This article presents the design and experimentation of a small-scale heliostat prototype, focusing on the analysis of tracking errors and optimization through artificial intelligence (AI). A heliostat model was fabricated using local materials, incorporating a robust structure and a motor control system for tracking on the azimuth and elevation axes. Five types of recurrent errors were identified and characterized individually and in combination for this heliostat, installed in Ouagadougou, Burkina Faso. These errors include time shifts, azimuth, elevation, mirror tilt, and pedestal tilt. The error characterization was performed through simulation with a mathematical model on the Arduino platform, adapted to the local context. The results show that the drift trajectories of the errors are similar from one geographical location to another, but the amplitude of the deviations varies significantly along the Z-axis. For example, for a delay error of 20 mrad, the maximum drift observed on the Z-axis is 4.6 mrad in Ouagadougou, compared to 1.2 mrad in Madrid and 1.5 mrad in Beijing. This difference can be explained by the variation in the sun’s elevation depending on the latitudes It is observed that the total combined error is similar to the sum of the individual deviations. On the X-axis, the drift is more pronounced with mirror tilt errors, while on the Z-axis, the reference elevation error is more pronounced. Experimental tests revealed an erratic trajectory of tracking errors, which does not conform to traditional error models, suggesting a complex combination of errors. We will explore the use of AI techniques, such as neural networks and reinforcement algorithms, to optimize these errors and improve the accuracy of the solar tracking system.