Every fitness app that has ever asked for your age, weight, height, and activity level has done the same thing: plugged your numbers into a population formula and handed you a calorie target. That number is a starting estimate. A reasonable one, maybe. But still a guess.
Body by AI Coach starts the same way. I want to be clear about that. On day 1, you are getting a formula estimate just like everyone else. What changes is what happens over the next 8 weeks.
What "Adaptive TDEE" Actually Means
Total Daily Energy Expenditure is the number of calories your body burns in a day — your resting metabolic rate, the thermic effect of food, and all movement combined. It is the number that determines whether you gain, lose, or maintain weight.
"Adaptive TDEE" means the engine does not assume it knows that number on day 1. Instead, it treats TDEE as a variable to be measured from your actual behavior and response over time. By week 8, if you have been logging consistently, the engine has replaced the formula's guess with a measured estimate specific to you — typically within a narrow confidence interval.
Why the Formula Is Wrong for Most People
The Mifflin-St Jeor and Harris-Benedict equations are built from population regression. They describe the average person in the dataset. Research on these formulas consistently finds they can misestimate an individual's actual TDEE by 500 or more calories per day. That is not a rounding error. At a 500-calorie error, you could be eating at maintenance while believing you are in a meaningful deficit. Your scale does not move. You assume something is wrong with your discipline. Nothing is wrong with your discipline. The formula was built for someone else.
The formula cannot see lean body mass versus fat mass — two people at the same total weight can have very different metabolic rates depending on their composition. It cannot see thyroid function, non-exercise activity thermogenesis (the fidgeting, pacing, and incidental movement that varies enormously person to person), or the metabolic adaptations that come with years of dieting history. A formula that describes an average captures none of that.
The Core Signal: Weight Trend Plus Logged Intake
The engine's measurement strategy is rooted in basic energy balance math. If you log an average of 2,100 calories per day over 21 days and your trend weight is flat — not going up, not going down — then your actual TDEE is approximately 2,100 calories per day. Full stop. No formula needed. The ground truth is in the data.
If the same person's Mifflin-St Jeor estimate came back at 1,900 calories, that formula is wrong by 200 calories — and the engine knows it because it observed the discrepancy directly. Going forward, targets are anchored to the measured 2,100, not the formula's 1,900. That adjustment means the difference between a real deficit and a phantom one.
Day-to-day weight is too noisy to use raw — water retention, digestion timing, and sodium intake all create swings. The engine uses trend weight (a rolling weighted average) to isolate the true signal from the noise. That is why 21 days of consistent logging produces a reliable read when a single week's data might not.
The Food Context Correction System
Not all logged calories are equally reliable. A home-cooked meal where every ingredient was weighed on a food scale carries a confidence rating near 1.0 — that entry is almost certainly accurate. A restaurant meal logged by eyeballing a portion from memory gets a systematic upward correction.
The research on self-reported food intake is consistent: people under-report actual intake by roughly 20 to 30 percent on average, and the error is largest for restaurant meals and estimated portions. The engine applies a context-based correction to logged entries — restaurant meals get a modest upward adjustment by default. More importantly, over time it learns your personal logging bias from your weight trend. If your trend weight consistently runs higher than the energy balance math predicts, the engine infers systematic under-logging and adjusts your effective intake upward to match the observed reality.
The Wearable Calibration: Why 80% Is the Starting Discount
Wrist-based wearables — Apple Watch, Fitbit, Garmin, Whoop — consistently over-report active calorie burn. Published accuracy studies show errors ranging from 20 to over 90 percent depending on activity type, with wearables biased high across the board. The engine starts by trusting only 80 percent of the active-calorie figure any wearable reports. That starting discount is conservative on purpose.
From there, the same weight-trend ground truth recalibrates the wearable discount for each user. If your trend weight is tracking exactly as the energy balance math predicts at the 80% discount, the engine holds it. If you're consistently losing faster than predicted, your wearable may be closer to accurate, and the discount narrows. If you're consistently losing slower, the wearable may be over-reporting by more than 20%, and the discount widens. Your wearable discount becomes personal to you over weeks of observation.
Why Month 2 Beats Month 1
In the first four weeks, the engine is accumulating signal. It has a formula estimate and early logging data, but the confidence intervals around your real TDEE are still wide. Targets are reasonable but still partly anchored to population averages.
By weeks 5 through 8, something shifts. The engine has enough consistent data to narrow those confidence intervals substantially. Your calorie target is no longer a formula output — it is a measurement. When the engine tells you to eat 2,050 calories to be in a 300-calorie deficit, that number is based on your observed metabolism, not a statistical average. Adherence finally produces predictable results because the target finally reflects your reality.
Month 2 feels different from month 1 because it is different. The math is working with actual data instead of educated guesses.
Calibration Is a Moat — It Can't Transfer
After 8 weeks, you have something no other app can replicate: a calibrated metabolic model built from your specific weight trend, your specific logging patterns, your specific wearable readings, and the systematic corrections that account for your personal under-reporting bias.
That calibration lives in Body by AI Coach. If you switch to another app, you are back to day 1 of someone else's formula. You start over with a population guess. Every week of consistent logging you have invested resets to zero.
I built this system because I lived through the frustration of formulas that did not match my body. The 8-week calibration process is how the engine earns the right to give you accurate targets — and why the coaching compounds in value the longer you stay.