The bounce rate that wasn't
Fernwood Yoga Studio has one location, 900 members and a new website with an online booking system. The owner reads her dashboard on the first of each month: homepage bounce rate 71%, average session duration 38 seconds, 2.1% of sessions convert to a booking, and 46% of conversions are attributed to "direct". She sends a monthly email newsletter but has never added tracking tags to it. The booking system sits on a subdomain and no tracking is installed there. She has concluded that "the homepage is bad" and wants to pay for a rewrite.
1. Which of the owner's conclusions does the data support, and which does it not? Be specific about what each metric can and cannot tell you.
Read the worked answer
Not supported: "the homepage is bad".
- A 71% bounce rate is not automatically bad. If the homepage is doing its job - giving opening hours, address and a booking button - many visitors will get what they need and leave without a second page view. For a single-location studio with strong local intent, a high bounce rate can indicate success.
- 38 seconds average session duration may be understated: with the booking system untracked, the most engaged visitors leave the measured site when they click through to book, and their time on the booking subdomain is invisible.
- 2.1% conversion is measured only on the main site, so it excludes every booking completed on the subdomain. The true rate is unknown and probably higher.
Supported: the measurement setup is inadequate for the decision being made. That is the real finding.
Specifically unsupported as a diagnosis: nothing in these numbers isolates the homepage as the cause. There is no comparison page, no funnel step, and no segmentation by new versus returning or by device.
2. What must be fixed in measurement before any redesign money is spent, and in what order?
Read the worked answer
1. Track the booking subdomain. Cross-domain measurement must be configured so a visit that starts on the main site and finishes in the booking system is counted as one session with one conversion. Without this, every conversion number is wrong and every channel looks worse than it is.
2. Tag the newsletter. Add tracking parameters to every email link with a consistent scheme (source, medium, campaign). This alone will reallocate a chunk of the 46% "direct" bucket to email and change the channel picture completely.
3. Define conversions explicitly. Booking completed, enquiry form, phone click, directions click - each as a separate event. "Conversion" as a single blurred number cannot guide a redesign.
4. Segment before judging. New vs returning, mobile vs desktop, and where the session started. The homepage may be fine for returning members and poor for first-time local searchers; a single average hides both.
5. Only then run a structured review of the homepage against those segments, ideally with a small usability test of five to eight real visitors.
Rationale: a redesign decided on broken data may destroy a page that is working, and will not be measurable afterwards either.
3. The owner wants a one-page monthly report. What should it contain, and what should be removed?
Read the worked answer
Keep - tied to a decision she can act on:
- Bookings completed (main site plus subdomain, combined).
- Bookings by source, with email properly tagged: email, organic search, paid, social, direct.
- New vs returning bookers - is marketing bringing in new members or only serving existing ones?
- Mobile vs desktop booking completion rate, since the mobile experience is a likely weak point.
- Enquiry form submissions and phone/directions clicks, as leading indicators for a local studio.
Remove or demote:
- Homepage bounce rate as a headline figure - it cannot be judged without context and invites exactly the wrong conclusion.
- Average session duration - distorted by the untracked subdomain.
- Total sessions and page views - vanity numbers for a 900-member studio.
Add one line of commentary: what changed this month and what will be tested next month. A report nobody can act on is a report nobody reads, which is how the original problem went unnoticed for so long.