Skip to content

Heatmaps & scroll depth

Roitrack can measure what happens on a hosted lander: where people tap, how far they scroll, and whether the page was ever actually looked at. It works on landers hosted on your tracker — nothing to add to the page, and nothing to install.

It is off by default. Nothing is collected until you turn it on for a campaign.

Turning it on

Open the campaign, find Lander tracking, and set a percentage above zero.

That percentage is the share of the campaign’s lander views that report back. It sits on the campaign rather than the lander because that is where the volume is: a popunder campaign can measure 1% of its views while a push campaign on the same lander measures all of them. Reports weight each measured view by its own rate, so the two still average correctly side by side.

A heatmap is a density map, so a small sample draws the same picture at a fraction of the cost. But see how much to measure before you rely on the percentage columns for decisions — the heatmap and the numbers need very different sample sizes.

Bots are never measured. The measuring script is not sent to them at all, which keeps the data clean and saves the bandwidth.

Reading the heatmap

Go to Landers, select one, and press Heatmap.

The lander is rendered live inside a phone, tablet or desktop frame, with the heat drawn over it. It is the real page, not a screenshot, so it can never drift out of step with the clicks shown on it.

Device tabs

Three tabs — Mobile, Tablet and Desktop — each with the number of views it holds. All three are always shown, so an empty one reads as nobody having arrived on that device rather than as a broken report, and the page opens on the busiest.

Each tab holds every screen width in its class and draws them together. That takes a correction, because a tap is recorded as a position across the visitor’s own screen and the same position means different things at different widths: on a lander whose content is a centred column, a tap on the column’s left edge is 14% across a 737px screen and 4% across a 390px one. Drawing one straight onto the other puts it out on the empty margin.

So before anything is drawn, the tracker measures your lander’s own layout at each width its visitors used, and moves every click to where the same part of the page sits on the width the tab is replayed at. Vertical positions are scaled the same way, since the page is taller on a narrow screen than on a wide one.

What the correction cannot undo is a layout that rearranges itself. A lander that moves a sidebar beside its column above 900px is not the same page above and below that line, and a click from one side of it can only be placed approximately on the other. Landers built as a single column — which is most of them — are placed exactly.

A fourth tab, Unknown, appears only when some views reported no screen size at all.

Controls

  • Heat power — how bright the map is. Sparse data is invisible at 1×, so push it up. The map normalises to the 95th-percentile cell rather than the single hottest one, so one visitor hammering one pixel cannot flatten everything else.
  • Spot size — the radius of each click blob.
  • Zoom — Fit, or a fixed percentage if you want to inspect one part of the page closely.

Scroll depth

The panel beside the heatmap shows the distribution, not an average. That matters: a mean depth of 3% reads identically whether everyone scrolled a little or almost nobody scrolled and a few read to the end. Those are opposite conclusions.

The numbers above the map

Meaning
LP viewsRedirects to this lander
MeasuredViews that reported back
Ever visibleMeasured views that were actually on screen at some point
Reached offerVisible views where the CTA was clicked
Avg visible timeTime the page spent in front of someone, averaged over views that were ever visible
Clicks mappedTaps drawn on the heatmap

Measured is normally lower than LP views, and the gap is information. An LP view is counted the moment the click is redirected, before the browser has fetched anything. A view that never reports back is one where the page never rendered: closed instantly, blocked, or a visitor who was never really there.

Finding bad zones

Three columns are available in any report, next to LP views and LP CTR. Add them from Settings → Columns.

ColumnQuestion it answers
LP Loaded%Measured ÷ LP views — did the page ever render?
LP Visible%Ever visible ÷ measured — did anyone look at it?
LP Scroll%Average scroll depth, among views that were visible
LP MeasuredThe sample size behind the other three

LP Loaded% is usually the one worth acting on. A zone where 95% of views never produce a measurement is a zone selling you traffic that does not arrive, and no other column in the tracker shows it.

A worked example, grouped by sub1:

ZoneLP viewsLP MeasuredLP Loaded%LP Visible%LP Scroll%
zone_good605795.0%91.2%79.4%
zone_meh603660.0%33.3%24.5%
zone_dead6035.0%66.7%30.5%
zone_none6000.0%——

Note zone_dead. Its LP Visible% of 66.7% is higher than zone_meh’s — but it is built on three measured views. LP Loaded% of 5% is what tells the truth. This is why LP Measured sits beside the ratios, and why you should sort by LP Loaded% before anything else.

zone_none shows 0.0% loaded and dashes for the other two: sixty views, not one of them rendered. The dashes mean unknown, not zero — with nothing measured there is nothing to say about scrolling.

Then blacklist them

Group the report by your zone sub, sort by LP Loaded%, mark the bad rows, and press Copy values. Paste the list into the flow’s Sub conditions with the not in operator, on the same sub.

How much to measure

The heatmap and the columns need very different samples.

  • For the heatmap, a few percent is plenty. It is a picture of density and it looks the same.
  • For blacklisting on the percentage columns, measure at 100%. A zone needs roughly 400 measured views before its percentages are stable to within a few points. At 1% sampling that means 40,000 clicks in that one zone before the number is worth acting on; at 100% it is 400.

Measuring at 100% costs very little. Each measured view sends a single small request when the visitor leaves the page, and only if something happened — a view nobody touched is about 400 bytes. A tracker handling a million lander views a day is sending about a dozen of these per second.

Limits worth knowing

The columns cover the last 14 days. The detailed per-view data is kept for two weeks and then discarded; the heatmap itself is kept indefinitely. A zone blacklist is a decision about recent traffic, so this is rarely a constraint in practice.

The columns work when grouping by sub1–sub20, campaign, flow, offer, lander or day. Grouping by country, OS or device shows no engagement figures rather than wrong ones.

Only hosted landers are measured. An externally hosted lander is a page Roitrack does not serve, so there is nothing to measure it with.

Scroll depth is a property of the page as much as the traffic. If a campaign rotates several landers of different lengths, compare zones within one lander rather than across them.

A visitor is measured once. If someone leaves the page, comes back and then clicks the CTA, that later activity is not reported.

There is no backfill. The data exists only from the moment you switch sampling on.