Eddy AI

Eddy AI matches incoming customer queries against a knowledge base of FAQ entries using keyword and semantic similarity matching, then generates or retrieves pre-configured responses. The system uses pattern-based intent classification rather than deep NLP, making it fast but less capable of handling paraphrased or nuanced variations of common questions. Responses are templated and deterministic, reducing hallucination risk but limiting conversational flexibility.

Eddy AI identifies qualifying signals in customer conversations (e.g., purchase intent, budget mention, timeline) using rule-based classification and intent scoring, then routes qualified leads to human sales representatives or support queues. The system uses configurable decision trees and keyword triggers rather than probabilistic models, making routing deterministic but brittle when customer language deviates from expected patterns. Handoff includes conversation history and qualification metadata to contextualize the human agent's response.

Eddy AI collects customer conversations from multiple channels (Shopify chat, Slack, web widget, email) and surfaces them in a unified inbox interface, preserving conversation history and metadata from each source. The system uses channel-specific adapters to normalize message formats and timestamps, then stores conversations in a centralized database indexed by customer identity. This allows support teams to view all customer interactions across channels without switching between tools, though the normalization process may lose channel-specific formatting or rich media.

Eddy AI connects to Shopify's API to access product catalog data, customer purchase history, and order information, enabling the chatbot to answer product-specific questions and provide personalized recommendations based on browsing or purchase context. The integration syncs product metadata (name, description, price, inventory) and customer data (order history, cart contents) into Eddy's knowledge base, allowing the bot to reference real-time product information and customer context when responding to queries. This reduces the need for manual FAQ updates when products change.

Eddy AI connects to Slack workspaces to receive customer inquiries posted in designated channels, respond directly in Slack threads, and escalate complex issues to human agents. The integration uses Slack's Events API to listen for messages, maintains conversation context within Slack threads, and allows agents to respond from Slack without leaving the platform. Responses are posted as bot messages with metadata tags indicating confidence level or escalation status, enabling teams to manage customer interactions entirely within Slack.

Eddy AI allows non-technical users to design multi-turn conversation flows using a visual builder or configuration interface, defining branching logic based on customer responses, keywords, or intent classifications. The system supports conditional branches (if-then rules), loops, and handoff triggers, enabling teams to create guided conversations that collect information progressively without requiring code. Flows are stored as configuration objects and executed by a state machine that tracks conversation state and applies rules at each step.

Eddy AI tracks metrics on bot conversations (volume, resolution rate, escalation rate, average response time) and surfaces them in a dashboard with filtering by time period, channel, or conversation type. The system logs conversation transcripts and metadata (intent, confidence score, customer satisfaction if available) to enable post-hoc analysis and performance optimization. However, analytics are limited to basic metrics; the platform lacks advanced insights like sentiment analysis, topic clustering, or predictive indicators of customer churn.

Eddy AI provides an embeddable web widget (JavaScript snippet) that can be deployed on any website to initiate customer conversations. The widget supports customization of appearance (colors, logo, position, greeting message) through a configuration UI or code, and uses a lightweight iframe to isolate the chat interface from the host page's styling. The widget persists conversation state in browser local storage, allowing customers to resume conversations across page navigations without re-authentication.

When a customer query falls outside the bot's FAQ scope or explicitly requests human assistance, Eddy AI transfers the conversation to a human agent, including the full conversation history, collected customer information, and bot confidence scores. The escalation process uses a queue or assignment system to route conversations to available agents, with optional skill-based routing if configured. The agent receives a summary of the conversation and any qualification data collected by the bot, reducing context-switching and enabling faster resolution.

Eddy AI can receive customer inquiries via email, process them through the same FAQ matching and intent classification logic as chat conversations, and send responses back via email. The integration uses email parsing to extract the customer message and metadata (sender, subject), matches it against the FAQ knowledge base, and generates or retrieves a response. If escalation is needed, the email is converted to a support ticket and routed to a human agent. This enables teams to handle email support without manually copying conversations into the chat system.