The most reliable method our species has for finding out what works is the controlled experiment. Scientists run it on molecules, engineers on systems, founders on markets. Almost nobody runs it on themselves.
Personal change usually gets the opposite design, one no evidence-minded person would accept anywhere else. The commitment is grand, permanent, and identity-staked, launched on a symbolic date with no pilot phase and no plan for negative results. From now on I delegate. No more phone in the evenings. This year I speak up in meetings. Then the first bad week breaks the streak, and the conclusion most people draw is “something is wrong with me,” when the fairer reading is “poorly designed study.”
There is a better design, and if you work in science or engineering you already trust it: small, reversible, low-risk experiments, run on the one subject you have unlimited access to.
Why resolutions collapse and experiments don’t
The problem with resolutions isn’t willpower – it’s threat. Change means uncertainty, and your brain treats uncertainty as danger, whether it comes as an unfamiliar behaviour, an unpredictable response, or a version of you that doesn’t exist yet. The bigger and more permanent the commitment, the bigger the perceived threat, and the harder the older parts of your brain work to pull you back to the known. That resistance shows up as procrastination, sudden urgent busywork, and remarkably persuasive reasons why next month would be better.
An experiment shrinks the threat until your brain lets it through. It’s small, so the worst outcome is trivial. It has an end date, so nothing is forever. It’s reversible, so your identity isn’t at stake. And it can’t fail the way a resolution fails, because a negative result is still a result. Instead of promising to become someone, you’re collecting data on one behaviour for one week.
How do you change a habit that actually sticks?
Treat the change as an experiment, not a resolution. Pick one specific behaviour that feels uncomfortable or uncertain, run it once as a small, reversible test, and name how it felt. Register that you survived or learned something, and let that evidence update what your brain expects. Then repeat, slightly bigger. Habits built from accumulated evidence stick; habits built from willpower last as long as the willpower does.
The full method has four steps, and each one is doing specific work.
The four-step experiment
Step one: choose something you perceive as threatening. Not objectively dangerous, just perceived as threatening, the kind of thing you keep finding reasons to postpone because it feels uncomfortable, uncertain, or ambiguous: handing a task to someone who’ll do it differently, asking the basic question in a room of experts, saying no to a meeting. The discomfort is the selection criterion rather than a side effect, because the experiment’s real target is the threat response, and the behaviour is how you reach it.
Step two: take the action, and label how you felt. Do the thing once, at small scale, then name the feeling precisely. Not “that was awful”: exposed, embarrassed, out of control, relieved. Precision matters because a named feeling is data, and data is what separates an experiment from an ordeal.
Step three: reinforce that you survived. This step feels unnecessary, and it’s the one most people skip. Deliberately register the outcome: I did the uncomfortable thing, and I’m intact. Or, when the outcome was genuinely bad: I did it, it went wrong, and I learned something specific. Either way, the discomfort was survivable. Your brain updates on the meaning you assign to an event rather than on the event itself, and an event left without an assigned meaning defaults to the old story.
Step four: reframe what happened. Your brain forecast danger. You acted anyway, and the danger either didn’t materialize or turned out to be something you could handle. That’s a forecasting error, and step four is where you correct the forecast: this category of thing is less dangerous than predicted. Run enough small corrections and the behaviour that used to need courage starts needing none, which is the only version of “habit” that survives a stressful quarter.
A worked example
Take the founder who can’t delegate. The resolution version (“from now on, I empower the team”) is grand, unfalsifiable, and staked on identity, and it usually collapses at the first sloppy deliverable. The experiment version runs like this. This week, hand one low-stakes task fully to one person, with the outcome defined and the method left to them, and don’t touch it. That’s step one, and the task is chosen precisely because handing it over feels risky. When the itch to check in hits, note it and name it: this feels like losing control. That’s step two. When the task comes back, register the result: it came back different from how I’d have done it, and nothing broke. Or: it came back late, and I learned my instructions were vague. That’s step three. Then correct the forecast: delegating this class of task is safer than my brain claimed. That’s step four. Next week, a slightly bigger task.
That’s the whole method. It looks almost too small to matter – too small, in fact, to trigger the defences that stopped the grand version. That is precisely why it works.
What looks like an experiment but isn’t
Some approaches resemble the experiment closely enough to pass for it, and each one quietly substitutes force for evidence. Forcing yourself through something doesn’t change what your brain expects to happen the next time.
The tough exterior, where you endure the discomfort but never process it, so nothing is learned. Anger as motivation, using frustration at yourself to power the change, which works briefly and teaches your brain to expect an attack every time the behaviour comes up. Abusive self-talk dressed up as coaching; “stop being pathetic” applies force, and force teaches your brain nothing. Extremes and too-much-too-fast, where everything is supposed to change at once, which spikes the threat instead of shrinking it. And linking your identity to the outcome, the most corrosive of them all.
Your experiment might fail. This is ok, and the design has to make it ok. A failed experiment is information about the design, never a verdict on the experimenter.
What a negative result means
A negative result means the test was too big, the conditions were wrong, or the hypothesis needs revising. So do what you’d do in any lab: shrink the next one. If asking a question in front of the whole company was too much, ask one in a five-person meeting. If a full day without checking email was too much, try two hours. No dose is too small to count, because the target is the forecast, and forecasts update on any real data point.
That reframes the usual story of failed change. The person who “has no discipline” is usually a person running experiments far too large, with no measurement, no debrief, and their whole identity riding on the result. Redesigned properly, the same person changes.
Run one experiment this week
Pick a behaviour you’ve tried to change before. Write four lines, the lab-notebook way. Hypothesis: I predict that if I do this once, at small scale, the consequences will be survivable. Procedure: what exactly, when, and how small. Observation: what happened and what you felt, named precisely. Conclusion: what your brain got wrong about the danger. One behaviour, one week, four lines.
This is also, incidentally, much of what coaching is: designing the next experiment with someone who has no stake in flattering your current design, and who holds you to step three when you’d rather rush past it. If there’s a change you’ve failed at twice, we can design its first experiment together on a free intro call.