Rep AI

Behavioral AI reads real-time storefront signals such as browsing depth, time, product interactions, and hesitation. Its role is to decide whether a shopper is showing enough intent or friction to enter the guided-selling flow; it does not choose the final recommendation by itself.

This module applies the store's timing and eligibility rules to start a conversation at a specific moment. It receives the behavior signal, opens the chat with relevant context, and hands the shopper to product discovery instead of displaying the same prompt to everyone.

The agent uses live catalog context to ask narrowing questions, compare options, answer objections, and recommend matching or complementary items. This is the main decision-support step and can lead to an in-chat cart action.

Teams can shape which products, messages, and selling motions the agent should use or avoid. These controls constrain the recommendation step and let operators turn repeatable selling guidance into a governed playbook rather than an unrestricted model response.

Conversation results show where engagement, recommendations, and conversion paths worked or failed. Operators can privately test answers and change instructions before relying on them in customer conversations, closing the loop from outcome back to timing and guidance.

The agent draws from the catalog, help center, policies, instructions, and connected store context. This module supplies the approved answer base and order context to every later step; it cannot compensate for stale policies or missing integration data.

Routine questions can be answered without creating human work when the knowledge and confidence conditions are met. This step is the resolution node of the workflow and produces either a completed answer or a reason to continue to an action or handoff.

For supported requests, Rep AI can use connected commerce context to carry out an order-related action rather than only describing what an agent should do. Permissions and source data bound the action, and unsupported requests must stay with a person.

Low-confidence, sensitive, or unsupported requests can move to a human with the conversation history retained. When the team also buys AI Helpdesk, the handed-off request can continue there as a structured ticket.

Repeated unresolved questions and conversation patterns surface what the current knowledge or instructions do not cover. Teams inspect those signals, approve corrections, and use Test & Train before the changed response returns to production.

The product connects Email, Instagram, Facebook, TikTok, and WhatsApp surfaces to Rep AI. Each enabled channel is a separate pricing selection and usage input, while all of them enter the same operating model.

Messages are interpreted with the same catalog, policies, instructions, customer context, and prior conversation history used by the core agents. Its role is to keep an answer consistent across channels instead of maintaining duplicate channel-specific brains.

For eligible intents, the system can respond in the originating channel; other cases can be prepared or routed for a person. Channel controls determine where automation is allowed, so a connection does not mean every message must be auto-sent.

Teams can test channel behavior privately, refine instructions, and decide which cases need review or handoff. This module governs the transition between automated answers and human work before the conversation reaches AI Helpdesk or another team process.

Channel conversations and their signals remain available to the shared history and learning layer. That handoff lets Deep Research examine cross-channel themes when the account also has AI Concierge and the Deep Research add-on.

Rep AI attempts routine resolutions before adding work to the human queue. This is the intake gate: an AI-resolved request ends there, while an unresolved or governed exception continues into ticket creation.

Unresolved conversations become tickets in a shared workspace with history and an AI-generated summary. This module turns a chat into actionable work and gives the next agent enough context to continue without asking the customer to repeat everything.

Operational rules classify a ticket and determine its tags, priority, owner, or team. They determine how tickets are routed and can separate urgent or specialist work from the general queue.

Human agents reply, change ownership and status, and monitor service timing in the ticket workspace. Its role is to keep human work measurable and accountable after AI hands off the request.

The Helpdesk can trigger a satisfaction survey after a conversation and associate the result with support reporting. CSAT records perceived outcome quality; it should be read alongside resolution and operational data rather than treated as proof that every answer was correct.

Ticket topics and outcomes feed the broader insight layer, while migration support helps a team bring existing service operations into the workspace. Insights close the improvement loop; migration moves records and setup but does not automatically repair weak policies or processes.

When activated, Deep Research analyzes the previous seven days of conversations and then continues processing new conversations daily. This layer supplies both topic analytics and Ask AI; its coverage therefore depends on the eligible conversation history available to the account.

The dashboard groups conversation patterns into topics and shows counts, shares, and trends. Its role is to make recurring issues browsable; users should still open the supporting conversations before using a pattern to inform an operating decision.

Teams can narrow the analysis by period and topic type or define a bounded custom topic. These controls shape the analysis set that later flows into evidence review or Ask AI, so a poorly scoped filter can change the meaning of the result.

Ask AI researches one selected conversation set after showing its size and credit cost. It generates a report retained in History with a summary, up to four findings, and follow-up prompts, giving operators a focused route from raw conversations to a question-specific result.

Completed Ask AI work can be revisited from history, continued with a follow-up, or exported as PDF. This module preserves the scope and output for later review or sharing; it does not turn an AI finding into verified customer research without human checking.

The merchant determines which supported clothing, footwear, or accessory products can offer try-on and provides suitable product imagery. This setup gates the entire shopper flow; unsupported categories should remain in normal product guidance.

An eligible product can surface a proactive prompt that invites the shopper into the try-on flow. Its role is to make the capability discoverable at the product-decision moment rather than requiring the shopper to know a special command.

The shopper uploads a suitable photo in chat, and the system combines it with the eligible product image to create a temporary visualization. Image quality and supported-product boundaries matter; the flow should request a better input or fall back when those conditions are not met.

After seeing the image, a shopper can request another color or variant and then use the in-chat Add to Cart action. If the item is unsupported or the image cannot be used, the agent returns to standard product guidance instead of presenting a fabricated try-on result.