Why only measuring the first order effects of an intervention can backfire

Decision-Making

In 1963, the UC Santa Barbara ecologist and economist Garrett Hardin proposed his First Law of Ecology: “You can never merely do one thing.” We operate in a world of multiple, overlapping connections, like a web, with many significant, yet obscure and unpredictable, relationships. He developed second-order thinking into a tool, showing that if you don’t consider “the effects of the effects,” you can’t really claim to be doing any thinking at all.

When it comes to the overuse of antibiotics in meat, the first-order consequence is that the animals gain more weight per pound of food consumed, and thus there is profit for the farmer. Animals are sold by weight, so the less food you have to use to bulk them up, the more money you will make when you go to sell them.

The second-order effects, however, have many serious, negative consequences. The bacteria that survive this continued antibiotic exposure are antibiotic resistant. That means that the agricultural industry, when using these antibiotics as bulking agents, is allowing mass numbers of drug-resistant

A retailer makes returns free. Orders leap. Somebody updates a slide to green.

Meanwhile, customers start ordering three sizes, the warehouse becomes a boomerang factory and support spends Tuesday locating refunds. The intervention has “worked”. It has also quietly mugged three other departments in the car park.

This is the seduction of first-order thinking: the person who presses the button gets the applause; the person cleaning up six weeks later gets a capacity-planning meeting. Response times fall, repeat contacts rise. Deliveries accelerate, breakages follow. A discount attracts customers whose loyalty lasts precisely until the discount does.

The useful question is not merely “what happens next?” It is “who changes their behaviour once this becomes normal?” Free returns alter ordering. Sales targets alter selling. A road bypass alters where people live. Humans are not tins of beans. Change the shelf and we shuffle about.

You cannot model every consequence without becoming the sort of committee that commissions a 94-page risk assessment for moving the kettle. But consequences that change incentives, spread costs to somebody else or are difficult to reverse deserve more than a fortnight’s victory lap.

I once watched a team celebrate fewer support calls after hiding the telephone number. Brilliant. By that logic, removing the fire alarm reduces fires.

Keep the fast metric. Just chain it to the slower one it is most likely to damage: orders to returns, first replies to actual resolutions, speed to breakage, clicks to profit. Otherwise the dashboard is not measuring success. It is photographing the getaway car.