Hard-to-find emails — voice-enabled semantic search across inboxes

Many knowledge workers—roughly 500 million globally—still spend hours each week hunting through email to recover past decisions, attachments, or context, and keyword-based provider search often misses semantically relevant threads. The resulting time waste and operational risk are concentrated in roles like sales, legal, customer support and program management where historical email is a primary institutional memory. You could build a voice-enabled semantic search layer that indexes multiple inboxes and attachments into encrypted vector stores, exposing natural-language and follow-up query capabilities across accounts with connectors to major providers and SSO for enterprises. Architecturally the product would combine incremental indexing, compact embeddings, and customer-controlled encryption or on-device indexes to keep latency low and minimize third-party data transfer. This is an attractive moment: recent advances in embeddings and retrieval tooling have materially lowered costs and latency, hybrid work has increased reliance on email as knowledge, and the addressable market is in the order of $30B (about $60 ARPU/year), indicating meaningful revenue potential. You can stand out by prioritizing privacy-first design (customer-managed keys or local indexes), enterprise compliance and data-residency options, and a voice-first UX tuned to threaded conversational retrieval rather than document search. Be honest about the hard parts: building and maintaining secure connectors, proving relevance across noisy inboxes, navigating provider terms and enterprise procurement, and investing in compliance will be necessary upfront investments before scaling revenue.