Measurability beat alignment

A steam-engine salesman in 1775 had a sharper business model than your consultancy does. Boulton and Watt did not sell you an engine. They installed it and took a third of the coal you saved by running it instead of the old Newcomen pump. The machine was, in effect, free; you paid them out of your own savings, for as long as the savings lasted. Their reward scaled with your benefit, and it kept scaling over time. We have spent two centuries getting worse at this.

The deals that shared the upside

The pattern shows up everywhere once you look for it, and it is always the same shape: structure the deal so your reward scales with the value you create, and ideally so it keeps paying for as long as the value lasts.

Rolls-Royce still run theirs. They do not sell you an engine, they sell you "power by the hour" and then sweat to keep your aircraft flying, because their revenue is your uptime.

The advertising commission was the same trick in a different suit: fifteen percent of the media spend meant a campaign that ran for a decade paid the agency for a decade. So you were rewarded in exact proportion to how right you had been, and you were paid for the idea not only in the quarter you had it but for as long as it kept working.

Josiah Wedgwood, two hundred and sixty years ago, was already running money-back guarantees, free delivery, self-service showrooms and royal endorsements. None of these were primitive merchants we have since surpassed. They had outcome pricing, risk-sharing and skin in the game figured out while we were still a century from the spreadsheet.

Boulton and Watt did not sell a machine. They sold a share of the future, and collected it one winter's coal at a time.

Then we installed the meter

Steve Jobs ran straight into the modern version of this and drew the right lesson from it. When he bought the computer graphics group out of Lucasfilm in 1986 and turned it into Pixar, the only thing the technology could make at first was thirty- and sixty-second films, so its clients were advertisers, and Pixar made some astonishing ads.

Then Jobs looked at the business model and called it what it was: the client might make a hundred million in brand value out of the spot, and Pixar was paid for its time and the cost of the stock. The upside walked straight out of the door.

So he stopped renting the genius by the hour and started owning the asset. Pixar would make its own films and keep the intellectual property, and Jobs took an executive-producer credit on Toy Story and a share of everything it went on to earn. It is the same move as taking a third of the coal: stop selling the machine once, start owning a piece of what it produces forever.

And then, mostly on purpose, we dismantled all of it. Procurement looked at the commission and the value-based fee and said, not unreasonably, I cannot audit that, I cannot control that, prove to me the advert is what sold the cars. So it pushed everyone onto retainers and day rates, which are legible, defensible, and sever the link between reward and value completely. The eighteenth century had risk-sharing and outcome pricing working, and we replaced it with timesheets because a timesheet fits in a spreadsheet.

We did not pick the better instrument. We picked the auditable one. Measurability beat alignment.

Tight and loose fitness functions

There is a clean way to name what we traded away. Rory Sutherland takes it from a remark by Stephen Wolfram, who observed that evolution works precisely because it runs a "quite loose fitness function": survive long enough to reproduce and you stay in the game, and that slack is what produces the whole riot of biodiversity. Sutherland turns it on the modern organisation, which by contrast is in thrall to a tight fitness function. The difference between the two is the difference between a meter that rewards the outcome and one that rewards the proxy.

An average-speed camera is a loose fitness function. It states the outcome it wants, do not average more than the limit between these two points, and then stays silent on the method. You keep the judgement: pull out to fifty-five for ten seconds to clear a lorry, drop to thirty on the patch of black ice where even the limit would be reckless.

A fixed camera is tight. It measures the proxy at a single instant; fifty-six miles an hour now is a fine, regardless of the lorry, the ambulance, or the ice. It is context-blind by construction.

The loose function is not a fuzzy version of the tight one. It is a categorically better instrument, because the slack between the measure and the goal is exactly where judgement, craft and adaptation live. The tight function measures the proxy so precisely that it punishes the very discretion the rule existed to protect. You optimise the map and bulldoze the territory, which is the same failure as steering by a dashboard.

The day rate is the fixed camera. It measures the input with brutal precision and is therefore perfectly indifferent to whether anything good happens at the other end. Worse, it inverts the incentive, because under hourly billing inefficiency is revenue. The better aligned you would want your supplier to be, the worse the meter serves you.

Goodhart is not a metric you can fix

Pull the same idea down one level, to how you measure the people inside a company, and it becomes Goodhart's law. The moment you measure the individual, you force every person to manufacture a legible personal reason to exist. So the work fragments into whatever happens to be measurable to me, and the connective tissue (covering for a colleague, the unglamorous integration work, the favour that lifts someone else's number) becomes invisible and therefore unrewarded and therefore starved.

Which is why you measure the team and not the person. The team can play to its own internal shape instead of the org chart's fiction that value decomposes neatly into person-sized parcels. Greg Jackson runs Octopus Energy on soft indicators rather than hard KPIs for exactly this reason: the moment you attach a bonus you need a metric to make it fair, and the moment you have a metric people optimise the metric and let everything it cannot see rot. He did not go looking for better metrics. He refused to install the meter.

Goodhart is not a bug that better metric design fixes. It is a property of attaching a reward to a measurement.

Stephen Wolfram gives the deepest version of why the shortcut can never work. The behaviour of a rich system is computationally irreducible: there is no formula that reads the answer off the inputs, and the only way to find out what a thing is worth is to run it and watch. The tight fitness function is the seductive lie that you can skip the running and price the inputs instead. You cannot, which is precisely why measuring harder makes firms worse, not better.

The defence, and the next mistake

The defence is not a cleverer metric. It is to keep the fitness function loose enough that gaming it and doing the job well are the same activity:

• Specify the outcome, not the method. Say what good means and stay silent on how, so the judgement stays with whoever is close enough to use it.
• Share the upside. Tie the reward to the value created so your supplier's interest and yours point the same way, the way Boulton and Watt's did.
• Measure the team, never the individual. Let the unit play to its own shape instead of forcing each person to defend a personal number.
• Price for value over time, not for hours. Pay in proportion to how right the work turns out to be, for as long as it keeps being right.

Which brings us to the mistake we are about to make again, at scale. We finally have a technology whose value is wildly variable and unusually easy to attribute, and we are pricing it by the meter: pay per token, pay per seat, charged the same whether the output was worth a hundred million or nothing at all.

The Boulton and Watt move is sitting right there and almost nobody is making it: let the model take a share of what it saves or earns, over time, so its price tracks its value. Instead we are reaching for the timesheet again, because the timesheet is easy to bill. Measurability is about to beat alignment one more time, unless we decide, just this once, to charge for the coal we save.