- AI-powered onboarding
- Notebook UI transformation
- The 'too easy' dilemma
- Competitive user research
In the competitive landscape of B2B SaaS, user onboarding is a make-or-break moment. Raziel Einhorn, a Product Manager at Pekan AI, shares their transformative journey tackling the pervasive 'blank page problem' – where new users struggle to initiate engagement with a complex product. Their story is a testament to innovative problem-solving, leveraging cutting-edge AI, and the critical importance of deep user understanding.
Pekan AI, a platform specializing in predictive analytics for businesses, initially faced significant hurdles with user onboarding. Despite attracting a target audience through a product-led growth (PLG) strategy, their SQL console interface and reliance on templates led to a dismal 5% conversion rate. Users were getting stuck, unable to bridge the gap between their business problems and the product's generic solutions, highlighting a fundamental mismatch in mental models.
The turning point came with a dual strategic pivot: adopting a data notebook UI, inspired by tools like Jupyter and Hex, and deeply integrating Large Language Models (LLMs). Instead of merely offering a chat pop-up, Pekan AI's LLM was designed to generate entire data notebooks, guiding users from a vague business question to a fully structured, model-ready dataset. This approach, validated through rigorous competitive user research on rival products, proved to be a game-changer.
The results were dramatic: Pekan AI saw its conversion rate skyrocket from 5% to 40% for its target users. However, this success unveiled an unexpected challenge – the 'too easy' problem. While users were successfully training models with minimal effort, qualitative feedback revealed a lack of true understanding. Users, empowered by one-click solutions, often couldn't edit or iterate on their models, resorting to restarting the entire process, a friction point not immediately visible in quantitative metrics.
Raziel Einhorn emphasizes several key takeaways from this journey: the imperative to ship fast and iterate, the invaluable insights gained from paid user interviews (combining quantitative data with qualitative depth), and the nuanced understanding required for co-pilot design. The lesson learned is that a co-pilot must truly empower the user, serving as a 'pilot' rather than just an assistant, and must cater to diverse learning styles to ensure genuine comprehension and long-term engagement.
“The risk was, in a place where you actually want, our goal is not just for our users to do the work, but we also want to empower our users.”
- Raziel Einhorn, Product Manager at Pecan AI
