Research and enrichment are genuinely faster. Sorting a raw list, checking a company against a profile, pulling the one public fact worth mentioning: that is work a model does well.
The first draft is fine as a first draft. What it cannot do is know which of the three angles is true for this reader, and that judgement is the entire value of the message.
The failure mode is not bad grammar. It is confident, fluent, slightly wrong, and sent to four thousand people before anyone notices.
How we do it
A person reads every message before the sequence goes live, and the same person answers every reply. A person reads the sequence before it goes live.