For pseudoalignment methods (like kallisto), it can't be a counting error (the EM algorithm will resolve multimapping transcripts; a count of 0 means that it just wasn't mapped). Pseudoalignment is one of the most sensitive mapping methods so if you see zero counts, I'd say it's most likely that your transcript of interest isn't present in your cDNA library.
Really hard to diagnose since we don't know your primer sequences, your qPCR protocol, your library prep protocol, your sequencing protocol, or even the commands/files you're giving to kallisto.