The hero says "give your team and customers AI-powered analytics," which is two different use cases in one sentence. The body then runs all three narratives in parallel: open-source deployment instructions, embedded SDK pitch, and self-serve BI features share equal real estate with no hierarchy. A visitor who landed because they're evaluating BI tools for their internal team has to mentally filter out the embedded analytics pitch; a visitor who came to embed analytics in their SaaS product has to wade through self-serve dashboard copy that isn't for them. Pick the buyer the homepage is built to convert first — the evidence in the testimonials (a CTO and a co-founder both describing internal team use) points to the self-serve startup buyer — and move the embedded analytics narrative to its own dedicated page, which already exists at /product/embedded-analytics.

"Open source" is one claim. "AI-powered" is another. "Deploy in minutes" is a third. "From pre-seed to post-IPO" is a fourth. These are four different reasons to choose Metabase, and the page treats them as equally weighted. The result is that the visitor finishes reading without knowing what Metabase is actually the best at. Tableau owns "visualization depth." Looker owns "governed metrics for enterprise." Apache Superset owns "free open source." The footer even lists comparison pages against all of them, which means Metabase is implicitly competing on every axis simultaneously. The page needs to pick one axis — the strongest candidate, given the testimonials and the 90,000-company number, is "the BI tool that non-technical teams at startups actually use" — and make every section defend that claim instead of adding a new one.

"Query data without writing code / Anyone can explore data, visualize results, and answer their own questions. Ask Metabot AI for conversational answers" appears, then two sections later: "Query data in plain English with AI / Anyone can explore data, visualize results, and answer their own questions. Ask Metabot AI for conversational querying." The "Guide people to the right answers" section is also repeated nearly word-for-word. A visitor scrolling the page hits the same content twice and reads it as either an error or a sign the product doesn't have enough to say. Cut one full set of these duplicate sections; the page is currently longer than it needs to be and the repetition undermines the credibility the testimonials are trying to build.

"Trusted by 90,000+ companies" is a significant number — it's the kind of number that ends competitive evaluations — but it appears as a small trust badge and then the page moves on. The two named testimonials (Derrick Mar at Pathrise, Peer Richelsen at Cal.com) are the most credible elements on the page, but they're positioned mid-scroll after three feature sections. A visitor who is comparing Metabase against Tableau or Looker needs to see the scale of adoption and a named customer's outcome before they read a single feature description. Move the 90,000-company figure and one named testimonial into the first thing the visitor sees below the hero; the current structure makes the visitor earn the proof instead of leading with it.