The agent successfully assembled a comprehensive overview of Innovorder's business, positioning, and pricing model, but had to work around a website that actively obscures pricing details and lacks standard navigation paths. The site is primarily French with thin English content, and critical pages like /pricing return 404s—forcing the agent to discover the correct URL (/en/price) via web search and to supplement homepage content with external sources (GetApp, SoftwareAdvice) to answer comparative and pricing questions.

• ›Step [1] (homepage) and [3] (English homepage) provided core positioning and meta descriptions but no structural nav to subpages; the agent had to guess or search for next URLs. Steps [2], [4], [5], [6], [7], [8], [9] all returned 404s on plausible paths (/tarif, /pricing, /solutions, /features, /about, /produits, /products), revealing a non-standard site structure.
• ›Step [11] (/en/price, discovered via web search in step [10]) was the only official pricing page found; it exists but is not linked from the homepage or discoverable through normal navigation—the agent had to use prior knowledge of competitor URL patterns to even search for it.
• ›Steps [12] (FAQ), [13] (e-wallet), and [16] (self-ordering-kiosk) were successfully fetched and provided feature details, integrations, and use-case information, but these are siloed pages; no clear product taxonomy or menu connected them.
• ›The agent relied heavily on web search (steps [10], [14], [15]) and external sources (GetApp, SoftwareAdvice, G2, blog posts) to fill gaps that the website itself should have exposed: clear pricing tiers, feature comparison, competitor differentiation, and customer social proof. These external sources account for much of the final answer's completeness.
• ›The website's language-switching (Webflow-based) works technically, but English content is thinner than French; some pages may not have English translations, and the agent had to navigate in both languages to find all information.
• ›No structured data (schema.org, JSON-LD) was detected in the response snippets; pricing, features, and alternative information had to be extracted from prose or inferred from external reviews.