Urdu is the language I have worked in most, and the one where the most goes wrong before a file ever reaches a translator. Most of what follows is about the second part.
The script is not a font choice
Urdu is written in Nasta’līq, a cursive style where each word cascades down and to the left along a sloping baseline. It is not Arabic, and it is not Naskh with different words in it. A translation set in Naskh is readable in the way that a legal contract set in Comic Sans is readable: an Urdu reader will understand every word and will also register, immediately, that nobody involved knew what they were doing.
The practical failures are always the same three. Naskh substituted for Nasta’līq because the production font had no Urdu face. Line height set from the English, which clips the descenders that Nasta’līq depends on. And numerals or embedded English flipped the wrong way, because bidirectional text was handled by hand instead of by the text engine. I check for all three before delivery, and I will tell you which one your existing file has if you send it.
What I am usually sent
Clinical: informed consent forms, patient information leaflets, questionnaires and the whole validation chain where a sponsor needs one. Legal: chargesheets and their supporting documents, moved in both directions, where a hedged verb changes what a document proves. Personal and immigration: certificates, transcripts, police records and the rest of what a USCIS filing asks for. Commercial: transcreated advertising, interface strings, and survey instruments.
Two of the four published samples are into Urdu, with a note on the decision behind each, which is the part that shows judgement rather than volume.
The vowels are not written, which matters now
Urdu is normally written without its short vowel marks. A human reader supplies them from context without noticing. A speech engine cannot, so it guesses, and a wrong guess turns one word into a different word.
If your Urdu text is going to be spoken, by a synthetic voice or by a voice artist reading cold, I put the marks back. This is ordinary editorial work that almost nobody offers, and it is the difference between an Urdu voice track that sounds like Urdu and one that sounds like a machine reading letters.
Where the pair actually runs
Into Urdu for 33 lines of the record, out of it for 4. Both directions are live, but the volume has always run one way, and back translation into English for regulated work is the main reason the reverse direction exists at all. I do not use a second translator to cover Urdu. It is my own language and my own hand.
What it costs. From $0.14 a source word for medical and legal work, from $0.12 for everything else, with a $50 minimum on anything up to 500 words. Certified documents are quoted on sight of a scan. The full rate sheet and the Trados match grid are public, because a rate you have to ask for is a rate that moves depending on who is asking.