What is a company really measuring when it says it is measuring value?
On 26 August 2026, Meta reached a $17 billion settlement. Among other things, the agreement provides for a daily two-hour limit on Facebook and Instagram for users aged 13 to 17 (adjustable only by parents), a default night-time block, a ban on displaying like counts to under-18s, and the option for minors to enable a feed without algorithmic recommendations.
The mechanism
Meta has always measured the success of its products through metrics that calculate the time spent on the platform and the rate of interaction with content presented to the user by the algorithm. In economics, numbers of this kind are called proxies: substitute indicators that represent a quantity which cannot otherwise be measured directly.
As long as the proxy and the objective remain aligned, optimising the former means moving closer to the latter, and the system works. But organisations are made up of human beings, inclined towards laziness and shortcuts, and sooner or later someone learns to push the proxy up without increasing the figure that the proxy was meant to represent. At that point, proxy and objective are no longer aligned, but overlap..
This mechanism had been described in the literature long before social networks arrived, and most people remember it as Goodhart’s Law.
In 1975, the British economist Charles Goodhart observed that a statistical regularity tends to lose its validity as soon as it is used as a target for controlling a system. A year later, Donald T. Campbell expressed a similar principle in the field of social evaluation: the more a quantitative indicator is used to make decisions, the greater the risk that it will be manipulated and distort the reality it is supposed to measure.
The best-known formulation today — when a measure becomes a target, it ceases to be a good measure — is, however, a paraphrase popularised by the anthropologist Marilyn Strathern in her 1997 essay Improving Ratings: Audit in the British University System.
Cobras and current accounts
We do not know whether the story of the snakes is true or merely a useful anecdote for describing the principle, so let us take it with a pinch of salt.
To reduce the cobra population, the British colonial administration in Delhi offered a bounty for every dead specimen handed in. The initiative seemed a success: the number of cobras handed in was increasing satisfactorily. Another number, however, was also increasing: that of snake farms. It goes without saying that the government closed the programme and the breeders released the now worthless snakes: the cobra population ended up higher than it had been when the bounty was introduced.
The following two cases, by contrast, are fully documented.
Marks and bonuses
School marks are a good proxy for measuring how much pupils have learnt, but if they are used to allocate funding and bonuses, they can easily turn into a target, as happened in 2011 in Atlanta’s public schools, where dozens of teachers were correcting pupils’ answers in the tests used to allocate funding and bonuses.
Sales targets
In 2016, Wells Fargo was fined $185 million because employees, under pressure to meet sales targets imposed from above, had opened accounts and credit cards without customers’ knowledge. The number of products sold was meant to represent the quality of the banking relationship, and ended up representing nothing but itself.
How to recognise a proxy that has lost its way
Distinguishing a proxy that is still sound from one that has taken the place of the objective is difficult, and no one notices while it is happening.
Do pay, promotions or organisational survival depend directly on an indicator?
Is there a second indicator, managed by someone who has nothing to gain if the first rises?
If the real objective were to deteriorate while the proxy continued to rise, would the system notice?