More clicks did not necessarily mean more customers
Across the ten-week test, version F is the clearest candidate to advance: its purchase rate appears higher than the current page. Version C drew substantially more CTA clicks, but that attention did not translate into more purchases.
Recommendation
Advance F to a controlled follow-up rollout, not an immediate all-traffic replacement. The supplied daily results point to more purchases per visitor for F than for A, while C’s very high click-through rate is not matched by a purchase lift. Keep a randomized control during rollout and confirm the exact visitor-level comparison before making the change permanent.
Purchase rate is the decision metric
Approximate, rounded directional rates from the supplied daily counts. Purchase rate uses purchases ÷ visitors; small differences should be verified against exact totals before rollout.
A = current page (gray) · F = leading candidate (green) · C = click-heavy, weaker downstream result (amber). Bars share a zero baseline.
CTA clicks: C wins this step
Approximate clicks ÷ visitors. A click is useful only if it leads to a valuable outcome.
What the funnel suggests
C appears to attract many low-intent clicks. Its click-through rate is far above the rest, but a smaller share of its clickers proceed to add to cart, and its carts appear less likely to become purchases. The extra clicks do not show up as extra orders.
F looks more balanced. It generates more clicks than A and appears to convert a somewhat larger share of clickers into carts. Its purchase advantage is not just a top-of-funnel effect.
E is a useful reminder to follow the whole journey. It has fewer clicks than A, yet its purchase outcome appears broadly similar. Improving click volume alone is not a goal.
These funnel-step rates are descriptive, not causal explanations: visitors who click or add to cart are self-selected after assignment.
Variant summary
Rounded directional estimates, intended to compare the overall pattern—not to report precision.
| Version | Clicks per visitor | Purchases per visitor | Readout |
|---|---|---|---|
| A · Current | ~36% | ~1.85% | Baseline |
| B | ~36% | ~1.84% | No clear purchase improvement over A |
| C | ~50% | ~1.78% | Clicks do not carry through |
| D | ~36% | ~1.84% | No clear purchase improvement over A |
| E | ~32% | ~1.85% | Fewer clicks; purchase result near A |
| F | ~38% | ~2.13% | Leading candidate |
| G | ~36% | ~1.84% | No clear purchase improvement over A |
| H | ~34% | ~1.62% | Potential downside |
What this means for the decision
- F is worth a confirmatory test. Its approximate purchase rate is about 0.3 percentage points above A—roughly 15% relative lift. At around 225,000 visitors per version, that would be on the order of several hundred additional purchases if the difference holds. This is an illustration, not a forecast.
- Do not select C because it wins on clicks. Purchases per visitor are the more direct measure of the team’s goal; CTA engagement is a diagnostic metric.
- Keep A as a randomized holdout during the next rollout. Monitor purchase rate alongside revenue per visitor, average order value, refunds/cancellations, and any page-performance or customer-experience guardrails.
- Before declaring a winner, recompute exact totals and uncertainty. Use visitor-level denominators, compare each candidate with A on the preselected primary metric, and account for the seven candidate-versus-control comparisons. Do not treat daily rows as independent visitors.
Important precision note
The charts and percentages in this report are rounded directional estimates from the supplied daily table; they are not an exact aggregation or a formal significance analysis. The experiment spans 70 days and appears to include roughly 1.8 million visitor assignments overall, but exact totals, confidence intervals, and adjusted comparisons should be calculated from the source counts before a final ship decision. Also confirm that purchases are attributed consistently to the assigned version and that purchase-lag/censoring does not differ across variants.