I Tracked My Macros With AI for 90 Days — Here's What Happened

The Setup: Why I Decided to Track Everything

I'm a 30-year-old male, 82kg, roughly 18% body fat by DEXA, with about four years of consistent resistance training behind me. I'm past the beginner gains phase but not an advanced lifter — solidly intermediate. The kind of trainee where progress is measured in months, not weeks.

My goal was body recomposition: gain muscle and lose fat at the same time, over 90 days. The literature says recomp is possible for intermediate lifters but requires tighter nutritional control than a straightforward bulk or cut. That sounded like the perfect test case for a data-driven approach.

For training, I used Arvo's AI programming — four sessions per week on a push-pull-legs-upper rotation with auto-regulated volume and load. For nutrition, I used a separate macro tracking app (Arvo doesn't offer nutrition tracking yet). I weighed everything I ate, logged every meal, and weighed myself every morning.

My starting targets came from a standard TDEE calculator: 2,650 kcal/day at a mild deficit, with 180g protein (2.2g/kg), 75g fat, and the rest from carbs. If you want the full breakdown of how TDEE formulas work, our TDEE guide covers the science in detail. For this post, the important thing is that I trusted the formula — and that trust cost me a month.

Month 1: The TDEE Reality Check

Weeks 1 and 2 went smoothly. I hit 2,650 kcal/day within a 5% margin on most days, tracked every gram of protein, and felt good in the gym. But the scale didn't move. Day 1: 82.0kg. Day 14: 82.1kg. Flat.

That shouldn't have happened. My calculator estimated maintenance at ~2,850 kcal. At 2,650, I was supposedly in a 200 kcal deficit — enough to lose roughly 0.2kg per week. After two weeks, I should have been down about 0.4kg. Instead: nothing.

In weeks 3 and 4, I recalculated my TDEE using my actual weight trend instead of a formula. The math was simple: if two weeks at 2,650 kcal produced zero weight change, then 2,650 was my maintenance. My actual TDEE was roughly 2,450 — not 2,650. The formula had overestimated by 200 kcal/day.

This is common, especially for desk workers. TDEE formulas ask you to self-report activity level, and most people overestimate. “Moderately active” feels right when you train four times a week, but if you spend the other 160 waking hours at a desk, in a car, and on a couch, your actual expenditure is closer to “lightly active.”

I adjusted to 2,250 kcal for a mild deficit of roughly 200 kcal below actual maintenance. The scale started moving immediately: 82.1kg at the end of week 2 dropped to 81.4kg by week 4. The lesson was clear: static TDEE calculators are starting points, not gospel. Real data beats formulas every time.

Month 2: Protein Is the Only Macro That Matters (Almost)

By month 2, I'd settled into a rhythm. I was hitting 2,250 kcal ± 10% on 82% of days — not perfect, but consistent. The question was whether the macro composition mattered as much as the total.

Protein was the hardest macro to hit consistently. My target was 180g/day; I averaged 172g. The gap came almost entirely from dinner — cooking at home, I'd underestimate portions, or I'd reach for carb-heavy convenience meals on busy nights.

Here's where the data got interesting. I tracked adherence by day and cross-referenced it with training performance. Days where I hit 170g+ protein correlated with better recovery: lower RPE on the next session's working sets, fewer sets flagged as “harder than expected” by Arvo's auto-regulation. The effect was modest but consistent across the full month.

Fat and carb distribution? Barely mattered. Some days I ate 60g fat and 280g carbs; other days 90g fat and 200g carbs. As long as total calories were in range, my training performance and weight trend didn't care about the fat-to-carb ratio. This aligns with the literature: for body composition, protein and total energy are the primary drivers. Fat and carbs are interchangeable within reason.

Strength during the deficit held up better than I expected. Bench press moved from 95kg to 97.5kg for a single. Squat went from 130kg to 132.5kg. Small gains, but gains nonetheless — a good sign that the deficit was mild enough to support muscle protein synthesis.

The real surprise from the data: weekly training volume predicted body composition changes better than calorie adherence. Weeks where I completed 16+ sets per muscle group showed faster lean mass preservation (measured by weekly weight trend and strength maintenance) than weeks where I nailed calories perfectly but trained with lower volume. Diet set the conditions; training drove the results.

Month 3: The Recomp Results

I got a DEXA scan on day 0 and another on day 90. DEXA isn't perfect — it has a 1-2% margin of error for body fat percentage — but it's the most accessible gold-standard measurement. Here's what changed:

The headline: +2.3kg lean mass and -1.8kg fat, with a net weight change of only -0.5kg. Body fat dropped from 18.0% to 15.9%. All three main lifts went up — bench by 5kg, squat and deadlift by 7.5kg each.

These results are good but not extraordinary. They're consistent with what the literature shows for intermediate lifters doing a controlled recomp: modest fat loss, modest muscle gain, strength trending upward. The magic isn't in any single intervention — it's in the consistency of applying multiple small optimizations over 90 days.

What Surprised Me

Four things stood out that I didn't expect going in:

Surprise 1: the 200 kcal TDEE overestimation. Without the week-3 recalculation, I'd have spent 90 days eating at maintenance and wondering why my body composition wasn't changing. That single adjustment — dropping from 2,650 to 2,250 kcal — was probably the most impactful decision of the entire experiment. If you take one thing from this post: recalculate your TDEE with real weight data after 2-3 weeks. Don't trust the initial number.

Surprise 2: protein consistency mattered more than calorie precision. I obsessed over hitting exact calorie targets early on. But the data showed that hitting my protein target (±5%) on 85% of days was more predictive of good outcomes than hitting calories within 3%. Protein was the constraint that actually bound — total calories had a wider margin of tolerance.

Surprise 3: training volume drove the recomp more than diet. Weeks where I hit 16+ sets per muscle group showed faster lean mass gain regardless of whether my calories were slightly over or under target. The training stimulus was the primary driver of muscle protein synthesis; the diet just set the energy availability. This makes sense physiologically, but experiencing it in your own data makes it real.

Surprise 4: weekends were the problem. 60% of my off-target days fell on Saturday and Sunday. Social eating, restaurant meals, alcohol — the usual suspects. My weekday adherence was 91%. My weekend adherence was 64%. The overall 82% was being dragged down almost entirely by two days per week. If I could fix weekends alone, my adherence would have been above 85%.

What I'd Do Differently

Ninety days of data gave me a clear list of adjustments for next time:

Start with a lower TDEE estimate for desk workers. If you sit for 8+ hours a day, subtract 10% from whatever the formula gives you. You can always add calories back if you're losing too fast. Starting too high and adjusting down wastes weeks.

Recalculate TDEE every 2 weeks, not 4. Faster feedback loops mean faster convergence to your actual expenditure. Two weeks of daily weigh-ins gives you enough data points to see a trend.

Don't stress about fat and carb ratios. Hit your protein target. Hit your calorie target. Let fat and carbs fall wherever they naturally land. Our macro guide explains the science behind optimal ratios, but for practical purposes, protein plus total calories covers 90% of outcomes.

Track training volume alongside nutrition. They're co-variables, not independent. A nutrition log without training data is only half the picture. The interaction between volume and energy availability was the most informative signal in my dataset.

Solve the weekend problem explicitly. Meal-prep Saturday lunch. Set a two-drink maximum. Pre-log restaurant meals. Treat weekends as a specific failure mode, not just “try harder.”

Use a system that adjusts targets automatically. Manually recalculating TDEE every two weeks works, but it's tedious and error-prone. An AI system that ingests your daily weigh-ins and adjusts calorie targets in real time would eliminate the lag entirely. This is something Arvo could do with nutrition tracking in the future. For now, our calorie deficit guide covers cutting-specific strategies if you want to go deeper.

Should You Track Macros?

Yes — but probably not forever. If you've never tracked macros before, commit to at least three months. The first month is learning: figuring out what 180g of protein actually looks like, discovering that your “small handful of almonds” is 300 calories, realizing that your homemade pasta sauce has more olive oil than you thought. The second and third months are calibration: building the intuition to estimate portions without a scale.

After three months, most people can “eyeball” their daily intake within 10-15% accuracy. That's good enough for maintenance and gentle bulking. For more aggressive goals — contest prep, rapid recomp, precise cutting — you'll want to keep tracking.

The ROI of tracking is highest for recomposition goals, where the margin for error is smallest. Bulking and cutting are more forgiving: if you overshoot a bulk by 200 calories, you gain a bit more fat but still build muscle. If you overshoot a cut by 200 calories, you lose a bit slower but still lose. Recomp requires you to thread the needle — and data is how you thread it.

Disclosure: this is a single-person self-experiment, not a controlled study. Results will vary based on training history, genetics, adherence, and starting body composition. Body composition measured via DEXA scan at day 0 and day 90. Arvo does not currently offer nutrition tracking—macro tracking was done with a separate app. See our TDEE guide and macro guide for the nutritional science behind these recommendations.