Why does Computer Vision extract text in a document in a different order if used locally or in production?

Question

I am using the Computer Vision service to extract text from a scanned document. And even though I use the same endpoint locally as in production, it returns some fragments of the text in a different order.

I have sent the same image scanned more than 10 times locally and in production, and always in local it reads well and in production, it reads in a different order.

Any idea why it happens?