Averages, often used in application performance monitoring, can be misleading due to their inability to accurately represent data distributions, especially when outliers skew results. Averages assume a bell curve distribution, which rarely occurs in real-world applications that often have long-tail distributions with a few outliers significantly affecting the average. This can lead to misinterpretations and ineffective performance management. Percentiles, on the other hand, offer a more accurate representation by showing specific points in the data set, capturing the true performance experience for the majority of transactions. They are particularly useful for automatic baselining and alerting, as they provide a clearer picture of performance trends and degradations without the volatility and false positives associated with averages. Percentiles also aid in performance tuning by allowing targeted improvements on specific transaction segments, making them superior to averages in understanding and optimizing application performance.