Sylvi Watches runs discovery, warranty and repair through one memory

A watch is bought slowly and owned for years, which creates two workloads at once. The pre-sale side is spec-led and budget-led: Product Discovery was the largest intent at over 2,500 chats, and the recorded theme is budget-led shopping for college and work, with shoppers opening on collection queries like "Help me find something in Watches Under 2000".

The second workload never stops. Repair and warranty chats ran to 291 across the window, with a further 94 return and replacement requests. Sylvi is the only brand documented here with a standing repair queue, and it sits in the same inbox as someone choosing a strap.

Traffic is diffuse and mostly silent. Sylvi draws one of the largest browsing audiences in this set from marketplace and social referral, and only 3.34% of sessions start a chat. The ones that do have to carry more weight.

The recorded objections are unusually useful as product feedback. The agent sometimes could not supply concrete dimensions (bezel size, case thickness, display behaviour) when a shopper asked. Colour-match and out-of-stock situations pushed shoppers toward other collections. Budget requests below the available range left the agent with limited closest-match options. And image or audio-only inputs frequently stalled the assistant, which on a visual category is a real gap.

Website Agent, WhatsApp AI, Instagram AI and Messenger AI run on a single shared customer memory, the widest channel footprint among the brands documented here. A customer who asks about a strap on Instagram and later messages on WhatsApp about a warranty is recognised as the same person.

The agent was trained on the collection structure and the specification detail that decides a watch purchase (case size, strap material, movement type, water resistance) and separately on the service path: warranty registration, the one-year cover, and what qualifies for repair versus replacement.

Routing between the two is the design point. A discovery chat stays with the agent; a repair that needs a physical assessment reaches a human quickly, which is why Sylvi's resolution rate deliberately sits below the highest in this set.

The Nudge Engine works the large silent audience, a browsing visitor who will never open a chat window may still respond to a product-aware prompt.

Four channels, two workloads, one memory. Three decisions shape the account.

Shared memory is the point, not channel count. Four inboxes with four separate histories is worse than one inbox. The value is that a chat started on Instagram continues on WhatsApp without the customer repeating herself, which matters most in the service tail, months after purchase.

Give warranty and repair their own path. 291 repair and warranty chats plus 94 return and replacement requests. A brand that routes service into the same undifferentiated queue as product questions will either frustrate owners or bury sales enquiries. Sylvi does neither.

Fix the spec data, the agent exposed the gap. The clearest finding in this account is a product one: shoppers asked for bezel size, case thickness and display behaviour, and the agent could not always answer because the data is not on the product pages. That is a catalog completeness problem the chat layer made visible.