Marketers should stop relying on hover rate

For many, the promise of digital advertising is the wealth of information that comes with every impression they buy. However, all this data becomes virtually useless when marketers employ the wrong success metrics for their campaigns.

What use is it to know your click-through rates if more than half of those clicks are fraudulent? And how much do you really gain by optimizing for lower CPMs if your impressions are all showing up in non-viewable locations?

Unless marketers are able to judge the effectiveness of their online spend using metrics that are proven to affect business results, there’s no reason for them to trust digital media any more than they do radio or print.

One major area where we are seeing marketers and some adtech vendors making a huge mistake is placing value in metrics based on how people maneuver their cursors while browsing a web page. Although conventional wisdom suggests that someone is likely to be looking at an ad if their mouse is hovering over it, recent research suggests this is not the case.

Today’s consumers have vastly different browsing habits.

A study we conducted in September examined billions of impressions and found that marketers who optimize their campaigns using such mouse-based metrics actually increase the likelihood that they’re reaching a bot mimicking human behavior. Hover rates were 27 percent higher on impressions that were blocked, such as high brand-safety risk or fraud. Simply put, these kinds of statistics are woefully inadequate for determining the success of an online ad campaign, and they need to be abandoned.

Browsing habits have changed

Much of the trust our industry places in hover rates and heat maps dates back to a 2001 study conducted at Carnegie Mellon University, which suggested a “strong relationship” between a person’s cursor location and where they were looking on the page. As a result, the authors concluded, companies could use mouse-tracking as an inexpensive alternative to eye-tracking systems.

Nearly 15 years later, the paper is now very much out of date, particularly when it comes to how people browse the web. At the time of its publication, virtually everyone was accessing the Internet from a bulky desktop computer, and mobile phones had little functionality beyond making calls and the occasional game of Snake. Not only do today’s consumers have vastly different browsing habits, but we are now also able to collect a great deal more information on how they use the Internet.

It’s no surprise that user experience experts from both Google and Microsoft have since produced research directly contradicting the findings of the 2001 study. In a 2010 presentation in Belgium, Google’s Dr. Anne Aula explained that 42 percent of the search users the company studied had no active mouse movement as they attempted to complete a task on the page.

Further, of the remaining 58 percent of participants who did move their cursors during the study, the eye movements of just 32 percent matched their horizontal mouse swipes, while only 10 percent exhibited a relationship between their eye gaze and their vertical mouse movements.

Mouse movement ultimately is useless in terms of predicting whether a user has seen an ad.

As a team of scientists from Microsoft and the University of Washington concluded in a paper published in 2012, “Cursor movements, scrolling, and other client-side interactions are easy to collect at scale, which many Web analytics services offer to do. But claiming that the cursor approximates the gaze is misguided — as we have shown, this is often not the case depending on time and behavior.”

Time to move forward

Not only is mouse movement ultimately useless in terms of predicting whether a user has seen an ad, optimizing campaigns for hover rate is actively damaging to marketers’ efforts to reach their target audience. Just as black hats have learned to take advantage of KPIs like CPM and CPC, fraudsters have in recent years found ways to game the system by programming bots to wiggle a virtual mouse over the area where an ad is supposed to appear, thus fooling measurement vendors into thinking a real person saw the impression in question.

The September research project from our data science team actually found a statistically significant correlation between hover rate and ad fraud; hover rates were 48 percent higher when served to bots rather than humans. While many marketers see a relatively high hover rate as a sign that they’re running a campaign with great viewability, the truth is they are more likely being swindled by fraudulent actors.