Every AI trading bot advertises a number. Usually it is the win rate: our bot wins 80 percent of its trades. That number sounds impressive, and it is almost useless on its own. A bot can win 80 percent of its trades and still lose money, if its few losses are much bigger than its many small wins. Judging a bot properly means looking at a small set of metrics together, understanding what each one actually measures, and knowing which combinations spell trouble. This guide explains the five numbers that matter, in plain language, with no maths degree required.

Win rate: the most quoted, least useful metric

Win rate is simply the percentage of trades that close in profit. A bot that wins 60 out of 100 trades has a 60 percent win rate. The problem is that win rate says nothing about how much each win or loss is worth. A scalping bot might win 90 percent of its trades for a few dollars each, then lose everything on one bad trade that runs 10 times larger. A trend bot might win only 35 percent of its trades and still be very profitable, because its winners are far bigger than its losers. Always ask for the average win and the average loss alongside the win rate. Without them, the win rate is marketing, not information.

Expectancy: the one number that predicts your future

Expectancy tells you how much you can expect to make, on average, per trade. It combines win rate, average win and average loss into a single figure. If a bot wins 50 percent of its trades with an average win of 100 dollars and an average loss of 60 dollars, its expectancy is 20 dollars per trade: half the time you gain 100, half the time you lose 60. A positive expectancy means the strategy makes money over many trades; a negative one means it loses money no matter how good any single month looks. Expectancy is the closest thing to a crystal ball in trading, and any serious bot provider should be able to show it, calculated after fees.

Profit factor: how much you make per dollar lost

Profit factor is the total profit from winning trades divided by the total loss from losing trades. A profit factor of 1.5 means the bot made 1.50 dollars for every dollar it lost. A factor below 1.0 means it loses money overall. Most consistently profitable strategies have a profit factor between 1.2 and 2.0. Be suspicious of profit factors above 3.0 on a backtest, because they usually signal overfitting — a strategy tuned so precisely to past data that it will not survive live markets. Profit factor is useful because it captures the balance of wins and losses in one number, but like every metric, it is only meaningful over a large number of trades, ideally hundreds.

Sharpe ratio and drawdown: risk-adjusted reality

The Sharpe ratio measures how much return a strategy earns for the risk it takes, by comparing its returns with their volatility. A higher Sharpe ratio means smoother, more consistent returns. A strategy that makes 20 percent a year with wild swings is less attractive than one that makes 12 percent steadily, and the Sharpe ratio captures that difference. Drawdown measures the worst peak-to-trough fall in the account: if your balance peaked at 10,000 dollars and then fell to 8,000, that is a 20 percent drawdown. Maximum drawdown tells you the worst pain the strategy has caused, and it is the number that decides whether you can actually stick with a bot through its bad periods. A bot with great returns and a 50 percent drawdown is a bot most traders will abandon at the worst moment.

How to read a bot's performance report honestly

When you read a performance report, check five things. First, the sample size: results from 20 trades prove nothing, while results from 500 trades start to mean something. Second, whether fees and slippage are included — a report that ignores them is fiction. Third, whether the results cover different market conditions, including at least one bad period, or only a lucky stretch. Fourth, whether the report shows live trading results or only a backtest, because backtests are easy to optimise and hard to trust. Fifth, whether the worst drawdown and the longest losing streak are disclosed. A provider that hides those numbers is hiding the information you need most.

Setting your own performance targets

Before you run any bot, write down your own targets: a minimum profit factor you will accept, a maximum drawdown that triggers a pause, and a review date when you will judge the bot honestly. If the bot breaches your drawdown limit, pause it and investigate, even if you still believe in the strategy. If it cannot beat a simple benchmark, like holding the asset, after a reasonable test period, stop paying for it. Measuring performance is not about finding the perfect bot. It is about knowing, with numbers rather than hope, whether the bot you have is earning its keep — and having the discipline to act on what the numbers say. Trading involves substantial risk of loss; never trade with money you cannot afford to lose.