I’ve been running a WhatsApp AI sales agent for a real estate company using n8n.
Replies now go out in under a minute. Before this, leads often waited more than two hours and many lost interest in the meantime.
The setup uses n8n, GPT-4o, Twilio WhatsApp, Airtable, and text-to-speech for voice notes. Here’s what caused problems and how I fixed them:
-
Voice notes
WhatsApp only plays certain audio formats as proper voice notes. Anything else shows up as a file. Switching to the right format (OGG/Opus) made the voice notes play smoothly inside the chat. -
Generating voice notes
Some text-to-speech services don’t give the audio right away, they give a job ID instead. I added a simple waiting loop that checks until the file is ready (with a limit on how many times it tries). -
Workflow branches
Sometimes one path in the workflow produces no data and the whole thing can get stuck. I now make sure every path always returns something, even if it’s empty. -
WhatsApp’s 24-hour rule
Outside the free 24-hour messaging window, normal replies stop working. For follow-ups I switched to approved message templates so conversations can continue. -
Silent delivery failures
Messages sometimes failed without anyone noticing. I connected Twilio’s status updates so failed sends are flagged instead of the lead just going quiet. -
Different languages
The AI prompt is organised in clear sections and works in multiple languages.
Happy to share more details about any of these parts if useful.
How are you handling failed message detection and the 24-hour messaging window in your own WhatsApp setups?