I’ve spent the last 10 years studying how people make health decisions, in research and in thousands of hours of coaching calls. Here’s the pattern I keep seeing: When someone misses a goal they genuinely cared about, skips a workout they’d really want to do, or breaks a streak they were proud of, it’s rarely a willpower problem. It’s that the system they’re running on was never designed to produce what they wanted in the first place.

There’s a quote often attributed to W. Edwards Deming: Every system is perfectly designed to get the results it gets. That’s true of factories, of algorithms, and of you. If a change feels hard to stick to, the inputs to your system weren’t aligned at that moment. Your algorithm ran, and it produced a predictable result, just not the one you wanted.

Most behavior change advice ignores personalization. It hands you the same playbook everyone else gets without checking whether those things actually address the specific places your system is leaking. That’s the gap I want to help you close, not with another one-size-fits-all habit formula, but with a way to diagnose your own system.

Understanding Change

Most behavior change models treat decisions as if they run on static inputs. Picture 1,000 people in a study trying to be more physically active. You give them an intervention, and on average, activity increases. Great! But zoom in on the individuals, some made big changes, some made small ones, some didn’t budge, and a few actually moved less. Same inputs, different outputs, even though the average said it worked.

This means that there are moderating factors the average is hiding. During my postdoc, I worked with a kind of analysis called multilevel modeling, which lets you see two things at once. First, people differ; this is what most research looks at. Second, the same person differs from themselves, depending on when, where, who they’re with, and what state they’re in. You might be more motivated to walk than your spouse on average, but you’re also more or less motivated than your own average from one day to the next. Sleep, stress, mood, weather, the conversation you had at breakfast, all feed into your algorithm at that exact moment

Think about how Google Maps works. It doesn’t just know your destination. It knows where you are right now, what traffic looks like on every possible route at this exact time, and which option will get you there fastest under current conditions. It doesn’t give everyone the same directions; it gives you directions specific to your starting point, your timing, and what’s in the way. When conditions shift mid-route, it recalculates. That’s not a simple program, and neither are you. You’re a personalized, adaptive system.

The question behavior change science has been slowly moving toward is: Which intervention works for this person, in this context, at this moment in their life? If you want to change, you have to start asking yourself this.

Your Personalized Algorithm

How do you answer that question? You need a model of how your system actually runs.

Your behavior is governed by what I call a personalization algorithm. In any given moment, this decision framework determines which behavior is most likely to win, given everything true about you at that moment. The reason I ate candy at midnight, or dropped the Portuguese streak I’d been protecting for months, or the reason you skipped the workout you’d planned, is not a character flaw; it’s that the conditions in that moment weren’t there to produce the desired result. Your algorithm did what it always does: produce the most likely behavior given the inputs it had.

Here’s the simplest way I’ve found to picture it. Imagine a tank, and inside the tank is water. The water represents your motivation to engage in a specific behavior, say, going for a walk after work. The level isn’t fixed. It rises and falls, and where it sits at any given moment determines whether the behavior happens. Let’s say your tank needs to be 55 percent full for you to act.

That graphic is the same person, with the same goal, across seven days. Monday, they’re at 45 percent. On Wednesday, they drop to 38 percent. Saturday, they hit 84 percent. Nothing fundamental about them changed in a week. Their context shifted, and the water (motivation) followed. This is what my postdoc work was actually measuring: that these psychological constructs can have dramatic daily swings.

Now, picture a pipe on the tank’s side. The opening of that pipe, the inflow point, sits at a specific height. When the water rises above it, water flows through, and the behavior happens. When it stays below, nothing moves.

Motivation isn’t operating in isolation, though. A network of tanks feeds into it. Self-efficacy, your beliefs about whether the effort is worth it, your emotional state in the moment, how rested you are, how much stress you’ve already absorbed that day, and whether the behavior fits who you think you are all feed into the level in the motivation tank, and they’re all moving at different rates throughout the day.

If social support and convenience increase, that feeds into the motivation tank, and the behavior is more likely to happen. The same person is different depending on when, where, who they’re with, and what state they’re in. Which means the question is never just, “How do I get more motivated?” The question is what is actually determining my water level right now, and what can I do about it.

That’s the inflow point, where the pipe meets the tank. The height of that opening is how much motivation the behavior actually demands. A one-hour workout sets the pipe high. A five-minute walk sets it low. A meal you’ve prepared a hundred times barely requires any water at all. When the water isn’t flowing, you have two options. Raise the water or lower the pipe. Increase your motivation, or make the behavior itself demand less.

Both options work the same way: by adding what computer science calls a module, a distinct, self-contained component that, plugged into a system, changes how the system runs. In psychology, we call the same thing a behavior change technique. Either way, you’re inserting a small, targeted intervention to change the algorithm.

Scheduling is a module. So are environmental design, habit stacking, and social accountability. Each one is a lever; it either adds water to a tank or lowers the pipe. The personalization part that most habit advice gets wrong is that the right module has to line up with the right tank—called parameters for effectiveness. If your motivation to cook a healthy dinner is low, a meal-planning module won’t help. That’s like asking: What salad do you want? When the person in front of you doesn’t want a salad to begin with. It’s the wrong module.

No single module fixes every system. What works is looking at where your water level is actually falling short, understanding why, and applying a module that addresses that specific gap. That’s what a personalized algorithm looks like in practice.

Here’s the diagnostic in action. I mentioned that I ate a piece of candy at midnight. Well, three pieces. In algorithm terms, my motivation to eat well wasn’t high enough to clear the threshold. The candy was very convenient; my goal of eating well wasn’t salient at that moment, and I wasn’t monitoring anything. The algorithm ran its course, and the system did its thing.

In this case, I had to add the right modules to change the algorithm. When I wanted to make better food decisions, it was not about trying harder; it was about changing the inputs. I made the cue less obvious (meaning the candy is not in a convenient place). I made the goal more salient I want to be fit for my wedding or vacation) I wore a continuous glucose monitor for real-time feedback, which improved self-monitoring. This reduced convenience, improved goal salience, and self-monitoring, which made my motivation to eat healthy rise to an adequate level for the behavior to happen.

You’ve probably read some version of this advice before. Lower the friction, set up your environment. These are good principles, but they just tell you the tools that exist; they don’t tell you what tools your system needs right now.

When you think about change, the best tool is the one that addresses the ‘tank’ that needs it. Address the inputs first, before tinkering with the behavior. The fact that the same person is different on different days, in different contexts, under different loads is a feature, not a bug. Under this lens, you don’t just change your behavior; you design it by producing the necessary conditions.