Area: mobieusAI (audit p11) · Surface: mobieusHelp reply suggestions (HelpdeskAI::replySuggestions, flag helpdesk_ai_reply_suggest / HD-1) · Dimension: competitor-gap · Severity: major
Zendesk's reply suggestions and Intercom Fin draft replies grounded in the help center plus historically successful resolutions, so the agent gets an answer that reflects the org's real prior solutions. mobieusHelp drafts a reply purely from the open ticket text, so it cannot reuse the team's established answers or canned macros and risks inventing a resolution the org never offers. The product already has canned responses (a canned-response generator exists, HD-5) and a KB (mobieusKnow) — they just aren't fed into the reply suggester.
Evidence
platform/src/Services/Helpdesk/HelpdeskAI.php:43-48 — `replySuggestions()` calls `$ctx = self::buildTicketContext($ticketId)` then `TextGenerator::generate('helpdesk_ai_reply_suggest', $ctx, ...)`. buildTicketContext (HelpdeskAI.php:204) assembles ONLY the current ticket's thread. By contrast autoCategorize() (HelpdeskAI.php:69-75) does inject extra context (valid help topics + queues), proving the pattern is available but unused for replies. No query against resolved tickets, canned responses, or knowledge_pages is made in the reply path.
Suggested fix. Before generating, retrieve the top-N most similar resolved tickets and the top matching mobieusKnow articles/canned responses, and pass them as labeled context (mirroring autoCategorize's topic/queue injection). Instruct the model to prefer the org's established answers and cite which canned/KB source it drew from.
Filed by the automated tenant-app audit and adversarially evidence-verified. Status: verified. Open — not yet actioned.
Patrick Bass
@mobieus