An AI trading bot can show a profitable backtest and still lose money live, and the usual reason is cost. Fees, spread, slippage, funding charges, and subscriptions each take a small slice, and for a bot that trades frequently, those slices add up fast. Beginners often judge a bot by its gross returns or its win rate, but the only number that matters is net return after all costs. Learning to calculate that number before going live prevents the common surprise of a bot that wins often yet slowly drains an account.

Every trade on an exchange incurs a fee, usually a percentage of the trade value, and a spread, which is the gap between the buy and sell price. For a bot making dozens of trades per week, even a small fee per trade becomes a large monthly cost. Spread matters most on less liquid pairs, where the gap is wider, and during volatile periods when it can widen further. To estimate this, multiply your expected trades per month by your average fee and spread cost per trade, and compare that total with your expected gross profit. If costs consume more than a third of gross profit in your estimate, the strategy has little margin for error, and a small change in market conditions can turn it negative.

Slippage is the difference between the price your bot intended to trade at and the price it actually got. It happens when price moves between signal and execution, or when an order is large relative to available liquidity. Backtests often assume zero slippage, but live bots, especially those trading on fast signals or thin pairs, regularly experience it. Slippage hurts more for market orders than for limit orders, and more during news or low-liquidity hours. Track your actual fill prices for the first weeks of live trading and compare them with your bot's signal prices. If slippage averages more than your backtest assumed, update your expectations before deciding whether the bot is working.

Beyond exchange costs, many bots charge a monthly subscription, a performance fee, or both. A subscription is a fixed cost that must be covered before you make any profit, which matters most for small accounts. If you run a bot on a virtual private server, add that cost too, along with any data or signal fees. For leveraged bots, funding rates on perpetual contracts can also accumulate, especially for positions held over days. List every cost in one place, monthly and per trade, and calculate the gross return your bot must earn just to break even. That break-even figure is a reality check that many marketing pages omit.

Choose liquid pairs with tight spreads for frequent trading, and avoid running a high-frequency bot on an illiquid pair. Use limit orders where your strategy allows, to control fill prices, and avoid trading during the most volatile news minutes unless your strategy is designed for them. Review your cost breakdown monthly, and pause any bot whose net return after costs is consistently near zero or negative, even if its gross chart looks attractive. A bot does not need to trade constantly to be useful. Often, a slower bot with fewer, higher-quality trades keeps more of its edge after costs than a fast bot that trades its profits away in fees. Review your bot monthly with a simple cost sheet, and compare net profit across months rather than judging a single good week. Small savings compound.