Do AI trading bots actually work?

The word covers two very different things. A genuine machine-learning system derives its rules from data rather than having them written, which is a real technique with real applications. A marketing AI bot is a set of conditional rules — moving averages, an oscillator, a filter — with the term applied because it sells. From outside, and often from the documentation, the two are indistinguishable.

Where machine learning genuinely struggles is exactly here. These methods excel when the signal is strong and the data is abundant; price data is mostly noise, non-stationary, and available in far smaller quantities than it appears. A model with thousands of parameters fitted to a few years of gold prices is the most efficient overfitting machine ever constructed.

There is a narrower and defensible version of adaptivity: tracking which of a fixed set of strategies is currently working and weighting toward it. That is measurement rather than learning, and it does not invent rules — but it does respond to changing conditions, which is what most buyers actually want from the word.

What we do instead, and why we do not call it AI

Our engine runs a tournament: several strategies produce candidate signals on each scan and exactly one is executed, chosen by score. The score is built on the Wilson lower bound of each strategy's measured win rate, which is a confidence interval rather than a raw percentage.

Using the lower bound matters. A strategy that won its only two trades has an observed win rate of 100% and a lower bound near 30%, so it does not outrank a strategy with a long and solid record. Without that correction the engine would chase whichever strategy got lucky most recently, which is a good description of how a lot of adaptive systems fail.

We call it a tournament rather than AI because that is what it is. Anyone selling you a black box should be able to describe its selection mechanism in a paragraph — if they cannot or will not, the label is doing the work the explanation should.