From Clunky to Conversational: Redesigning a Support Bot for an Early-stage SaaS

AI · 5 min read

WorkHub, a project-management SaaS, launched with a scripted support bot that returned rigid, often irrelevant answers. The redesign replaced brittle scripts with a hybrid model: a lightweight LLM for conversational phrasing plus a deterministic retrieval layer for precise product help and linkable KB answers.

Designers focused on three UX patterns: clear confidence signals when the bot was certain, visible fallback actions (like ‘connect to human’) when uncertain, and session summaries that provided next steps and ticket IDs. The product decision also limited the bot’s scope explicitly to routine workflows to avoid hallucinations in high-stakes issues.

After deployment, first-response query resolution rose 34%, and NPS for support interactions increased by 11 points. The study underscores that adding generative capabilities is most effective when constrained by reliable retrieval and when the UX communicates limitations and next actions plainly.