5 min read

Cohort Bounce

Cohort Bounce answers a question a single bounce-rate number never can: which acquisition cohort bounces hardest on which landing page? A site-wide "42% bounce" hides the fact that paid traffic to /pricing might bounce at 85% while organic to the same page bounces at 9%.

To open it, select your site from Domains, open the Retention tab, and switch the view to Bounce (alongside Retention and LTV).

Cohort Bounce is part of the Retention tab, a Pro feature. Free-tier teams see the Retention upgrade card. It is never shown on public/shared dashboards (the endpoint requires an authenticated session).

How to read the grid

The grid is shaped rows × columns:

Rows are your top 20 landing pages by sessions for the selected date range. A landing page is the first page of a session.
Columns are the cohort: either traffic channel or device (toggle top-right). At most 9 columns are shown, ordered by volume.
Each cell is the bounce rate for that (landing page × cohort) pair. Cell colour scales with bounce: pale = low, deep red = high (legend at the bottom).

An n/a cell means that page/cohort pair had no sessions in range. Hover any cell for the number of visits that started on that page in that cohort, the exact bounce %, and how far it deviates from that page's own average ("±X pts vs page avg").

LANDING PAGE        DIRECT   ORGANIC   PAID
/pricing             23%       9%      85%   ← paid bounces hard here
/                    50%      n/a      n/a
/blog/post-a         14%      n/a      n/a

Channel vs Device

The Channel / Device toggle re-pivots the same sessions:

Channel: Direct, Organic, Paid, Social, Referral, Email (whatever is present in range).
Device: Desktop, Mobile, Tablet (and any other device class present).

Switching the toggle re-fetches; the date range comes from the Retention tab's period selector.

Biggest bounce leaks

The right-hand rail is the grid's executive summary: the highest-bounce (page × cohort) pairs, ranked. To stay trustworthy it only ranks pairs with at least 5 sessions: a lone single-session "100%" is noise, not a leak. Each row shows the page, the cohort, the bounce %, the session count, and the delta from that page's average. Click any leak (or any grid cell) to drill straight into the matching sessions.

How bounce is defined

A bounce is a single-page visit: a visit (session) with exactly one page view. Time on page, scrolling, clicks and events do not change that. The grid is entry-based: a cell's bounce rate is the number of visits that started on that landing page in that cohort and viewed no other page, divided by all visits that started there. This is the same definition as the site-wide bounce rate on the Analytics tab and the bounce rate of each entry page on the Entry Page tab of the Pages card. For the same period and without filters, adding up the bounced visits and all visits over every landing page and cohort (not only the ones shown) gives exactly the site-wide bounce rate.

The grid is computed by one aggregation over your stored pageviews and visitor rows for the range, closed days and today alike, so a 30-day view returns in well under a second even on high-traffic sites.

Drill-down

Clicking a cell or a leak applies two filters, the landing page and the channel (or device), and switches you to the analytics view, pre-filtered to exactly those sessions so you can see who bounced.

Device drill-down is exact. Channel drill-down is approximate: the grid groups by the channel stored on each visitor, while the session filter re-derives channel from UTM/referrer/click-IDs. The two definitions agree the vast majority of the time but can differ by a few percent. The UI labels this on the grid.

Privacy

Cohort Bounce honours every Zenovay privacy commitment:

It is read-only aggregation of data you already collect. It adds no cookies, no localStorage, and no tracker code; data-cookieless="true" setups are unaffected.
The grid and rail only ever show counts and rates, never individual visitor data.
It introduces no new behavioural processing: GPC-respected ingestion upstream is unchanged.

Limitations

The grid caps at 20 landing pages × 9 cohort columns. Lower-traffic pages and rarer cohorts fall below the cut to keep the view legible and the query fast.
The leaks rail needs ≥5 sessions in a pair before it will rank it.
Channel drill-down is approximate (see above). Device drill-down is exact.
New websites see data as soon as sessions exist for the range; there is no overnight wait.