The recent Color 22 conference had an interesting speaker: dr. Terry Wu, about neuromarketing. My LinkedIn feed was full of it! And neuromarketing is an exciting topic. E.g. emotions can influence our perceptions. Knowing you are tasting Coca-Cola makes it taste much better (yes, it does!). And color is, of course, also an important input for our brain. But don’t get overexcited… Let’s take a deeper look at the Google ’41 shades of blue’ experiment (which became the 50 shades of blue over time). It has supposedly increased the advertising revenue by 200 Million US$! However: when put into perspective, it’s less spectacular. And even more: the test was fundamentally flawed.
After seeing all these nice comments on his presentation, I contacted dr. Wu and he responded pretty fast. He also pointed me to his Ted Talk a few years ago. Fascinating! And I agree: neuromarketing is undoubtedly a thing, only last week I read a scientific report on how you can make cookies taste better: just state that it is a ‘new and improved’ formula. It’s that simple. Really! It’s something called ‘framing’.
What I also found intriguing, is the Google (close to) 50 shades of blue story. By finding ‘the right’ blue, they made 200.000.000 US$ more in advertising revenue. Sounds impressive.
So, I wanted to learn more about this test, e.g. how they executed it and their findings. With the Google offices filled with the best and brightest, this must be well-designed research. With rock-solid results.
Putting it into Perspective
The first thing I checked, the easy part, was that 200.000.000 US$ and, more specifically: what does that represent? It’s a lot of money, I would be happy with even a tiny fraction of that. But putting it in the Google perspective, it’s not that much. In 2009, the year of the study, Google had revenue from search advertising totaling 22,89 billion US$. So, the 200.000.000 US$ is even less than 1%. If the Google engineers had stated that their research showed that a different color would result in a less than 1% increase, you might have looked at it differently. That’s such a slight increase, you could argue other factors were influencing the results.
Next, I tried to find more information about that experiment, but it was hard to find. Except for one presentation, called Metrics Driven Design by Joshua Porter (you can also download a PDF-version with some annotations here). This one shows a bit more information, although not entirely accurate. Slide 5 shows a gradient from a certain green to blue and states that they had 41 buckets.
When looking at the next slide, it offers more information about the shades of blue. When looking at the different graphs, you can make up the different color codes used. And there isn’t that much green in it… (the numbers are very low: 00 to 33; please keep in mind that hex codes have 256 shades).
The color that was eventually selected was, depending on the source, #2200CC or #2200C1.
Suppose that the red stayed the same, I can create 35 shades of blue based on the green (00; 07; 11; 18; 22; 29; 33) and blues (aa; bb; cc; dd; ee). I put them into a Photoshop document, and the first thing that came to my mind was that it’s tough to see a difference between many of them. Changing the green (especially the low numbers) doesn’t make much difference to me (and I do have ‘normal’ vision). Changing the blue makes it lighter.
And these were blocks in a Photoshop document, so areas filled with the same color. What would that look like when using text? Here are seven variations, the text is the color code. Do you see a difference?
#2200aa; #2207aa; #2211aa; #2218aa; #2222aa; #2229aa ; #2233aa
Input known, but output unknown
But then came the ‘aha’ moment: these numbers represent the input, the color code that Google sends out to the users. But how do they know what color the monitor is showing? As clever as they are, they can not control the monitors used (unless it is a strictly controlled environment, which wasn’t the case). That monitor could be AdobeRGB capable and calibrated. But it could also be a less than sRGB, uncalibrated one (remember, the test was done in 2009, when many monitors weren’t even capable of representing the complete sRGB gamut). Or what about people that have adjusted the brightness of the monitor, which will result in a different color? So, what is the value of a color-related study if you don’t have control over the most critical element: the monitor?
This is the page I used for the picture above: https://www.insights4print.ceo/2200CC/ Check it out yourself! Open this page on different devices, like laptops, tablets, smartphones, monitors and see what that same input looks like on these different devices.
To make an analogy: this would be the same as print buyers sending out PDFs and only judging the print quality based on that PDF, not the actual print, the ink on the substrate. Would that be ok? And dear printer: make this analogy if you ever have a customer (or consultant) using this Google 41 shades of blue story to enforce you to print within an unrealistic margin.
So, again, we need to disqualify a color-related study that sounded like music in our ears… Please, don’t draw any color-related conclusions from this Google study!
This unknown, uncontrollable output is a technicality, dr. Terry Wu probably isn’t aware of it. But the engineers at Google should have known it.
BTW: if you look closely at the graphs, you can e.g. see that one is about the CTR (click-through rate). And what is the delta between the worst and the best? 0,15. Is this significant? If you use it as a multiplier for your turnover, it shows a nice figure. But is this really more than just random fluctuations? Please enlighten me!
When looking at these different colors, what might have made a difference is the legibility of the text. Some colors are easier (or more comfortable) to read than others, especially on a monitor. But that doesn’t necessarily translate into print and logos.
One more thing…
And one more thing. Not only Google did this kind of research, but Microsoft also did this, for their Bing search engine. Probably one year later. And their ’80 million US$ color’ is another blue… It’s #0044CC. Are Bing users so different from Google users that they prefer a different blue?
BTW: their 80 million US$ blue seems much less than Google’s 200 million US$ blue, but remember that the market share of Bing in 2010 was much smaller than Google. So the increase in revenue of the Microsoft blue was, percentage-wise, much higher.
Why is this important?
We all want to feel important, we want to feel that our job matters. And in print, that translates to print quality, to color. So we love to hear stories that our products can make a big difference. Unfortunately, this desire can disable our critical thinking, as in Google’s 41 shades of blue story. People from the printing industry should have seen the flaw on the output side: Google had no control over the color shown on the monitors. Therefore turning the results useless.
Now for the record: I’m not saying print isn’t important. I love print. I love that the people in the printing industry are so passionate about their job and their work. But we need to get real about the alleged importance of slight color differences. I haven’t seen solid proof yet that these matter. The (in)famous Loyola study is on ‘business graphs’ (spreadsheets). Studies showing that color enhances brand recognition are on black & white advertising versus color. And now the Google study on the perfect blue has a serious technical flaw in it, turning it into a waste of time.
PS: now that we discussed that you don’t have control over what the participants see when doing color-related experiments online, I must tell you something. Something I’ve known from the start, but nobody (or maybe 1 person) addressed until now. The Coca-Cola Iconic Color Memory test that I have on my blog. I have no control over the visualization of the images. That’s why I also included the question if people were looking at it on a calibrated screen… Did it make a difference? There was a slight redistribution of the results, but the right one was still not the most popular.