Using AI to Go From User Insight to Better Backlogs - Mike Cohn

Using AI to Go From User Insight to Better Backlogs - Mike Cohn

AI is rapidly changing how product teams work—but the biggest opportunity isn’t replacing product thinking. It’s reducing the friction between understanding users and turning those insights into high-quality backlog items.
To make the ideas concrete, I use a consistent example throughout: a team building software for valet-attended parking garages, initially selling to independent operations like boutique hotels. Each step builds on the previous one, showing how AI outputs can feed naturally into your existing agile practices.
With a straightforward prompt, AI can help you build a detailed persona—including hopes, concerns, emotional triggers, and decision criteria. In my example, the persona that emerged was a garage owner/operator with high staff turnover, contract-renewal anxiety, and a strong desire for predictable labor costs. Several of these insights are things I might have missed or deprioritized on my own.
Understanding a persona’s aspirations—not just their functional needs—turns out to be especially valuable.

Once a persona exists, you can ask AI to role-play that person and let you interview them. This is not a replacement for real user interviews, but it’s a great way to explore assumptions, test questions, and uncover gaps in your thinking.
AI is also excellent at preparing interview guides for real users who match a persona. With the right prompt, it can generate a structured guide that covers:

• Opening context (confidentiality, purpose, time commitment)
• Current workflows and pain points
• Desired future state and success criteria
• Constraints (including regulatory or operational)
• Thoughtful wrap-up questions

Looking at the results, I was struck by how much better prepared I could have been for many interviews over the years if I’d had this kind of support.

Once you’re ready to move into backlog work, AI can help generate user stories and job stories that follow well-established agile guidance.
By being explicit in the prompt—format, INVEST criteria, and output rules—you can get clean, ready-to-use stories that are easy to import into a backlog tool. AI can also correctly choose between user stories and job stories depending on whether the situation or the role is more important.
In the valet parking example, this resulted in stories about vehicle handoff tracking, damage-claim protection, wait-time monitoring, staff accountability, and remote visibility into operations.

I prefer to add acceptance criteria as a separate step, and AI handles this easily. You can ask for:

• A simple bullet list (great for user reviews), or
• Gherkin (given-when-then) format for more formal specification
  

You can even convert between formats later. Either way, this step quickly raises clarity and testability.

AI isn’t just for generating content—it’s also useful for critique.
With a structured prompt, AI can evaluate user and job stories against the INVEST criteria, identify only what’s missing, explain why, and suggest a focused improvement. This works whether the stories were written by AI or by you.
Over time, you can even build a library of good and bad examples to further improve the quality of feedback you get.

AI won’t replace talking to users, making judgment calls, or exercising product sense. What it can do is help teams move faster from vague ideas to concrete artifacts, surface blind spots, and raise the baseline quality of their work—especially when time or experience is limited.
Used well, AI becomes a tireless collaborator: one that remembers persona details, never gets impatient with rewrites, and can move effortlessly from big-picture thinking to precise backlog items.
The key mindset shift is this: don’t ask whether AI can replace parts of product discovery or backlog refinement. Ask how it can help you arrive better prepared for the conversations that still matter most.