Different accuracy on deployed model

Is there any reason why on the same sentence I would get a different ‘confidenceScore’ than I do on the online training interface? Every time I deploy my model on a system I get a different accuracy.

Thank you and let me know if you need any more information.