Recent advances suggest that webcam eye-tracking can work for consumer research. But a simple in-house demo shows exactly how bad it can be. The demo also shows just how good Neurons' models are.

Market research is going through a data-driven transformation. Big data is being combined with click data, biometric, and other sources, and mined using powerful analysis tools. New tools are invented, allowing us an unprecedented insight into customer thought and behavior.

But not all methods are ready for prime time. For example, we have long followed the trends in more affordable eye-tracking solutions. One recent addition of such methods is webcam-based solutions for eye-tracking. The reasoning here is simple: we can easily see where people's eyes are during calls and recordings, so why not use this for reliable eye-tracking?

To test this, we have demoed many vendors of online/webcam eye-tracking solutions. We have tested these methods against high-end eye-tracking methods.

The truth? Webcam eye-tracking doesn't work very well!

The results are pretty staggering when we look at how it works. Below, you can see the results from one of the leading online vendors of webcam eye-tracking. Here, we tested people using standard to high-end webcam solutions with a PC-based webcam. To make matters as easy as possible, we tested fixation to a single cross, which we are using for calibration during our mobile eye-tracking tests on phones. Here, we used PCs to give the method the best possible way to track eye-movements.

‍

‍Results from a webcam-based eye-tracking study. Each green dot is a single person's pair of eyes. As you can see, the results are extremely noisy and do not reflect what should otherwise have been precise eye-tracking data.
‍

As you can see from the above data, the webcam-based eye-tracking looks almost like random variation around a central place. When the fixation cross moves, there is definitely a trend that the noise around the mean moves along with it. But the fixations generally seems to be lower than the cross.

By contrast, when we look at our predictive algorithm, it clearly demonstrates that it can track expected visual attention:

‍

‍The Neurons machine learning algorithm clearly demonstrates where the expected attention should be. This also reflects what happens during eye-tracking results to this and other types of stimuli.‍

Taken together, these results clearly show that webcam eye-tracking still has a long way to go. It's a simple demo that clearly shows that more work is needed before business decisions can be based on this method.

This also shows the power of attention prediction models such as both Neurons' new AI algorithm and the present visual salience algorithm. Both clearly demonstrate good visual prediction capabilities.

Moreover, using an automatic method has several other advantages, such as:

• No learning effects: when testing people, there is a clear learning effects if people are shown the same or similar items after each other. An automatic prediction algorithm is not affected by such learning. Because of this, you can test 100 design iterations and know which works best, without having to test 3000 people (if you have a sample size of 30 people per design, and 100 designs)
• Turnaround time like a charm: a Neurons analysis is in seconds, and a video analysis finishes in minutes. Eye-tracking research with a representative sample takes 1-2 weeks at least.
• Low cost appeal: eye-tracking studies, even online, can cost around $2,000 per image. An image analysis with Neurons will cost you around $5.
• Data-driven creativity boost: optimally, designers within any industry would love to have some indication as to whether their design will be seen they way they intend it to be. But these groups rarely have research budgets. Neurons offers a tool for allowing data-driven design, so that designers can try out many more options, and even try out risky designs, and gain better certainty that their design will work, before it actually is released.
• Shelf-design on the fly: retailers would love to know whether their signs are seen, or whether the intended planogram setup works as intended. Neurons allows them to test, change and optimize their design through direct Neurons feedback.
• E-commerce design that lead to checkout: one of the challenges in e-commerce web design is that many visitors leave without finalizing the purchase. By running Neurons analyses, web designers can secure that the visual path to checkout works as intended, and purchases are no more lost.

These and other aspects make Neurons the chosen tool for optimizing visual design. Want to try it out for yourself? Book a demo with us to explore all of Neurons' capabilities.