Taskly Uses LLMs to Personalize Onboarding, Cutting Setup Time by Half

AI · 6 min read

Taskly, a productivity startup, struggled with high abandonment during multi-step project setup. Traditional guided tours were ignored, and templated wizards failed to match varied workflows. The team experimented with a lightweight conversational assistant powered by a fine-tuned large language model to ask a few clarifying questions and generate a tailored project skeleton.

Design prioritized brevity and transparency: the assistant asks three focused questions, shows the proposed project structure before creating anything, and offers an option to opt out. Privacy choices were explicit — data used in prompts stays in-session unless the user consents to store templates. The engineering team built a hybrid pipeline combining on-device prompt handling for interface text with server-side LLM inference for synthesis.

A randomized trial showed onboarding completion time dropped by 51 percent and time-to-first-task fell by 47 percent. Users rated the experience higher on perceived control and relevance, while the support team saw fewer setup-related tickets. The case highlights how conversational AI, when constrained to specific decisions and paired with clear user control, can accelerate setup without eroding trust.