Crypto trading bots work at execution, not at prediction. They follow a strategy faithfully, which only helps if that strategy already had an edge after costs.
The research is blunter than the marketing. A 2025 study of grid trading by Chen, Chen, and Jang works through the math of the classic spot grid and concludes that its expected return is “essentially zero” under simple assumptions. The grid takes profit on the way up and buys more on the way down, and the two halves cancel.
The same paper does not stop there. Its authors built a dynamic version that resets grid positions as conditions change, and in backtests on minute-level Bitcoin and Ethereum data from January 2021 to July 2024, it beat both the traditional grid and buy-and-hold.
That is a research result, not a forecast. The words that matter are “in backtests.” A backtest tunes parameters against a price history that already happened, so a curve that looks perfect usually describes the past precisely and the future not at all. Overfitting is the name for that failure, and the only honest defense is testing on a period the parameters never saw.
Software also does not know when it is out of its depth. In a 2026 stock trading competition run by Trade.xyz and cited in a DWF Ventures report, each account started with $10,000, and the top human beat the top agent by more than 5x. That same report counts more than 17,000 AI agents launched since 2025. Seventeen thousand agents is a head count, not a track record.
Then there are the costs, which most bot comparisons leave out. Take a grid bot running 40 round trips a month on a venue charging 0.1 percent per side. Those 80 fills, two for every round trip, come to roughly 8 percent of the value traded before the strategy has done anything at all.
Four costs decide the result, and only the first is obvious.
- Exchange fees. Charged per fill, so they scale with the strategy’s trade count rather than its returns.
- Subscription fees. A fixed monthly cost, which bites hardest on a small account.
- Slippage. The gap between the price your rule wanted and the price the order got, which widens exactly when the market moves fastest.
- Funding. Paid or received on every perpetual position held through a funding interval.
The grid figures above are illustrative, and your own venue’s schedule will differ. The shape holds either way. Set the total against holding the asset and doing nothing because a bot that cannot beat that baseline after all four costs has bought you activity rather than return.
So, are crypto trading bots profitable? Some users profit and many do not. The bot is rarely the variable that decides which, and anyone promising otherwise is selling something other than software.
Discipline, not profit.