You run an experiment for 5 days. The variant is up 8%. You ship it.
Two weeks later, the metric is back to where it started. What happened?
You got fooled by noise. The 8% lift was random fluctuation, not a real effect. You saw a pattern in randomness and acted on it. This happens constantly, and it's the number one reason teams lose trust in experiments.
Statistical significance is the tool that prevents this. It tells you: "Given the data you have, how confident should you be that this result is real and not random noise?"