How Long Before You Know If a Change Worked?
The short answer
Give a change two to three weeks before you judge it, and judge it by comparing the average of the days before against the average of the days after — never by what the scale said the next morning. The first three or four days after any change are dominated by water and gut contents, and they tend to point in whichever direction flatters or panics you rather than the direction that turns out to be true.
That is the whole discipline: one change, one waiting period, one verdict. Everything below is why those numbers are what they are.
Why do the first few days always mislead?
Almost every change you can make produces an immediate response that is not the thing you were trying to measure.
Cut carbohydrate, and you draw down stored glycogen — which is held in the body along with a good deal of water. The scale can fall a kilogram or more in under a week, and almost none of it is fat. Add carbohydrate back, or go from a week of cooking at home to a weekend of restaurant food with far more salt in it, and the same kilogram walks back on just as fast.
Start a new training programme and you get the opposite illusion: unfamiliar work means fluid retention in the worked muscles, so the scale sits flat or rises for the first fortnight while body fat is quietly going down.
All of these settle within days to a couple of weeks. None of them is fat. We have written about the mechanism in more detail in why your weight went up overnight, and the short version is that a single morning's number carries so much noise it cannot be used as evidence about anything.
How many days does a fair comparison need?
Here is the arithmetic that decides it.
A kilogram of body fat holds about 7,700 kcal. So a genuine deficit of 500 kcal a day — a fairly aggressive one — produces about 3,500 kcal a week, or roughly 0.45 kg of fat. That is the real signal you are hunting.
Now consider the noise. Day-to-day body weight commonly swings by half a kilogram, often more, from food volume, salt, fluid, the timing of your last meal and your last bowel movement. In other words, a single day's wobble is frequently bigger than a whole week's worth of actual fat loss.
That is why point-to-point comparison cannot work, no matter how carefully you weigh yourself. Tuesday against Tuesday is one noisy number minus another noisy number.
Averaging fixes it. Take seven consecutive morning weights and average them, and most of that daily swing cancels out; the average of a week moves far more steadily than any day inside it. Compare one seven-day average against another seven-day average two weeks later and you are looking at something like 0.9 kg of expected change against a few hundred grams of residual wobble. That is a difference you can actually read.
Three practical consequences:
Weigh most mornings if you can, under the same conditions. You are not collecting daily verdicts; you are collecting the raw material for an average.
The smaller the change, the longer the wait. A 200 kcal-a-day adjustment is worth about 0.18 kg a week. That needs three to four weeks before the two averages separate convincingly, not two.
Fourteen days is roughly the floor for anything you intend to act on. Shorter windows are not wrong so much as unreadable.
This is also the logic behind how Day So Far sets a daily target: rather than trusting a formula's guess at what you burn, it compares what you logged against what the scale actually did over a fortnight, because a fortnight is about the shortest stretch that can tell you anything.
Why only one change at a time?
Because two changes in the same window cost you the window.
Suppose you drop your intake by 300 kcal and start running three times a week, both starting Monday. Three weeks later the average is down 1.2 kg. Which change did that? If you now want to keep the easier one and drop the harder one, you have no basis for choosing — and if the result had been nothing, you would not know whether one change worked and the other cancelled it, or neither did anything at all.
You have spent three weeks of real life and acquired no information. Do it four times in a row and you have a four-month record that answers no question you can name.
The only reasonable exception is a change to how you log rather than what you eat. If you have started weighing your cooking oil, or you were previously guessing at drinks, that is not an intervention to be judged on the scale — it is a correction to your measuring instrument, and the thing it changes is your logged total, visible within days. Oil and sauces and drinks are the two places where a hundred or two hundred uncounted calories a day most often hide.
What does a verdict look like?
Write down, before you start, what you would accept as a yes. Something like: the average of days 15–21 is at least 0.4 kg below the average of days 1–7.
Then leave the change alone for the full window. Then read it, and do exactly one of three things: keep it, undo it, or extend it by another week because the result sat on the line. Sitting on the line is a real and common outcome and is not a failure to decide.
The condition that trips people up is a stall after several weeks of steady progress, which feels like evidence that the plan has stopped working long before it is. How long a stall has to last covers that case specifically.
What impatience actually costs
This is not a willpower problem, and it is not fixed by caring more. It is a measurement problem, and the failure mode is making changes faster than results can arrive.
If you adjust something every five days, every adjustment is judged on data made almost entirely of water weight. You will reliably conclude that whatever you did on day one worked brilliantly and whatever you did on day six ruined everything, and both conclusions will be wrong. Worse, you will keep moving, which means the one thing a verdict requires — a stretch of weeks where only one variable changed — never exists.
The counter-intuitive move is that doing nothing for a fortnight is the productive part. You are not waiting for motivation to build. You are waiting for the noise to average out, which it does on its own schedule and not on yours.