Characters sometimes get replaced by different fonts

We create pdf files from a Word template and JSON data. Sometimes (app 5%) we find pdf files where some characters are replaced by a different font, and font size. If we submit the same data several times it will eventually get correct formatted, despite no changes at all in the source data. The affected font in the Word template is Arial Unicode MS 10pt. The font is fully embedded in the Word template.

Grateful for help how to get this 100% stable.