The retrospective baseline worksheet

For measuring an AI initiative you deployed without recording a baseline. Six steps, one initiative at a time. You set every value on this page yourself, including the verdict. Nothing here is scored or decided for you.

Everything you type stays in your browser. Nothing on this page is stored, sent anywhere, or seen by Revenue Experts. Closing the tab clears it, so download the CSV before you leave.

Step 1. Name the task, not the tool

One sentence describing the recurring piece of work. "Quoting a custom order," not "the AI assistant." If the tool touches several processes, run the worksheet once for each. Which tool or initiative

Step 2. Set the two windows

Use the same calendar window one year before deployment, not the weeks immediately before it. The weeks immediately before are usually why the tool was bought, and measuring against them inflates the result.
Before window
After window
Which systems hold your before period records Tick the ones you have checked and found usable. Three of the five is normally enough.

Step 3. The metric, before and after

One metric, from one source system, measured the same way in both windows. Cycle time in days, items processed, hours of meeting time, or supplier spend per month. Metric and source system
Before value
After value
Raw difference, in your own words

Step 4. Rival explanations, written down before you decide anything

Mark each one present or absent, then write what it could explain on its own. Fill this in before you interpret the difference above. A list written afterwards is a list of things you have already dismissed.
Surviving benefit after subtracting the rivals, generously Give the rivals every benefit of the doubt. What is left is a floor, not an estimate.

Step 5. The full run cost, not the licence

Annual figures in your own currency. The last two lines are the ones normally missing from the budget.
Licence or subscription
Implementation, amortised
Internal maintenance time
Mediation time
Total appears here once you enter numbers. It is arithmetic on the four figures you typed, nothing else.

Step 6. Your verdict

Three permitted answers. You choose. Cannot be determined is legitimate, and it carries a required action so that it does not become a way of avoiding the question.

If you have more than three initiatives, or would rather not run it yourself

The AI ROI, AEO and Data Activation Assessment applies this across your whole program, using records from your own systems rather than your rating of yourself, and returns a written finding on what your AI spend is producing. If it finds nothing you did not already know, you do not pay for it.

Book a fit conversation