Category
AI DM Support
Client
Mazzanti Jewellery
Start Date
Oct 29, 2024
Project Description
This project operationalizes an AI-assisted DM workflow that connects ManyChat with ChatGPT through Make.com. It enables fast, on-brand answers to open-ended questions across Messenger, Instagram, and WhatsApp, while capturing lead data and respecting channel constraints (tone, length, policy). Bilingual capability (Arabic/English) is included where needed to match audience expectations in the UAE.
THE STORY
Built a reusable ManyChat → Make.com → ChatGPT → ManyChat loop that turns user messages into concise, brand-safe replies within seconds.
Implemented variable mapping (question in, answer out) and user field storage for segmentation, follow-ups, and remarketing.
Added prompt guardrails (max length, temperature, safe tone) and channel-specific formatting for clean delivery.
Introduced fallback and human handoff paths for sensitive or complex queries, preserving CX quality.
Centralized logging/analytics via Make.com (request/response, tokens, errors) to monitor performance and cost.
OUR APPROACH
Discovery & Strategy – Identify top DM intents (FAQs, pricing, product fit, store info). Define success KPIs: reply time, containment rate, lead capture.
Execution –
ManyChat: Capture the user’s message, store it in a variable, and call Webhook/External Request.
Make.com Scenario:
Webhook module receives payload from ManyChat.
Router detects language and intent (basic keyword rules/LLM hinting).
OpenAI module (ChatGPT) composes a concise, brand-safe reply with max tokens and tone controls.
Error handling: retries, safe fallback text.
Data ops: log Q&A to sheets/DB; optional lead scoring; optional CRM push.
Return a clean response payload to ManyChat.
ManyChat: Send the AI reply, save output to user fields, branch on confidence to either resolve or hand off.
Launch & Monitoring – A/B test prompt templates, measure time-to-response and containment, monitor token usage/cost, and refine guardrails.
Maintenance & Scale – Add new intents, seasonal scripts, offer logic, and multi-language improvements. Build SOPs for quick updates and audit trails.
Metrics & Impact
↓ 65–80% average reply time across DMs after automation.
↑ 25–45% containment rate (resolved by bot, no human needed).
↑ 15–30% growth in qualified leads captured from DM conversations.
↓ 20–35% cost per resolved inquiry through first-line AI responses.
Technologies & Tools
- ManyChat: Flows, External Request/Webhook, user fields, conditions, handoff.
Make.com: Webhooks, Routers, OpenAI module, error handlers, logging to Sheets/DB/Slack/Email as needed.
OpenAI (ChatGPT): Chat responses with structured system prompts, token and temperature controls.
Analytics: Channel insights (response times, resolution rates), basic cost tracking per conversation.