Product Management Blog – Pankaj Bisht

Use AI to write PRDs, analyse feedback, and validate ideas

I’ll be honest. When AI tools first became popular, I wasn’t excited — I was sceptical.

After years in product management, I’ve seen plenty of so-called game changers that didn’t change much at all. So I took my time before really trying them.

But then I started using AI in my daily work, and I used it to tackle real day-to-day problems, like:

* Messy stakeholder feedback

* Half-baked user stories

* Unfinished requirement documents

* Simplifying technical jargon

* Ideas that needed quick validation before starting discovery

AI works best as a structured thinking partner, not a magic wand. One of the biggest productivity gains I’ve seen is in documentation, particularly when drafting PRDs.

---

You don’t need a week to write a PRD

Every PM has faced this moment: discovery is complete, stakeholders have agreed, engineers are waiting, and you’re looking at a blank page, trying to organise your thoughts without inviting a flood of questions in the next refinement session.

Here’s a prompt that helps:

You are an experienced senior product manager supporting a cross-functional delivery team (engineering, design, QA, data, stakeholders).

I will describe a feature. Create a clear, structured Product Requirements Document (PRD) using plain language. Be specific and practical. Stay clear of jargon.

Include these sections:

* Feature Overview — what it is, who it’s for, why it matters

* Problem Statement — user pain and business impact

* Goals

* Users and key use cases

* User Stories (As a… I want… so that…)

* Acceptance Criteria (clear and testable)

* Success Metrics (how success will be measured)

* Edge cases along with risks

* Open questions and assumptions

If information is missing, make reasonable assumptions and label them clearly. Format with clear headings and bullet points.

The results aren’t perfect, but you get an 80% draft in about fifteen minutes instead of several hours.

Instead of writing from scratch, you start by refining, and that’s a big shift. In practice, it reduces documentation effort by around 60–70%.

---

Analysing Customer Feedback

If you’ve ever been handed a spreadsheet with 500 rows of NPS comments and told to “find the themes,” you know how tough it is. It’s important and doable, but it can be exhausting.

My current AI approach is as follows:

I will paste customer feedback from our NPS survey. Analyse all responses and generate structured insights.

1. For each individual comment:

* Assign a primary theme (and secondary theme if relevant)

* Identify the sentiment (positive, neutral, negative, mixed)

* Flag if it indicates an urgent or high-risk issue (with reason)

* Map it to the relevant product area or feature

2. Then aggregate insights by theme:

* Number of comments per theme

* Overall sentiment trend

* Key customer pain points or drivers of satisfaction

* Notable patterns or repeated concerns

3. Recommend actions:

* Recommended product improvements or investigations

* Priority level (high/medium/low) based on frequency and severity

4. Output format:

A detailed table with one row per feedback item showing:
Feedback | Theme | Secondary Theme | Sentiment | Urgency | Product Area | Notes

A theme summary table showing:
Theme | Volume | Sentiment Summary | Key Insight | Recommended Action | Priority

If feedback is ambiguous, infer the most reasonable interpretation and note the assumption.

I paste in a batch, usually 50 to 100 comments at a time, and within seconds, I get a structured breakdown that would have taken me half a day to do by hand.

Is the categorisation always perfect? No. Sometimes it groups things in strange ways or misses the details.

I’ve also started using AI to compare feedback from different time periods. I feed in Q3 and Q4 data, ask it to find new themes or changes in sentiment, and I have the start of a quarterly insight report without the usual two-week delay.

Where it saves time: finding themes, analysing sentiment, and comparing different periods, what used to take two days now takes just two hours.

---

Validating Ideas Before You Build Anything

This stage is especially valuable, and I believe many product managers underestimate AI’s potential here.

Before moving an idea into discovery, I use a ‘stress test’ prompt:

I’m considering building [feature].
Target users: [description]
Problem being solved: [problem]

Analyse this idea critically from a product discovery perspective. Be direct and evidence-driven.

1. Failure Analysis
What are the top five reasons this idea could fail? Consider:

* User value proposition and desirability

* Usability or behaviour change required

* Adoption barriers

* Revenue/Cost factors

* Technical or operational complexity

* Competitive or market factors

2. Assumption Mapping
List the key assumptions behind this idea and classify each as:

* critical/important/minor

* currently validated/untested/weak evidence

Identify the highest-risk assumptions.

3. Validation Plan (before engineering investment)
For the highest-risk assumptions, recommend:

* What to test

* Fastest way to test it (experiment, interview, prototype, data analysis, etc.)

* Success criteria

* Estimated effort level (low/medium/high)

* Priority

4. Alternatives
Are there simpler or lower-risk ways to solve the same problem?

5. Recommendation
Based on the risks and evidence gaps, should we:

* Proceed

* Validate first

* Redesign the concept

* Abandon the idea

Explain why.

The output consistently surfaces blind spots. It asks questions I should have been asking myself. It’s as if having a partner who’s read every product management book but has no political agenda.

I also use AI to draft survey questions, create interview scripts, and simulate responses from different user personas. While this does not replace direct user research, it accelerates the preparation and improves research quality.

AI saves time in assumption mapping, risk identification, and research preparation, reducing early-stage planning from days to hours.

---

A Few Things I’ve Learned Along the Way

1. Be extremely clear in your prompts. Garbage in = garbage out. If you set the context very clearly (who it is for, what your constraints are, how you want it formatted), you’ll get WAY better results. Clarity is the multiplier.
2. Don’t abdicate your judgment. AI is simply a tool; it does not have decision rights. It doesn’t know how your team works together, what your company’s risk tolerance is and it sure doesn’t have any idea of the political and human factors at play. Leverage it to generate ideas, but ultimately, you are responsible for the decision.
3. Iterate, don’t settle for the first answer. Treat the AI as you would any other person you’re bouncing ideas off. Accept the first output as a starting point only. Ask follow-up questions, ask for clarifications and push it to re-frame.

FYI: My team knows I’m using AI to be more productive. I want them to know so they trust what I’m doing and understand I’m not trying to cut corners, I’m trying to deliver better.

---

Top 10 AI Use Cases for PMs:

1. Extract Themes from user interviews, NPS follow-up comments, or any type of qualitative feedback.
2. Write User Stories and Acceptance Criteria: Help turn quick ideas into well-framed stories with clearly defined acceptance criteria.
3. Feature prioritisation help: Let AI be your partner when you’re trying to compare trade-offs and quickly identify high-impact features.
4. Customer interview questions: Have the AI propose targeted questions that you can ask customers to gather meaningful insights.
5. Define MVP scope: See what the smallest number of features you can test your assumptions with.
6. Competitive analysis: Save time summarising what features your competitors are offering and what’s trending in the industry.
7. Create journey maps and opportunity solution trees: Quickly visualise customer flows and opportunities.
8. Risks: Identify potential risks along with proposed mitigation for each.
9. Survey analysis/sentiment summarisation: Turn both quantitative and qualitative survey data into actionable insights.
10. Meeting summaries & Decision Logs: Reduce admin work and ensure everyone is aligned on next steps.

---

The Real Shift

Over the next few years, the PMs who will excel aren’t the ones who know their PRD in and out or spend all weekend filtering through support tickets alone. It’ll be those who leverage strong product thinking with the right tools to work faster and make better decisions.

AI is not going to make you a better product manager overnight. What it can do is remove the friction that’s stopping you from doing the things that require your unique talents and skillsets: deeply understanding your users, making difficult decisions, and executing.

TL;DR: If you’re using AI to work smarter on product work, I want to hear about it. Haven’t tried it yet? Pick one task this week, give it a go, learn, and share what you discover.

---

PS. If you’re a PM trying to learn how AI can fit into your work, let’s talk. I’m always happy to swap prompts.