This study aims to evaluate the efficiency of robo-advisors in financial services and propose corresponding optimization strategies. By constructing multiple indicators, such as the Sharpe ratio, mean-variance optimization model, and Kalman filter model, we conduct backtesting of the investment strategies of robo-advisors, using various experimental indicators such as annualized return, risk-adjusted return, maximum drawdown, and volatility to comprehensively analyze their performance. The experimental results indicate that machine learning-based robo-advisor strategies outperform the traditional mean-variance optimization model in terms of annualized return and risk control, particularly excelling in controlling maximum drawdown and volatility. Reinforcement learning optimization strategies and diversified asset allocation strategies further enhance the adaptability and stability of robo-advisors. Overall, robo-advisors demonstrate significant potential in improving investment returns and optimizing risk management, but challenges such as data quality and market volatility must still be addressed in practical applications.