luxury-tone AI: how to train an agent for jewelry.
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. In jewelry, a tone that sounds like a generic chatbot ruins the brand's perception before the customer even gets to the price. This article details how to configure an agent that sounds like an advisor, not automated support, for high-value sales.
Tune formality and humor with sliders, not trial and error
The agent is trained by uploading your brand guidelines and examples of real responses, and adjusting formality, warmth and humor with simple sliders from the panel. For jewelry, that usually means high formality and humor close to the minimum: an agent that treats the customer warmly but never casually, with no extra emojis or out-of-place jokes in a conversation about a piece worth several thousand dollars. The outcome doesn't depend on rewriting the prompt by hand every time something sounds off.
Hard brand rules in the knowledge base
Beyond tone, you define explicit rules in the trainable knowledge base: what is never said (unauthorized discounts, comparisons with other brands), what is always clarified (certification, warranty, provenance) and how each product line is described. The agent answers within those limits; if a question takes it outside them, it doesn't improvise an answer that could compromise the brand — it hands it off.
Immediate escalation on high-value sales
With escalation rules and keywords, you configure that any conversation about a piece above a certain value, or with clear purchase intent, goes to a person on the team immediately — not as a replacement for the agent, but as the point where a big sale needs a human touch. Steff assembles the full context before escalating (what the customer asked, which piece they're interested in) so the advisor comes in ready to close, without repeating questions already answered. With copilot mode, that same person can take over the conversation in one click and the bot pauses itself.
Payment-proof verification with vision
In high-value purchases, a bank transfer is common instead of a card payment. Steff has vision: when the customer sends the photo or PDF of the transfer receipt, the agent reads it and confirms the payment details within the same conversation, instead of asking the customer to wait for someone to check the email manually.
Private appointment booked inside the chat
For pieces the customer wants to see in person before deciding, built-in scheduling lets them book a private appointment at the showroom or the jewelry store, synced with Google Calendar, with an automatic reminder before the visit — all inside the same conversation where they were already asking about the piece.
Because scheduling lives in the same agent that already handled the initial inquiry, the appointment arrives with context: the advisor who receives the customer at the showroom knows in advance which piece caught their eye, what they asked and what budget they mentioned, instead of starting the conversation from scratch in person.
Choosing the AI model for a luxury brand
heysteff lets you choose the language model per workspace (Anthropic, OpenAI, DeepSeek or OpenRouter). For a brand where tone is part of the product itself, it's worth spending time testing which model, combined with your formality sliders and your brand rules, sounds closest to how your sales team actually speaks — not every model interprets "formal but warm" the same way.
◆ the star flow
The immediate escalation with context already assembled is what separates a well-configured luxury agent from a generic one: the big sale always reaches a person, with all the groundwork already done.