How much do crypto trading bots actually make? Honest numbers from a quant desk

We run trading algorithms live on our own capital, and we have killed far more of our own strategies than we have shipped. That gives us an unpopular qualification for answering this question: we know what real numbers look like, and we have no bot marketplace subscription to sell you. The honest answer has three parts — what most bots actually return (losses), why the advertised numbers are arithmetic nonsense, and what realistic success looks like when it does exist.

Start with the arithmetic that kills the marketing

"1% per day" — a modest-sounding claim from bot marketplaces — compounds to over 3,700% per year. Sustained for three years on a $1,000 account it produces more than $50 million; a few more years and it exceeds the capitalization of major banks. Nobody quietly operating this sells it for $29/month. Every "% per day" claim dies on this arithmetic before any backtest is needed. Realistic systematic returns are measured monthly, in single digits, with losing weeks and real drawdowns — that is what our own live account produces and what any audited track record you can find looks like.

Why the typical retail bot loses

• The entry logic has no edge. Marketplace bots overwhelmingly run textbook indicators — RSI, MACD, EMA crosses, grid variations. In our public benchmark, 0 of 24 such strategies beat a random-entry control after real fees with statistical significance. A bot automates the edge it is given; automating zero edge yields zero edge, minus costs, at higher speed.
• Costs compound against activity. Round-trip fees on futures run ~0.1–0.2% per trade. A bot making five trades a day pays roughly 15–30% of its account per month in fees at 1x notional. High-frequency retail bots are, functionally, donation machines to exchanges. From our own live cost audit: in a low-activity month, commissions alone can exceed the gross edge of a genuinely positive strategy on a small account.
• Small accounts amplify everything. On sub-$5K accounts, minimum position sizes, fees and slippage form a fixed-cost floor that a thin edge cannot outrun. This is a structural reason most retail bot journeys end red regardless of the strategy.
• Backtest ≠ live. Advertised backtests routinely omit fees, use look-ahead data, or cherry-pick the parameter corner that worked. We wrote a field guide to honest backtesting — and a guide to spotting fabricated track records — because these failure modes are the industry norm.

What profitable actually looks like

Systems that survive honest testing share a profile, and it is nothing like a marketplace grid bot: they are event-driven (they wait for rare, specific market states rather than trading constantly), built on non-price information — positioning, order flow, liquidation dynamics — rather than chart patterns, capacity-constrained (the edge exists precisely because it is too small for large funds to harvest), and validated out-of-sample against random controls. They trade a few times per day at most, often less. Our two production algorithms fit this profile; the hundred-plus candidates that did not survive are documented, kills included, in the public research log. Realistic expectation for such systems: single-digit percent monthly on average, uneven, with drawdowns that test your discipline — and that is the good outcome, achieved by a small minority of attempts.

The checklist before you run any bot

• Demand a time-stamped live track record with drawdowns — not a backtest screenshot.
• Check the strategy class against our free Reality Check benchmark — if its logic is a textbook indicator, the base rate says no.
• Compute the fee load: trades/day × round-trip cost × 30. If that number rivals the claimed monthly return, the bot's true customer is the exchange.
• Never grant withdrawal permissions on API keys; profitable operators never need them.
• Treat "% per day" as a full disclosure that the seller is lying — the compounding math above is not debatable.