conversational AI glossary for ecommerce.
heysteff is an AI platform for customer support and sales across WhatsApp, Instagram, Messenger, Gmail and Shopify; Steff is the AI agent that runs it. This glossary defines, in alphabetical order, the 34 terms that come up most when talking about AI agents, business messaging and conversational customer support.
AI agent
An AI agent is an artificial-intelligence system that doesn't just answer questions — it reasons with the context of a conversation and executes actions (tools) to complete it: look up an order, book an appointment, recover a cart. Unlike a rule-based chatbot, it doesn't follow a fixed decision tree; it interprets natural language and decides what to do on each turn. In heysteff, the AI agent is called Steff and works across WhatsApp, Instagram, Messenger, Gmail and Shopify from a single inbox.
AI copilot
Copilot mode (AI copilot) is an operating mode in which the AI doesn't send replies directly to the customer; instead it suggests a reply to a human agent, who reviews it, edits it if needed and sends it. It suits teams that want AI support without handing off the final customer relationship. In heysteff it coexists with autonomous agent mode and can be turned on per conversation, per channel or for a whole workspace.
AI sales agent
An AI sales agent is an AI agent specialized in sales tasks inside the conversation: answering product questions, offering alternatives, recovering an abandoned cart or closing a purchase without taking the customer out of the chat. It differs from a generic support assistant in that its main goal is to move toward conversion, not just to resolve a question.
API
An API (application programming interface) is the set of rules that lets two software systems communicate — for example, so an AI agent can check a store's inventory or an order's status. In heysteff, the tool builder can connect any system that exposes an HTTP API, without needing a native integration.
Automation flow
An automation flow is a visually configured sequence of steps that triggers on an event (an incoming message, a Facebook lead, a booked appointment) and runs actions like sending a message, waiting for a reply, updating a field or calling a tool. In heysteff, flows are built with a visual, no-code builder.
Baileys
Baileys is an open-source library that lets you connect an existing WhatsApp number without going through Meta's official API, by emulating a WhatsApp Web session. It's the route for using an unofficial number the business already has, as opposed to the Cloud API, which requires formal verification with Meta. heysteff supports both connection paths.
Broadcast
A broadcast is a bulk send of outbound messages to a list of contacts — for example, to announce a promotion or re-engage inactive customers. On WhatsApp, a broadcast to contacts outside the 24-hour window requires a template (HSM) approved by Meta, not a free-form text message.
Cart recovery
Cart recovery is the automatic process of contacting a customer who added products to a Shopify cart but didn't complete the purchase, to remind or nudge them to finish the order. heysteff does this over WhatsApp and email, with an AI agent that can answer questions inside that same conversation instead of only sending a generic reminder.
Checkout link
A checkout link is a direct payment link that an agent (human or AI) can send inside a conversation so the customer can complete the purchase without leaving the chat, avoiding extra steps like hunting for the product on the website.
Conversational commerce
Conversational commerce is the sales model where the entire process — discovery, questions, payment and post-sale — happens inside a chat (WhatsApp, Instagram, Messenger) instead of a traditional website or form. It depends on the agent handling the conversation, whether human or AI, being able to resolve it end to end without pushing the customer to another channel.
CSAT
CSAT (Customer Satisfaction Score) is a customer-satisfaction metric, typically measured with a short question at the end of an interaction ("how satisfied were you?") and expressed as a percentage or average. It's distinct from operational metrics like first-response time or the resolution-without-a-human rate.
Echo detection
Echo detection is the mechanism that identifies when a human agent has already replied to a customer from the native channel (for example, WhatsApp directly from a phone) so it can automatically pause the bot and avoid sending a duplicate or contradictory reply in the same conversation.
HSM / WhatsApp template
An HSM (Highly Structured Message), or WhatsApp template, is a message format pre-approved by Meta used to write to a customer outside the 24-hour window or for notifications (confirmations, reminders). Unlike a free-form text message, its content and structure must be approved before it can be sent.
Human escalation
Human escalation is the process by which a conversation moves from the AI agent to a team member — whether because the customer asks for it, because a sensitive keyword is detected (complaint, urgency), or because the agent can't resolve the query. heysteff lets you define escalation rules and keywords per workspace.
Knowledge base
The knowledge base is the trainable set of information — catalog, policies, FAQs, brand tone — that an AI agent uses as context to answer. The more specific and up to date it is, the more accurate the agent's replies are and the less it needs to escalate to a human.
Lead qualification
Lead qualification is the process of collecting and validating a contact's data — budget, need, urgency — inside the conversation itself, to decide whether it moves to a booking, a purchase or is dropped. An AI agent can ask these questions naturally and store the answers without the customer feeling like they're filling out a form.
LLM
An LLM (Large Language Model) is the AI model trained on large volumes of text that generates natural-language responses and can reason over instructions and context. It's the engine behind a conversational AI agent; different providers (Anthropic, OpenAI, DeepSeek) offer LLMs with different strengths and costs.
Model Context Protocol (MCP)
Model Context Protocol (MCP) is an open standard, created by Anthropic, that defines how an external AI (for example, the AI a customer uses in their own assistant) can read and configure third-party software in a structured way, instead of through a web interface built for humans. heysteff exposes an MCP server that lets you configure an entire workspace — agent, flows, tools, channels — from the customer's own AI. More on heysteff's MCP server.
Multi-tenant
Multi-tenant is a software architecture in which a single instance of the application serves multiple customers (workspaces or brands) while keeping their data completely isolated from each other. In heysteff, this lets you run several brands, stores or countries from one account, each with its own configuration, agent and data, without them mixing.
NLU
NLU (Natural Language Understanding) is a system's ability to interpret the intent and the entities inside a message written in free-form language, rather than relying on predefined commands or buttons. It's what lets an AI agent understand a question written casually or with typos.
Omnichannel inbox
An omnichannel inbox is the category of tool that centralizes conversations from different messaging channels into a single workspace for the support team. It's the general concept behind what heysteff implements as its unified inbox, with the difference that in heysteff each conversation can also be handled by an AI agent, not just routed to a human.
Opt-in / opt-out
Opt-in is a contact's explicit consent to receive messages from a brand (required before sending WhatsApp templates or email campaigns); opt-out is the action by which that contact revokes that consent and stops receiving communications. heysteff's email templates always include a one-click unsubscribe (opt-out) link.
Prompt
A prompt is the set of instructions, context and examples given to an LLM to steer its response: who it is, what tone to use, what information it has available and what it shouldn't do. The quality of the prompt, together with the knowledge base, largely determines how well an AI agent responds.
RAG
RAG (Retrieval-Augmented Generation) is a technique where an LLM looks up relevant information in an external knowledge base before generating its response, instead of relying only on what it learned during training. It lets an AI agent answer with up-to-date, business-specific data, such as its catalog or its policies.
Resolved conversation
A resolved conversation is heysteff's billing unit: a distinct customer served within a 24-hour window, regardless of how many messages are exchanged or which channel they write on. You're not billed per message or per team seat, but per that complete conversation. More detail in the dedicated article.
Rule-based chatbot
A rule-based chatbot is an automated-response system built on fixed decision trees: if the customer writes X, reply Y, or it presents buttons with predefined options. It doesn't understand free-form language or reason over new context; outside its scripted path, it gets stuck or escalates. It's the model behind most conversational-marketing flow builders.
Speech-to-text
Speech-to-text (transcription) is the automatic conversion of spoken audio into written text. In heysteff, this capability lets the AI agent understand a voice note a customer sends on WhatsApp or Instagram without a human having to listen and transcribe it manually.
Tool / function calling
A tool (or function calling) is a concrete action an AI agent can run during a conversation, beyond generating text: book an appointment, look up an order, save a qualification field or call an external API. The agent decides when to use each available tool based on the context of the conversation.
Unified inbox
The unified inbox is the central view where messages from every connected channel (WhatsApp, Instagram, Messenger, Gmail) arrive and are answered in one place, without switching tabs or tools. It shows the customer's full history regardless of which channel they wrote on, and it's where the AI agent's replies and the human team's replies in copilot mode live side by side.
Vision (multimodal AI)
Vision, in the context of an AI agent, is its ability to interpret images sent by the customer — a product photo, a payment receipt, a screenshot — and respond or act based on what it sees, without a human having to review the image first. It's a form of multimodal AI, combining text and image in the same conversation.
Webhook
A webhook is an automatic notification one system sends to another the moment an event happens (a new message, a paid order, a captured lead), without the receiving system having to constantly poll for updates. It's the mechanism WhatsApp, Shopify and Meta use to notify heysteff in real time, and the one heysteff exposes to connect to a business's own systems.
WhatsApp 24-hour window
The WhatsApp 24-hour window is the period, counted from the customer's last message, during which a brand can reply with free-form text. Once that window closes, the only way to restart contact is with a template (HSM) approved by Meta, not a normal text message.
WhatsApp Business Platform
WhatsApp Business Platform is Meta's official name for the enterprise WhatsApp infrastructure aimed at medium and large volume, which includes the Cloud API, message templates (HSM) and verified-number management tools. It's distinct from the WhatsApp Business App, the free single-device application designed for small businesses.
WhatsApp Cloud API
The WhatsApp Cloud API is Meta's official infrastructure to connect a verified WhatsApp Business number to external software, without depending on the WhatsApp Business app installed on a phone. It's the recommended channel for high volume and for sending approved templates (HSM) outside the 24-hour window.
◆ also useful
Head back to the resource center, read the FAQ, or dig into heysteff's MCP server.
◆ next step
Get a demo and we'll show you these concepts running on your real channels.
Get a demo →