevant.
A baseline should include ordinary and difficult cases. Comparing an easy AI example with an average manual case can make a project look more useful than it is.
Count the whole process.
Evaluate preparation, review, correction, and follow-up together. A quickly generated answer can still create more work if a person must repeatedly reconstruct its sources or fix its assumptions.
- Compare the same task and similar inputs.
- Include the time required to verify the output.
- Track exceptions and unsuccessful attempts.
- Separate implementation cost from ongoing operating cost.
- Check whether people actually use the new workflow.
Distinguish capacity from cash savings.
Time released is capacity. It becomes a cash saving only when an expense actually changes. It can also create value by enabling more work, faster service, or better coverage—but that value needs its own evidence.
Use the evidence to decide what comes next.
Set a review point and a continuation criterion before expanding. Depending on what you learn, the next step may be broader use, a narrower scope, better source material, more testing, or a pause.
For predictive models, include error costs and a meaningful baseline. A better model score alone does not establish a better business result.
Explore WAIV’s delivery approach or use the starting-point planner to identify a candidate workflow and measures.