Caching the questions that repeat
A large share of first-contact WhatsApp messages are the same handful of questions asked in slightly different words. The first inbound message of a new conversation is checked against a cache of previously-answered questions before calling Claude at all — an exact match returns instantly, with no AI call and no added cost.
Scoped so it can never leak between customers
Only the very first message of a brand-new conversation is eligible for a cache hit — before any customer-specific context exists — and only for general enquiries, never a booking, order, or lead reply, which is specific to whoever asked it.
Where this is headed
Today's cache matches on exact wording. Semantic matching — catching rephrased questions like "do you deliver?" and "can you ship this to me?" as the same question — is on the roadmap, gated on choosing an embedding provider.