Productboard: doing product discovery with AI without losing the human element

Productboard’s March 2026 article addresses a practical question for product managers: if AI can automate discovery tasks, which ones should it handle, and where should humans stay in charge? The framework they propose organizes the answer around three buckets — gathering, analysis, and framing.

Gathering covers pulling signals from multiple sources: usage data, sales calls, support tickets, market research, and competitive intelligence. AI is well-suited here because the volume is high and the value of each individual signal is low. A PM reading every support ticket is expensive; an agent that clusters them and surfaces emerging themes takes seconds.

Analysis sits in the middle. AI can group themes, separate what customers say from what they actually do, and quantify the relative frequency of problems across a dataset. The article draws a useful distinction: AI is good at volume-based pattern recognition, but deciding whether a frequently requested feature fits the roadmap — or contradicts it — still requires a human with context.

Framing and shaping is where the article gets most specific. It describes a three-part narrative structure for building a product business case: the villain (the problem and its cost to the customer), the protagonist (a specific customer with documented evidence), and the stakes (what happens if nothing changes). This structure is useful not because AI generates it — it doesn’t — but because having a clear template helps AI pull the relevant supporting evidence from unstructured data faster and more accurately.

The article estimates that most product teams have between 40 and 60 steps in the process from insight to launch, including data gathering, analysis, review cycles, prioritization sessions, and handoffs. The argument is not to automate all of these, but to identify the steps that are high-frequency, low-judgment, and well-defined — those are where AI saves meaningful time without introducing meaningful risk.

The piece avoids the common trap of treating AI as a general-purpose productivity accelerator. The goal it articulates is more time for actual product thinking, not just faster execution of the same process. The guidance is specific enough to be actionable without being so platform-dependent that it only works inside Productboard’s own toolset, though Spark, their AI platform, is mentioned as one implementation path among several.

Useful for product managers who are past the curiosity stage and want a structured approach to deciding where in their current workflow to experiment with AI first.