TL;DR. First-order thinking asks what a decision will do. Second-order thinking asks what happens after that: how people will react, what the change will accumulate into, and what will be true a year later. Most people agree it matters; far fewer do it well. In a study of 173 graduate students at MIT and Harvard, 96% could read a simple graph of people entering and leaving a store, but only 44% could say when the most people were inside. History is full of well-meant decisions that failed at the second step: in 1902, Hanoi paid a bounty for rat tails, and people began cutting off tails and releasing the rats to breed. The good news is that consequence reasoning responds to training. In forecasting tournaments, less than an hour of training improved accuracy by 6 to 11 percent. This article turns second-order thinking into a workout: 10 exercises, each anchored in a documented case, with a four-week plan and a rubric to score your progress.
What second-order thinking is, and why it is hard
The investor Howard Marks popularized the term “second-level thinking”. His example, in a 2015 memo drawn from his book The Most Important Thing: “First-level thinking says, ‘It’s a good company; let’s buy the stock.’” The second-level thinker asks what everyone else already believes and whether the price already reflects it. “Second-level thinking,” Marks writes, “is deep, complex and convoluted.”
Marks’s version is about markets and consensus. In everyday decisions, the same idea has a broader form: every action changes the situation, and people and systems respond to the change. First-order effects are what you intended. Second-order effects are the responses, side effects and accumulations that follow.
Three things make this hard:
- Responses are invisible when you decide. You can see your plan; you cannot see how others will adapt to it.
- Accumulation is counterintuitive. People confuse flows (what comes in and goes out) with stocks (what builds up). The store study above is the classic demonstration: when asked when the fewest people were in the store, only 31% answered correctly.
- Measures invite gaming. As the social psychologist Donald Campbell wrote in 1976, “The more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures.”
We introduced this model briefly in our curated index of mental models for the AI era. This piece is the practice manual.
A case library: when the second step overturned the first
The exercises below are built on documented cases. Each one shows a first-order effect that was real and a second-order effect that changed the outcome.
Table 1. Documented cases where the second-order effect changed the outcome
| Case | Intended first-order effect | What followed | Source |
|---|---|---|---|
| Hanoi rat bounty, 1902 | Pay a few cents per rat tail to reduce the city’s rat population | Daily kills ran between 7,000 and 14,000 and passed 20,000 on the worst day, but people also cut off tails and released rats to keep breeding: the bounty rewarded tails, not fewer rats | Vann (2003), French Colonial History; Vann interview (2020) |
| Wells Fargo sales goals, 2009-2016 | Aggressive cross-selling targets to grow accounts per customer | Employees opened roughly 1.5 million deposit accounts and 565,000 credit card accounts that may not have been authorized; a later review raised the potential total to about 3.5 million accounts; fines of $185 million in 2016 and $3 billion in 2020 | CFPB (2016); Wells Fargo 8-K (2017); US Department of Justice (2020) |
| Interstate highway expansion, US | Add lanes to reduce congestion | Vehicle kilometers traveled increased roughly in proportion to lane kilometers, so congestion returned | Duranton and Turner (2011), American Economic Review |
| More efficient steam engines, 19th-century Britain | Burn less coal per unit of work | Cheaper power expanded coal use; Jevons: “It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption” | Jevons (1865), The Coal Question |
| Broadway pedestrian plazas, New York, 2009 | Improve traffic flow and safety in Midtown by closing parts of Broadway to cars | Injuries to motorists and passengers fell 63% and to pedestrians 35%; traffic effects were mixed, with northbound taxi speeds up 17% and southbound down 2% | NYC Department of Transportation (2010) |
| Car safety regulation, US | Save lives through mandated safety equipment | Peltzman (1975) argued drivers compensated by driving less carefully; later work on seat belts (Cohen and Einav, 2003) found no significant support for that compensation | Peltzman (1975); Cohen and Einav (2003) |
Two notes on this table. The last two rows are included on purpose: second-order effects are not always negative, and a claimed second-order effect can itself be wrong. Good second-order thinking tests the chain of consequences; it does not assume the worst. And the famous “cobra effect” story from colonial Delhi is not in the table because we could not find a primary record of it. Treat it as a parable. The Hanoi case is the documented version.
The 10 exercises
Each exercise takes 5 to 20 minutes. Do them on a real decision you are facing, not a hypothetical one.
1. The “and then what?” chain
Write one decision at the top of a page. Write its most likely consequence below it. Then ask “and then what?” four more times, each time about the line above. Stop when you reach something that matters in a year. Most people stop after one step; the exercise forces the next four.
2. The reaction map
List every group affected by the decision: customers, colleagues, competitors, your future self. For each, write one sentence on how they are likely to adapt. The Hanoi administration planned for rats, not for the people collecting the bounty.
3. The gaming test
For any target or metric you set, ask: “What is the cheapest way to hit this number without doing what I actually want?” If you can think of one in two minutes, so can the people being measured. Steven Kerr’s classic 1975 paper called this “the folly of rewarding A, while hoping for B”. Wells Fargo is what it looks like at scale.
4. The stock-and-flow sketch
Draw two boxes: what flows in and what flows out. Then draw what accumulates between them. Apply it to your inbox, your savings, your team’s technical debt or your skills. The question to ask is not “is the inflow good?” but “is the stock rising or falling?”
5. The rebound check
When something becomes cheaper, faster or easier, ask how much more of it people will use. This is the Jevons question. Apply it to AI: if drafting a report takes a tenth of the time, will you write fewer drafts, or ten times as many reports that someone now has to read?
6. The 10/10/10 horizon scan
Write down the effect of the decision in 10 days, 10 months and 10 years. The three answers are often different in sign: pain now, gain later, or the reverse.
7. The consensus check
This is Marks’s original exercise. Before acting on a conclusion, ask: “Who else already knows this, and what have they already done about it?” If the answer is “everyone”, the opportunity has probably been priced in.
8. The second-order pre-mortem
Imagine it is a year from now and the decision succeeded on its first-order goal but caused a problem you did not anticipate. Write the story of that problem. This is a variation of the pre-mortem: instead of “why did it fail?”, ask “how did success create a new problem?”
9. The reversal test
Run the chain backward, the approach we described in our guide to inversion. What would guarantee the worst second-order outcome? Then check whether any part of your plan resembles it.
10. The consequence journal review
Once a month, reread the “and then what?” chains you wrote four weeks earlier. Mark which predicted consequences happened, which did not, and which ones you missed entirely. Without this feedback loop, the other nine exercises are guesses. A decision journal is the simplest place to keep them.
Table 2. The 10 exercises at a glance (CEOtudent editorial framework)
| # | Exercise | Trains | Time | Documented case to study |
|---|---|---|---|---|
| 1 | “And then what?” chain | Depth of consequences | 10 min | Hanoi rat bounty |
| 2 | Reaction map | Other people’s adaptation | 15 min | Hanoi rat bounty |
| 3 | Gaming test | Metric and incentive design | 5 min | Wells Fargo |
| 4 | Stock-and-flow sketch | Accumulation | 10 min | Store study (Cronin, Gonzalez and Sterman) |
| 5 | Rebound check | Demand response | 5 min | Jevons; highway expansion |
| 6 | 10/10/10 horizon scan | Time horizons | 10 min | Highway expansion |
| 7 | Consensus check | What others already know | 5 min | Marks’s stock example |
| 8 | Second-order pre-mortem | Side effects of success | 20 min | Broadway plazas |
| 9 | Reversal test | Worst-case paths | 15 min | Wells Fargo |
| 10 | Consequence journal review | Calibration through feedback | 20 min monthly | Your own decisions |
Does this kind of thinking actually improve with practice?
The evidence is encouraging but calls for patience.
- Forecasting training works. In the Good Judgment Project’s geopolitical forecasting tournaments, Chang and colleagues (2016) found that a training module of less than an hour, which taught principles such as using comparison classes and base rates, improved forecasting accuracy by 6 to 11 percent over four years.
- Feedback works, slowly. In the store study, researchers gave participants feedback and let them try again. Scores on the stock-flow questions rose from 28% and 25% at first to 81% and 84%, but only by the sixth attempt. Even highly educated adults needed repeated corrected practice to master a simple accumulation problem.
- Future orientation is a measurable trait. Psychologists measure it with the Consideration of Future Consequences scale, first published with 12 items by Strathman and colleagues in 1994 and later revised into a 14-item version with separate “future” and “immediate” subscales. A 2018 meta-analysis found it linked to healthier behaviors, with effects that were significant but small.
The practical conclusion: a short introduction helps, but the gains come from repetition with feedback. That is why the workout has a monthly review built in, and why it should run for weeks, not a single afternoon.
Your four-week workout plan
Table 3. Four-week second-order thinking plan (CEOtudent editorial framework)
| Week | Focus | Exercises | Output |
|---|---|---|---|
| 1 | Depth | 1 and 6 on one real decision every weekday | Five written chains with three time horizons |
| 2 | People and incentives | 2 and 3 on every target or rule you set or receive | A list of metrics with one gaming risk each |
| 3 | Systems | 4 and 5 on one personal system (time, money, skills) and one work system | Two stock-and-flow sketches |
| 4 | Stress test | 7, 8 and 9 on your biggest open decision | A one-page second-order pre-mortem |
| Monthly after that | Feedback | 10 | Scored journal review |
How to score yourself. For each chain you review in exercise 10, give one point for each predicted consequence that happened, and write down every important consequence you did not predict. Track two numbers each month: the share of your predictions that came true, and the number of important consequences you missed. You are improving if the second number falls.
Why this matters more in the AI era
AI makes first-order actions cheap. A strategy memo, a hiring rubric, an automated workflow or a new performance metric can be produced in minutes. The second-order effects are not cheaper to predict; they arrive at the same speed and scale as the action. When a metric can be rolled out to a whole organization in an afternoon, so can the gaming of that metric. The rebound check applies directly: when AI makes an output cheap, expect more of it, and ask who has to absorb it.
AI can help with the workout itself. Ask a model to generate reaction maps from the perspective of each stakeholder, or to challenge your “and then what?” chain. But treat its answers as a list of hypotheses, not a forecast; then check them against base rates, as we described in our guide to the outside view.
The CEO and the student in second-order thinking
The CEO side of CEOtudent is responsibility for consequences, not intentions. The Wells Fargo sales goals were intended to deepen customer relationships; the consequence was millions of accounts that may not have been authorized and billions of dollars in penalties. A leader is judged on the second step.
The student side is the discipline of practice and review. Second-order thinking is not a personality trait some people have; it is a set of questions you can rehearse until they become automatic. Ten minutes a day for four weeks, followed by a monthly review, is a modest investment for a skill that changes the quality of every other decision you make.
Frequently asked questions
What is the difference between first-order and second-order thinking?
First-order thinking considers the immediate, intended effect of a decision. Second-order thinking considers what happens next: how people and systems respond, what accumulates over time, and what side effects follow.
Is second-order thinking the same as pessimism?
No. Second-order effects can be positive, as the pedestrian injury reductions on Broadway show, and a feared second-order effect can turn out to be weak, as research on seat belts suggests. The goal is to test the chain of consequences, not to assume the worst.
How long does it take to get better?
Studies suggest that short training helps a little and repeated practice with feedback helps much more. Plan for at least four weeks of daily exercises and a monthly review, then judge your progress by the consequences you miss.
Was the cobra effect real?
We could not find primary historical evidence for the story of cobra breeding in colonial Delhi. The Hanoi rat bounty of 1902 is a documented case of the same mechanism.
Can AI do second-order thinking for me?
It can generate candidate consequences quickly, which makes it a useful sparring partner. It cannot know how your specific stakeholders will respond, so its output should be treated as hypotheses to test.
Sources
- Marks, H. (2015). It’s Not Easy. Oaktree Capital Management memo, September 9, 2015; Marks, H. (2011). The Most Important Thing. Columbia University Press.
- Vann, M. G. (2003). Of Rats, Rice, and Race: The Great Hanoi Rat Massacre, an Episode in French Colonial History. French Colonial History, 4, 191-203.
- Consumer Financial Protection Bureau (2016). Consumer Financial Protection Bureau Fines Wells Fargo $100 Million for Widespread Illegal Practice of Secretly Opening Unauthorized Accounts, September 8, 2016; Wells Fargo & Company (2017), Form 8-K, August 31, 2017; US Department of Justice (2020), Wells Fargo deferred prosecution agreement, February 2020.
- Duranton, G., Turner, M. A. (2011). The Fundamental Law of Road Congestion: Evidence from US Cities. American Economic Review, 101(6), 2616-2652.
- Jevons, W. S. (1865). The Coal Question. Macmillan.
- New York City Department of Transportation (2010). Green Light for Midtown Evaluation Report. January 2010.
- Kerr, S. (1975). On the Folly of Rewarding A, While Hoping for B. Academy of Management Journal, 18(4), 769-783; Campbell, D. T. (1976). Assessing the Impact of Planned Social Change. Occasional Paper 8, Dartmouth College.
- Cronin, M. A., Gonzalez, C., Sterman, J. D. (2009). Why don’t well-educated adults understand accumulation? Organizational Behavior and Human Decision Processes, 108(1), 116-130.
- Chang, W., Chen, E., Mellers, B., Tetlock, P. (2016). Developing expert political judgment: The impact of training and practice on judgmental accuracy in geopolitical forecasting tournaments. Judgment and Decision Making, 11(5), 509-526; Peltzman, S. (1975), Journal of Political Economy, 83(4), 677-725; Cohen, A., Einav, L. (2003), Review of Economics and Statistics, 85(4), 828-843; Joireman, J. et al. (2012), Personality and Social Psychology Bulletin, 38(10), 1272-1287; Murphy, L., Dockray, S. (2018), Health Psychology Review, 12(4), 357-381.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
