6 Hard-Won Lessons from Building an Instagram DM Automation Engine on Meta's API

Targenix, a free automation platform for businesses in Uzbekistan, rebuilt its Instagram and Facebook messaging engine this year, expanding from simple hard-coded replies to a system with 9 triggers, 17 step types, and an AI fallback layer. The team discovered that private replies to ad comments are tied to the comment itself rather than a user ID, meaning the customer's messaging ID is only available after the first message is sent, requiring a rethink of how conversation runs are initiated and keyed. A separate timing issue caused flows to silently fail because the standard 24-hour messaging window does not apply to private comment replies, which instead carry a 7-day allowance — a distinction their original single-function check did not account for. The team validated the new engine using a shadow mode, running both old and new systems in parallel for 48 hours per page and comparing decisions before enabling live sending, while also learning to store shadow results in a database rather than logs to survive container redeployments. Additional lessons included never using the page owner's account for end-to-end testing due to Meta's role-based sending restrictions, and combining deterministic flow steps with a grounded AI reply layer to handle both scripted sales paths and unexpected customer questions without losing conversational context.