How to use the first AI trial worksheet
A guide to defining the task, recording a comparison and explaining the decision made by the person responsible.
Define, record and decide
First define a task and how you will check its output. Consider a template, rule or formula without AI. Record practice, training and initial costs separately.
For evaluation, set aside cases of comparable difficulty and explain how they were selected, their order and their differences. Separate practice from evaluation. Alternating the order does not eliminate learning; where possible, assign comparable cases at random. Keep unsuccessful attempts and the work needed to finish them.
For each condition, record the minutes spent preparing, producing, reviewing and correcting. Record waiting separately. An empty field means the time has not been recorded; enter 0 only if you observed no work at that stage. Totals require all four fields to be valid and never include waiting or training.
The difference is AI-assisted work minus usual work. A negative value describes less work in that record; it does not establish causation, quality or cash savings. Review quality and failures before deciding.
Answers are not saved automatically or sent to a server. Use an HTML copy to reopen the worksheet with its data included, or import a JSON copy using “Open a copy”. Printing also lets you create a PDF from the browser. The downloaded copy contains what you entered; clearing the view does not delete those files.
What to check before recording times
- Confirm that you are allowed to use the information and that it is current. Keep a reference to its version.
- Define the output and quality standard with the reviewer. Record which failures prevent the output from being used.
- Consider an alternative without AI. A template or formula may handle the task with less preparation.
- Set aside practice and training separately. Do not combine them with the minutes recorded for evaluation cases.
How to record cases
Each record brings together a usual condition and an AI-assisted condition for a case or pair of cases established as comparable beforehand. Describe their difficulty, selection, order, participants, dates and tool version. If several people are involved, retain a record of how working minutes were obtained and avoid counting the same activity twice.
Waiting may overlap with another activity and does not automatically count as human work. Record it separately. Complete the observed quality and failures even if the attempt does not produce an accepted output.
How to read the totals
The total for each condition adds exactly four fields: preparation, production, review and correction. Negative and non-numeric values are rejected. An empty field leaves the total pending. The difference appears only when both totals are complete and valid.
The subtraction does not assess quality, initial costs or how the time is used afterwards. It does not estimate cash savings or the outcome of extending the tool across the company. Review those aspects before choosing a decision.
How to save and recover your work
- Use “Save a filled-in HTML copy” to download a worksheet with its data included. Open it as an HTML file or use “Open a copy” from another worksheet.
- The JSON export is a copy of the fields, rather than a diagnostic report. Restore it using “Open a copy”.
- Use the print button or save as PDF through your browser's print dialog. The printed PDF is not guaranteed to be fillable.
- The worksheet asks for confirmation before clearing or replacing a view that contains data. Copies saved on your device remain there until you delete them.
Answers are not transmitted, autosaved or automatically copied to the cloud. If you close the page without downloading or printing, you may lose what you entered. A saved HTML copy includes the quick guide so you can consult it without a separate guide file.