A customer knowledge base only delivers value if it is fresh, searchable and multilingual. In most companies, KB articles are written manually once and never updated — so customers search, find stale answers, and open a ticket anyway.

Ticket deflection fails when the KB does not match real customer questions. Utilyx models the KB as a living workflow: AI auto-generates articles from resolved tickets, semantic search serves answers in 7 languages, and analytics track which articles actually deflect tickets.

The workflow is structured as a decision tree with 3 branches (auto-draft from ticket, gap-topic backlog, negative-vote review) and 7 steps end-to-end. Every action is logged in the kb_audit subtable. Customers find answers themselves only 20% of the time — AI KB deflection targets 60%.

• Step 1 (Start — Ticket resolved): when a support ticket is closed, the workflow captures case_id, ticket_category, resolution_text, customer_id, agent_id. The copilote IA extracts the problem statement, root cause and resolution steps.

• Step 2 (AI draft generation): the copilote IA drafts a KB article in 7 languages from the case content — title, summary, step-by-step resolution, tags, related_articles[]. The draft is routed to the knowledge owner.

• Step 3 (Branch — content decision): Branch A (new topic): the draft is published as a new article after owner review. Branch B (existing topic update): the draft is merged into an existing article via versioning (diff view). Branch C (duplicate): the draft is discarded and linked to the existing canonical article.

• Step 4 (Semantic search): customers and agents search the KB in natural language via sql-query against the kb_article_index (vector embeddings); semantic search returns the top 5 most relevant articles across all 7 supported languages, ranked by relevance score — not keyword matches.

• Step 5 (Self-service deflection): when a customer starts a ticket via the portal or chatbot, the workflow suggests 3 matching KB articles before submission. Sub-branch A (article resolves): the customer confirms, the ticket is deflected and deflection_event is logged with article_id, customer_id, channel. Sub-branch B (still need help): the ticket proceeds with the suggested articles attached for agent context.

• Step 6 (Feedback loop): after reading an article, the customer votes helpful/unhelpful. Negative votes trigger a review task for the knowledge owner with the article_id and verbatim comment. The kb_gap_analysis subtable logs searches with no matching article to build the backlog.

• Step 7 (Analytics & archive): the dashboard tracks article_views, helpfulness_rate, deflection_rate, gap_topics[] per week. Article versioning is archived in useArchive with index fields article_id, version, author, publish_date for compliance.

Teams using Utilyx report a 40% reduction in ticket volume thanks to AI-powered self-service deflection, and a 50% faster time-to-answer for customers who find their solution directly. The deflection rate climbs from 20% to 60% as the KB matures on real ticket content.

The no-code visual designer lets a knowledge manager own the workflow without developers, and the AI copilot auto-generates article drafts in 7 languages from resolved cases — 60% faster deployment at 70% lower TCO than legacy KM suites. The gap-topic backlog ensures the KB stays aligned to what customers actually ask.

Semantic search across 7 languages via vector embeddings means a single article base serves a global customer base. Article versioning ensures every change is auditable and reversible for compliance.

Gap analytics close the loop between what customers ask and what the KB covers, so the knowledge base stays aligned to real demand, and the kb_gap_analysis subtable prioritizes the next articles to write.