Productboard Spark — lessons from building AI products

Ravi Mehta’s guest post from the Productboard product team is one of the more practical accounts of building AI features that ship to paying customers. The piece covers the evolution from Productboard Pulse (customer feedback analysis) to Spark (AI-powered product management assistant), with candid detail about what worked and what required rethinking.

Key insights

Two observations stand out. First, the distinction between guiding and automating: users say they want AI to do tasks for them, but the real value comes from AI that helps them think better. Productboard learned this through user research — when they automated entire workflows, adoption dropped. When they restructured AI to surface insights and suggest next steps while leaving decisions to the PM, engagement increased.

Second, the concept of AI quality as a “third dimension” in product development. Traditional features have two dimensions — UI and backend logic — with relatively predictable development timelines. AI features add a third: model quality. The team found they could build the UI and backend in two weeks, then spend two months iterating on AI output quality before reaching the threshold where users trusted the feature.

Why this matters

These are not theoretical frameworks. They come from a product team that shipped AI features to thousands of product managers and observed real usage patterns. The guiding-vs-automating insight alone can save teams months of building the wrong thing.

The honest timeline — two weeks for the “product” part, two months for the “AI quality” part — is valuable for PMs estimating AI feature development. Most roadmaps dramatically underestimate the iteration cycle required for AI output quality.

Who should read this

PMs building AI-powered features in SaaS products, particularly those working on knowledge work tools where the AI assists human judgment rather than replacing it.