You cut prices to win back customers, and a competitor matches you within a week, so now everyone earns less for the same volume. You set a sales target to lift revenue, and within a quarter your best people are gaming the target instead of serving customers. Each decision was sensible at the first level. What went wrong lived one level down, in the reaction the decision provoked.

The quick version

  • First-order thinking asks "what happens?" Second-order thinking asks "and then what?", how will people and systems react to what I just did?
  • Most costly leadership mistakes are second-order: the plan worked exactly as designed, but the reaction to it didn't.
  • The whole tool is one habit, after you've decided, run two or three rounds of "and then what?" before you commit, especially for things that are hard to undo.
  • Go two or three orders deep, then stop. Past that, you're guessing, not reasoning.

The idea in depth

The plain version of this idea is old, and it comes from sociology, not management. In 1936 Robert K. Merton published The Unanticipated Consequences of Purposive Social Action in the American Sociological Review, the first systematic attempt to explain why deliberate actions so reliably produce effects nobody intended. Merton named the usual culprits: ignorance (you can't analyse what you don't know), error (you reason from a wrong model of how the world works), and what he called the "imperious immediacy of interest", the pull of the result you want right now, which crowds out attention to everything else the action will touch. Those three still describe almost every preventable mistake a leadership team makes.

The move, then, is to treat your plan as an input to other people's behaviour, not as the end of the story. A decision lands in a system of customers, competitors and colleagues who all respond, and their responses are where the real consequences live.

The investor Howard Marks gave this its sharpest practical form. In The Most Important Thing (2011) he distinguishes first-level thinking from second-level thinking. First level: "It's a good company; let's buy the stock." Second level: "It's a good company, but everyone already thinks so, it's overpriced, so let's sell." Marks's point is competitive: first-level thinkers reach the same obvious conclusion as everyone else, so they can't outperform. The edge comes from reasoning one move further than the room. That logic isn't only for investors. Any leader operating in a market, for customers, for talent, for attention, is playing the same game.

flowchart LR
    A(["Your decision"]) --> B("First order: the intended effect")
    B --> C("Second order: how people react to that effect")
    C --> D("Third order: how those reactions interact and compound")
    D --> E(["Net result, often not what you planned"])
Each order is a different question. Most planning stops at the first box. Leaders Loop

The most useful one-liner belongs to the ecologist Garrett Hardin, whose first law of human ecology was blunt: "You can never do merely one thing." Any change to a connected system has many effects, and you don't get to choose which ones arrive. This is where the systems literature earns its place. Donella Meadows, in Thinking in Systems (2008), explains the mechanics Hardin only gestured at: systems run on feedback loops and delays. A reinforcing loop amplifies your action; a balancing loop pushes back against it; and a delay means the reaction often shows up long after you've moved on and stopped watching. The classic case is building more roads to cut congestion, which lowers the cost of driving, draws more drivers, and refills the road. The intervention didn't fail. The system absorbed it and routed around it.

"You can never do merely one thing.", Garrett Hardin

So for any change, ask: what loop am I pushing on, and how long until it pushes back? Lower the cost of a behaviour you don't actually want more of, and you'll get more of it. This is the same instinct behind cross-disciplinary mental models, borrowing the feedback-loop lens from ecology because it explains your org chart better than your org chart does.

Where it breaks down

Honesty matters here, because this tool is easy to over-sell. Going deeper does not mean going further, the branches multiply faster than your knowledge. By the third order you are reasoning about reactions to reactions, and your confidence should drop accordingly. Treat distant effects as scenarios to watch for, not predictions to plan around.

There's a subtler failure too. Merton himself warned about the self-defeating prophecy, a forecast that, by being acted on, prevents itself. Second-order thinking can talk a team out of a good, reversible decision by conjuring vivid downstream disasters that were never likely. The fix is to match the depth of analysis to the cost of being wrong. For a cheap, reversible call, stop at first order and just try it. Save the "and then what?" passes for decisions that are expensive to unwind or that change what people are rewarded to do, because incentives are where second-order effects bite hardest.

A worked example

The cleanest real-world illustration is the Wells Fargo cross-selling scandal. The bank set an aggressive internal goal, eight financial products per customer, the so-called "Gr-eight" initiative, and tied compensation and job security to hitting it. First order: cross-selling numbers would rise. They did. Second order: when you pay people to open accounts and punish them for missing the target, some will open accounts that customers never asked for. In September 2016 the Consumer Financial Protection Bureau found employees had created more than 1.5 million unauthorised deposit accounts and over 500,000 unauthorised credit-card accounts; the bank paid $185 million in penalties and the reputational damage ran far higher. The target worked exactly as designed. The reaction to it is what cost them.

Now run the same situation through the habit before launching, with illustrative numbers to make it concrete. Suppose you're a sales director and you propose a bonus for every new account opened, say £50 a head, aiming for a 20% lift in new accounts.

flowchart TD
    A(["Pay £50 per new account opened"]) --> B("1st order: reps open more accounts")
    B --> C("2nd order: some open low-quality or unwanted accounts to earn the bonus")
    C --> D("3rd order: churn and complaints rise; service team overloaded; trust drops")
    D --> E(["Net: more accounts, worse customers, higher cost to serve"])
    A -.->|the move| F("Reward 90-day active accounts, not opened ones")
The dotted branch is what the "and then what?" pass surfaces: change the measure before you change the incentive. Leaders Loop

The second-order pass takes about ten minutes and changes the design. Instead of rewarding accounts opened, you reward accounts still active and funded at 90 days. Same goal, growth, but the incentive now points at the outcome you actually want, so the reaction it provokes works with you instead of against you. That's the whole discipline: not predicting the future, just refusing to stop at the first move. It pairs naturally with hypothesis-driven problem solving, state what you expect to happen, then watch the delayed reactions to see whether the system agreed.

Frequently asked questions

Isn't this just overthinking?

No. Overthinking is looping on the same decision without new information. Second-order thinking adds one structured pass, what does this cause, and how will people respond? and then stops. The discipline is bounded: a few rounds of "and then what?", not infinite ones.

How many orders deep should I go?

Usually two or three. First order is the intended effect; second is how people and systems react to it; third is how those reactions interact and compound. Past about three, the branches multiply faster than your confidence, so you're guessing, not reasoning.

Doesn't this slow everything down?

Only if you apply it everywhere. Reserve it for decisions that are hard to reverse or that change what people are paid or rewarded to do. For small, reversible calls, first-order thinking is fine. The cost of one second-order pass is minutes; the cost of a perverse incentive can be years.

What's the difference between second- and third-order effects?

Second-order effects are the direct reaction to your action, the behaviour your change provokes. Third-order effects are what those reactions then trigger in turn, often in a different part of the system and on a delay. The further out you go, the less certain and the more systemic the effect.

How is this different from just having a risk register?

A risk register lists things that might go wrong by chance. Second-order thinking traces the things that will go wrong by design, the predictable reactions to your own decision. It's not about bad luck; it's about the system doing exactly what your incentive told it to.

Related in the Toolkit

Where to go next