SleekAI ingests SupportCandy's wpsc_ticket table, ticket replies and custom fields into an OpenAI Files vector store. A customer asking about a refund for order #2491 hits past ticket #1820 where another customer worked through the same Stripe refund flow. The bot summarizes the resolution in 1-3 sentences and cites the ticket number, so the answer is verifiable rather than plausible. Tickets that an agent closed as resolved get a ranking boost, so verified answers come first.

Display conditions scope a bot per SupportCandy category, so a Billing category can run a different system prompt than a Technical category, each drawing only from its own ticket history. Custom fields on tickets, things like order ID, payment method or affected version, are part of the prompt context. That lets the bot answer 'how long do Stripe refunds take?' with the actual policy from your past tickets rather than a generic 5-10 day estimate. Smart escalation calls SupportCandy's REST endpoints to open a real ticket pre-filled with the chat conversation, the category and any custom fields the visitor mentioned.

Agent-side bots are supported, so a logged-in agent can ask the same bot to summarize a long ticket thread or suggest a draft reply from similar resolutions. Multilingual setups pair with WPML, Polylang and TranslatePress so the bot replies in the visitor's language even when the source ticket was written in another. Logs land alongside the cited ticket, giving team leads an audit trail and a worklist of recurring issues that should be turned into docs.

SupportCandy installs accumulate ticket history fast. By month six, most teams already have the answer to the majority of incoming questions sitting in resolved tickets, but no customer is going to scroll through 300 past tickets to find theirs. A generic chatbot does not know wpsc_ticket exists, ignores your custom fields, and cannot tell a 2022 ticket about a long-fixed bug from a 2024 ticket about the current one.

A ticket-trained bot reads the actual archive, surfaces the canonical resolution for the most common questions, and cites the ticket number so the answer is auditable. The economics are clear: refunds, shipping, payment confirmations and account questions all have exact precedent in your tickets. Each deflected tier-one ticket is 10-20 minutes back to the team, and each escalation arrives with the conversation, the order ID and any custom fields already attached.

The bot's job is not to replace agents but to remove the predictable questions from their queue, so they spend more time on the genuinely new ones. Logs become a worklist of recurring issues that should be turned into docs or auto-closed templates.