For several years now, we’ve been talking very regularly with our clients and prospects about a fairly well-known phenomenon in behavioral economics called Goodhart’s law. And usually, we lump it together, which is actually a mistake, with another phenomenon, the cobra effect.
Something to show off with at dinner parties
Goodhart’s law says “when a measure becomes a target, it ceases to be a good measure”. Put another way, when you set up performance indicators, Key Performance Indicators, the organization will pivot toward those indicators at the expense of quite a few logical considerations, and even of its overall performance. An extremely common example: the newsletter sign-up pop-ins we’ve had on just about every website for the past 10, 15, 20 years. They are a very reliable sign that there is, in all likelihood, a metric on the number of newsletter subscribers somewhere in that site’s marketing department. And the fact that it degrades the experience, that it most likely degrades part of how users perceive the site, hasn’t really been factored in. Because what matters is meeting that performance indicator. That’s Goodhart’s law.
Then there’s the complement, the cobra effect, which says, roughly, that a well-intentioned solution sometimes produces results that are often perverse. You try to get a good result, even if it produces negative side effects. The example that gave the effect its name comes from British India (opens in new window), a very long time ago (the anecdote has never been proven, but a very similar case, involving rats, is documented in Hanoi). When the British arrived, unlike the locals, they weren’t really used to coming across cobras, and they set up a reward policy for anyone who brought back cobra heads, as proof that a specimen had been killed, and that there would therefore be fewer of them and safety would be ensured. Fairly quickly, they realized that more and more heads were being brought in, until they noticed that cobra farms had been set up in response. And the day they decided to scrap the reward, the cobra farms closed, the cobras were released into the wild, and there ended up being even more of them than at the start.
Another example that often comes up: in Spain in 2010 (opens in new window), installing solar panels and producing electricity were also rewarded. Very large solar farms sprang up, produced very large amounts of electricity, and were very handsomely paid, until someone looked closely at the figures and noticed they were producing in the middle of the night, and that in fact it was generators producing that electricity, not the solar panels.
It’s a phenomenon that happens very often: you try to maximize a metric, and since the metric becomes a target, you put in place all sorts of things that very often go against common sense, against the common good, and even against the original objective, if you look at things a little more broadly.
There’s “index” in EcoIndex
For us, this is something we ran into very directly while redesigning our website. We had several objectives, including of course refreshing our identity, but also being more exemplary on things that matter to us, notably accessibility and eco-design. And it was on eco-design that we quickly ran into an absurdity.
We aim to have an eco-designed website. You can agree with the principle or not, and we know there are debates on the subject within the eco-design and RGESN community. We can understand the questions, but the fact is that in 2026, unless you bring in a company like GreenSpector for a multifactor audit with measurements, the main eco-design indicator for any website is the EcoIndex.
The EcoIndex is available through an online site. You submit a page from your site, it churns for a few minutes, and it produces a score out of 100, translated into a letter from A to G. It gives an indicator of the page weight, in megabytes (the page needs to be light), of the page’s simplicity, where it looks directly at the DOM, and finally of external requests. It’s not a black box at all: the weighting of these indicators gives the score between 0 and 100. The higher it is, the more you show you’re able to be exemplary on the subject, which was exactly what we wanted.
What we wanted came up against an imperative: we still have many clients who tell us that making things accessible or eco-designed means making them ugly, static, degrading the user experience. There was the imperative of showing our expertise on the subject, but also of living out what we claim to sell our clients: experience, ease of use, design, aesthetics, motion. The goal of this website was really to show that it’s possible to be usable, well designed, offering an experience that isn’t ugly, and at the same time accessible and eco-designed.
We decided to do the front-end development ourselves, well, when we say ourselves, it was with the help of AI, and when it comes to environmental issues, we know AI is a problem. So over the past few months we found ourselves iterating dozens, maybe even hundreds of times on some pages, optimizing them component by component, design pattern by design pattern, submitting the page to the EcoIndex calculator countless times, only to see that sometimes you gain a little on one side and lose a little on the other. A very concrete example, SVGs: is it better to have an inline SVG, which avoids a request with the DOM unchanged but makes the page a touch heavier, or an external SVG, which counts as a resource and therefore lowers the score a bit? You end up with this kind of somewhat absurd trade-off, and with the cobra effect we’re subject to.
The double bind: do better without doing less
To reach the respectable score of EcoIndex B, 78 out of 100 on the home page, it’s quite likely that we consumed several tens of millions of tokens. When you set that against our traffic, there’s the makings of a huge contradiction here. We wanted to show that you could build something eco-designed without it being degraded, and we most likely didn’t build something eco-designed if you take a holistic view of the subject. We realize it, and everyone involved in this subject realizes it.
Maybe the EcoIndex, or the next digital sobriety indicator, will have to factor in the cost of building and designing, and websites will have to publish a token consumption statement on their site, just as they present the RGESN criteria they meet?
The illustrations in this article were generated by AI (see our AI use statement).
