Queue incoming emails in Zapier (Storage or a Sheet), batch them (e.g., groups of 10) and send one prompt to ChatGPT that returns compact JSON: {intent, urgency:1–5, tags[]}. Run batches every 1–5 minutes and cap processing to ~10–20 emails/min to be safe. Use a smaller model (gpt-3.5-turbo), limit response tokens, and prefer batching to reduce calls. On 429s use exponential backoff with jitter (wait 2^n seconds), retry up to 3 times, then surface for manual review.

Compare Zapier integration options: Compare Zapier and Make