Why this matters
A funnel tells you where people leave, and almost every team then works on the wrong step: the one with the lowest rate or the biggest count. The funnel analysis tutorial gives the method for finding the step that matters; the explorations tutorial covers building the funnel in GA4Google's event-based analytics platform.. This exercise asks you to apply both to one set of figures and to stop at a hypothesis rather than a redesign.
The checkout data
Ember and Ash Homeware is an invented online shop selling lamps, throws and tableware. Its tracking was verified last
month: each checkout event fires once, at the right moment, on every template, and purchase reconciles with the
order system. The analyst exported a closed, indirectly followed funnel for four full weeks.
Ember and Ash checkout funnel, four weeks, users
| Step (event) | All devices | Mobile | Desktop |
|---|---|---|---|
Product view (view_item) | 60,000 | 39,000 | 21,000 |
Add to basket (add_to_cart) | 7,200 | 4,500 | 2,700 |
Begin checkout (begin_checkout) | 4,000 | 2,500 | 1,500 |
Delivery entered (add_shipping_info) | 3,000 | 1,750 | 1,250 |
Payment entered (add_payment_info) | 2,100 | 1,050 | 1,050 |
Purchase (purchase) | 1,850 | 920 | 930 |
The delivery-to-payment step, all devices, by session default channel group:
| Channel | Delivery entered | Payment entered | Share of these users on mobile |
|---|---|---|---|
| Paid social | 900 | 540 | about 75% |
| Paid search | 850 | 680 | about 45% |
| Organic search | 700 | 525 | about 45% |
| 550 | 355 | about 70% |
Other evidence: elapsed time on the delivery-to-payment step is roughly twice as long on mobile; the next action panel shows most mobile leavers at that step returning to the basket page. On mobile the order summary is collapsed, so the delivery charge first appears in the total on the payment step. On desktop it is visible in a sidebar from the start of checkout.
Building the funnel
Specify the funnel before you read it, the way the tutorial asks: steps named for the journey, each mapped to one event, closed so that step rates read as a sequence, indirectly followed because shoppers move back and forth, with the purchase last. Read rates, not counts. Break down by device first and by source separately, because both at once splits the sample into cells too small to trust. Then do the arithmetic: for each candidate step, the extra purchases a realistic improvement would produce at the downstream rates.
Task
- Write the funnel specification: steps, events, open or closed, direct or indirect, date range and breakdown.
- Compute the step-to-step rates for all devices, mobile and desktop.
- Identify the step that loses the most users in absolute terms, then the step with the largest gap between mobile and desktop.
- Estimate the extra purchases if mobile matched desktop at each step with a gap, and choose the leak to work on.
- Read the source breakdown and decide whether it shows a separate problem or the same one.
- Write one hypothesis on the experiment hypothesis template, with a primary metric and guardrails.
Worked answer
The specification is the table above with a closed, indirect funnel over four full weeks, broken down by device category. The rates:
Step rates and the leak arithmetic
| Transition | All devices | Mobile | Desktop |
|---|---|---|---|
| View to basket | 12.0% | 11.5% | 12.9% |
| Basket to checkout | 55.6% | 55.6% | 55.6% |
| Checkout to delivery | 75.0% | 70.0% | 83.3% |
| Delivery to payment | 70.0% | 60.0% | 84.0% |
| Payment to purchase | 88.1% | 87.6% | 88.6% |
The largest absolute loss is view to basket: 52,800 users. That is normal browsing on a product catalogue, and the rate barely differs by device.
If mobile matched desktop:
- Delivery to payment: 1,750 × 84.0% = 1,470, so 420 more reach payment; × 87.6% gives about 368 more purchases.
- Checkout to delivery: 2,500 × 83.3% = 2,083, so 333 more; × 60.0% × 87.6% gives about 175 more purchases.
- View to basket: 39,000 × 12.9% = 5,014, so 514 more; through the remaining mobile rates, about 105 more purchases.
Delivery to payment on mobile is the leak: about 368 purchases over four weeks against 1,850 in total.
By source, paid social keeps 60.0% and email 64.5% at this step, against 80.0% for paid search and 75.0% for organic search. The weak sources are the ones with the most mobile traffic, so the source gap is mostly the device gap seen from another angle.
The hypothesis, in the template's form:
- Evidence: the mobile step rate is far below desktop on delivery to payment, the transition takes longer on mobile, and leavers go back to the basket.
- Problem: mobile shoppers meet the delivery charge for the first time on the payment step, and go back to reconsider.
- Hypothesis: if the mobile checkout shows the delivery charge in the collapsed order summary from the delivery step, then the mobile delivery-to-payment rate will rise, because the charge will no longer surprise people at the point of paying.
- Primary metric: the mobile delivery-to-payment step rate, judged in an A/B TestControlled comparison of two variants..
- Guardrails: overall purchase conversion rate (CVRShare of visits or clicks that convert.) and average order value.
The funnel is a location, not a cause. The evidence here makes the delivery charge a strong candidate; recordings of the leavers' segment or a short poll on the payment step would strengthen it before the test is built.
Common mistakes
- Working on the biggest count. The first step always loses the most people; the arithmetic decides the leak.
- Reading the source table as a new finding. Check the device mix before blaming the channel.
- Breaking down by device and source at once. The cells become too small to read.
- Writing a redesign instead of a hypothesis. Name one change, one mechanism and one primary metric.
- Rebuilding the exploration every week. If the step rate needs watching after the change, put it on a monitoring page designed with the dashboard design guide and keep the exploration as the analysis behind it.
The measurement plan and dashboard capstone brings this level together, and the intermediate analytics level shows where you are.
