Finding the drop
You have the flow. Now you need the numbers, and the data team's queue is eleven days long.
A funnel is a multiplication, which is why it is brutal
Funnel steps multiply rather than add. Four steps that each keep 80 percent of people keep 41 percent overall. Four steps that each keep 40 percent keep 2.5 percent.
Which means the step to fix is almost never the one with the smallest absolute number. It is the one with the worst rate, because that is the one leaking a proportion of everyone who reaches it.
A funnel locates loss; it does not explain the cause. Read the arithmetic correctly before choosing an intervention.
Use your chosen course project throughout. The additional examples below are fictional practice cases; transfer the method to your own evidence.
Keep the population consistent
Define who enters, the allowed time window and the order of events. A same-session funnel answers a different question from a seven-day funnel. If a step can be skipped, repeated or completed on another device, state how you count it before calculating any percentages.
Separate step rate and total conversion
The step rate divides completers by the people who reached the previous step. Overall conversion divides final completers by entrants. In a simple sequential funnel the overall rate is the product of the step rates. Improving a late step cannot recover people who already left earlier; calculate the expected absolute gain as well as the local rate.
Use the drop to choose the next investigation
Split the largest loss by provider, device or new-versus-returning users before redesigning a screen. A provider-specific failure suggests a different next move from a broad comprehension problem. Treat the split as a hypothesis check, and keep the original population definition attached to every chart.
Read the rates before the story
- Population: Who enters?
- Step rate: Kept from prior step
- Overall: Kept from the start
- Investigation: What explains the loss?
Change one step. Follow the effect.
Interactive teaching example · All numbers and tickets are fictional.
1,000 × 80% × 50% × 80% × 90% = 288.0 expected completions. Baseline: 288. Rounded counts are illustrative; all other rates stay fixed.
Doing it with AI, and where it breaks
Give a model your step counts and ask it to compute step to step rates, then identify which single step improvement would produce the largest absolute gain at the end of the funnel.
Models are unreliable at multi step arithmetic in prose and will confidently produce a wrong total. Make it write the formula, then compute it yourself in a sheet. Never let a chat window be your calculator.
Your AI workbench
Start with your own notes or clearly labelled practice data. Remove private details before sharing. Replace the placeholders, run the prompt in your chosen AI tool, and keep the output beside its source.
Audit this funnel table. Show entrants, step conversion, overall conversion and lost users with formulas. Flag missing denominators, inconsistent time windows and impossible increases. Calculate the result if only [named step] improves to [rate]. Do not invent a causal explanation for the drop. Data: [paste].
Before you use the output
- Recalculate one row manually
- verify the event order
- check whether the segment split changes your recommendation.
Stuck? Try this next
Ask for spreadsheet formulas with cell references and paste a tiny sample first. A plausible-looking chart is not evidence that the formula uses the correct denominator.
Keep a brief AI log: input used, useful output, what you checked, and what you rejected. The decision remains yours.
Build it
A funnel analysis with step rates and a ranked fix list.
Google Sheets
The worked solution
Stuck, or want to compare?
There is a worked solution. Try the build first. Reading it before you have attempted anything is the fastest way to learn nothing.
Checkpoint
If you did the build, these take two minutes. If you cannot answer one of them, that is the part to go back to.
Moving on marks this lesson complete. Finish the build first, it is the part that counts.