We Tracked 100 Binance Auto-Invest Trades for 60 Days

1. Experiment Setup #

The experiment had exactly one goal: treat Auto-Invest as a "deployment channel" and see whether it can quietly accumulate position across different baskets and frequencies. We were not trying to measure "how much did we make" — 60 days is way too short to say anything meaningful about long-run returns in crypto. What we wanted to see was: does the execution layer leak, do slippage and asset rotation eat too much of the return, and is the AI-picked basket worth the extra cognitive cost?

The four baskets, by design:

The "AI picking" workflow in basket D is fully transparent: every Monday evening we feed Claude a snapshot of the prior 7 days of on-chain data (Glassnode plus DefiLlama) and the latest regulatory and project news that Perplexity surfaces, then ask Claude for 5 coins plus weighting rationale. The whole loop is about 30 minutes of human-plus-AI collaboration. Important: the AI proposes a basket of candidates — the actual orders are executed by the Auto-Invest template, not by AI directly.

2. Aggregate Results Across 100 Trades #

By the end of day 60, the 100 trades had deployed a total of 1,950 USDT (30×10 in basket A, 25×20 in B, 15×50 in C, and 30×20 in D), with a mark-to-market value of 2,073 USDT. Aggregate unrealized P&L: +6.31%. That headline number by itself means nothing — over the same window BTC was up about 4.8%, ETH up 9.2%, SOL down 3.1% — but the distribution is more revealing.

Sixty-six percent of the trades closed green. That sounds nice, but notice this: the four outlier trades sitting beyond the left tail of the distribution all came from basket D, the AI-picked one. We will pick those apart in section 4. What the distribution actually tells us is this: in 60 days, regardless of which basket, the vast majority of Auto-Invest trades cluster inside ±5%. Auto-Invest is not a tool for making big money — it is a tool for spreading out your entry points.

3. Performance by Basket #

Basket D had the best return at +7.33%, but it also had the widest swings. The Bluechip basket was second on return with much lower volatility — that is the empirical version of "why most people doing long-term DCA should just hold the bluechip basket." BTC-only is the steadiest and the lowest-returning basket, which is exactly what you would expect.

One number deserves its own line: total Binance fees across all baskets came to 0.41% of total capital deployed. Auto-Invest itself is fee-free (Binance's terms spell this out), and that 0.41% all came from manual spot rotations we did inside basket D. That gap is the real cost advantage of Auto-Invest over manual DCA.

4. The Worst 10 Trades #

The wins are easy to celebrate. The losses are where the information density lives. Here are the 10 worst trades of the 100:

Seven of those 10 came from basket D, the AI-picked one. Our reading is straightforward: the coins AI flags as "high upside" are structurally high-beta, which means they get amplified on the way down too. The WLD and TIA blowups both lined up with "unlock day plus a piece of bad news" — when the AI picks on Monday, it sees the prior 7 days of on-chain data but not Wednesday's unlock calendar. That is a data-recency problem, not a flaw in the picking logic.

It did force a change in our process. We added a hard filter to the basket D workflow: any coin with more than 3% of circulating supply unlocking in the next 7 days gets skipped, even if the AI recommends it. The rule is blunt, but it eliminates roughly 60% of the "DCA right before an unlock" traps.

5. Did AI Picking Beat the Passive Basket? #

This is the question we most wanted to answer. Basket D at +7.33% versus basket C at +5.80% — AI picking won by 1.53 percentage points. But that comparison is not fair on its own: basket D ran 30 trades while basket C ran 15, with different frequencies and different sizing. We re-ran it on an "equal-weight, risk-adjusted" basis:

The risk-adjusted verdict is sober: the AI-picked basket's Sharpe ratio is actually below the bluechip basket's. Put differently, the extra volatility was not compensated proportionally with extra return. Add in 30 minutes of human work per week, and AI picking did not win inside the 60-day window.

That does not invalidate AI picking as an approach. Sixty days is far too short, and single-window results get dominated by luck. We intend to keep all four baskets running out to 6 months and publish another report then. What we can say right now: using AI as "a second pair of eyes in the Monday review" is fine; treating it as "a thing that will reliably beat the bluechip basket" is not.

6. Our Post-Mortem #

The most valuable output of running 100 trades over 60 days is not the return number. It is a handful of process-level findings:

First, Auto-Invest's real job is to cancel your timing instinct. When we were designing the experiment, the initial plan was to manually buy 10 USDT of BTC every day. But every time the market dipped, we would hesitate and ask, "Should I skip today?" Switching to Auto-Invest deletes that hesitation by force. The disappearance of that psychological friction is worth far more than the 0.41% in fees.

Second, Auto-Invest does not protect you from unlock days. Binance's Auto-Invest templates do not automatically dodge token unlocks, FOMC meetings, or any macro events. They simply execute on schedule. That is a feature, not a bug — if you do not like it, either pre-filter through an AI basket or just exclude high-unlock-risk coins from the basket entirely.

Third, the "high-beta tilt" of the AI basket is structural, not coincidental. Any LLM (Claude, GPT-4o, take your pick), when asked, "Which coins are likely to outperform the market over the next month?", will gravitate toward names with strong narratives, clear catalysts, and high volatility. That "narrative bias" is an advantage in a bull market and a liability in a chop. If you choose the AI basket, you need to mentally accept that you are not holding a conservative portfolio — you are holding a high-beta one.

Fourth, the most useful "zero-loss" observation is about when you stop. Across all 100 trades, we never took profit or cut losses early — that is the design philosophy of Auto-Invest, and DCA neither takes profit nor stops out. But the +4.80% to +7.33% unrealized gains we have over 60 days could easily turn into a -20% basket within 3 days if something extreme happens, like BTC suddenly dropping 25%. DCA is not a strategy. DCA is just an entry method. Your exit strategy has to be designed separately.

For the next report, we plan to keep all four baskets running out to 180 days and add a new control group: basket E = AI-picked + hard unlock-day filter + 4% monthly take-profit reinvested. The question we want to answer is whether that workflow can pull basket D's Sharpe ratio back above 0.7.

If you want to replicate this experiment yourself: open an Auto-Invest template on Binance, pick BTC, ETH, SOL, BNB, and LINK, equal-weight them, run it weekly at 50 USDT per trade — that is our basket C. Stick with it for 6 months and revisit. That single move will teach you more than any "AI picking" exercise will.

Try Binance Auto-Invest → See tool reviews →

— AI Trade Lab, 2026-05-12