Turn scattered website docs into an on-site AI support agent

Many websites—especially SMBs, SaaS vendors and ecommerce businesses—struggle to make scattered help content across knowledge bases, blog posts, product pages and changelogs useful for real-time customer questions, which increases support costs and hurts conversions. This is addressable at scale: with roughly 100M business websites and an estimated $20.0B addressable market (100M sites x $200 ARR), even modest improvements in self‑service adoption move meaningful revenue and ticket-volume metrics. You could build an on‑site AI support agent that crawls and indexes a site’s structured and unstructured docs, uses LLM-enabled retrieval to provide context‑aware, citation-backed answers on the page, and exposes actions like opening a ticket or linking to purchase flows. Focus on low‑friction deployment (WordPress/Shopify plugins + one-line JS), an admin console for editable responses and analytics, privacy modes (on-device or customer‑controlled retention), and a pricing path that supports self‑serve SMBs with enterprise upsells—targeting the $200 ARR per site economics. This market is attractive now because retrieval-augmented LLMs materially reduce hallucinations, conversational UX is mainstream, and SMBs are shifting to self‑serve subscriptions—so buying friction is low while expected value is high. Competition is medium; you can differentiate by prioritizing accuracy and citations, vertical templates, and stringent privacy/compliance controls rather than attempting to out‑feature large incumbents. Be honest about the key risks: controlling hallucinations, keeping content fresh across diverse CMSs, and building defensible integrations—success will depend on execution in retrieval quality, demonstrable ROI for buyers, and a frictionless install-and-forget user experience.