A Proven AI PRD Template by Miqdad Jaffer (Product Lead @ OpenAI)
Build and scale AI-powered products successfully. A Product Requirements Document template with an AI PRD case study for Shopify Auto Write.
Hey, Paweł here. Welcome to the free archived special edition of The Product Compass Newsletter.
With 103,731 Product Managers from companies like Meta, Amazon, Google, and Apple, this newsletter aims to be #1 source for learning and growth as a PM.
Expect actionable tips, templates, resources, and insights.
No fluff. Every week.
Consider subscribing and upgrading your account for the full experience:
In 2025, the AI market is estimated to be around $638 billion. It’s expected to reach $1,811 billion in 2030 with a CAGR (compound annual growth rate) of 36.6%.
It’s an excellent opportunity for AI Product Managers. In the U.S., Senior AI PMs already earn $259K per year on average.
But not every AI implementation is successful.
Given the hype around AI, many implement AI without a clear, justified business case. To make things worse, AI-specific considerations are often overlooked.
Today’s guest,
, works as a Product Lead at OpenAI, helping OpenAI’s partners succeed. He previously worked on AI as the Director of Product at Shopify.For the first time, Miqdad shares his AI PRD (Product Requirements Document) template, tested in the trenches, that:
Addresses common problems with AI implementations.
Provides critical guidance to help you make better decisions.
Can be applied to most AI products and features (especially LLM-powered).
It will help you implement and scale AI-powered solutions successfully.
We discuss:
CHAPTER I: How to Work With the AI PRD
CHAPTER II: AI PRD Template: Key Sections & Guidance
🔒 CHAPTER III: AI PRD Template to Download (Google Docs)
🔒 CHAPTER IV: AI PRD Case Study for Shopify Auto Write
🔒 CHAPTER V: Conclusions
Before we proceed, I’d like to recommend the AI Product Management Certification. It’s a six-week cohort taught by instructors like Miqdad Jaffer.
I (Paweł) participated in the cohort in Spring 2024. The next session starts on April 26, 2025.
I was able to secure a $250 $500 discount for our community if you use this link to sign up:
This post was not sponsored.
CHAPTER I: How to Work With the AI PRD
Before we dive into the AI PRD template, let’s briefly discuss how to use it.
1. AI PRD as a Powerful Alignment Tool
In my experience (Miqdad), an AI PRD is the core of the AI product development lifecycle. In many organizations I work with, it is the primary alignment tool used during Product Review Meetings (PRMs) to achieve cross-functional consensus around product strategy, design, implementation, and go-to-market activities.
The AI PRD sits at the intersection of strategy, planning, detailed product development, and effective go-to-market execution.
It covers three areas:
Strategic Context: Defines why your company should build this product or feature, clearly identifying strategic alignment, market growth, and competitive advantage.
Product & Technical Excellence: Clearly outlines how your product creates value, detailing key features and capabilities, technical specifications, and AI-specific considerations that are often overlooked.
Go-to-Market: Guides how you effectively communicate, launch, measure, and continuously improve your product to maximize customer and business value.
2. AI PRD as a Living Document
The AI PRD is a living document. In our AI Product Management course, we discuss the concept of "continuous PRM mode" - a mindset in which Product Managers constantly operate in a state of testing assumptions and gathering insights, as visualized by Paweł:
Every interaction, whether a Slack conversation, hallway discussion, formal meeting, or customer feedback session, becomes an opportunity to test assumptions about user value, usability, business viability, and feasibility.
These insights should continuously update your AI PRD, ensuring it remains accurate and relevant.
CHAPTER II: AI PRD Template: Key Sections & Guidance
Here, I outline each PRD component with a clear understanding of its purpose, recommended tools, examples, and common traps.
The specific sections of your AI PRD should be customized to your context, including the company's strategy, market position, product scope, and user base.
In the next chapter, we discuss how this framework was applied to Shopify Auto Write.
1. Executive Summary
Purpose: Briefly summarize the initiative and success criteria for those who won’t read the entire document.
Example:
“This product requirement document describes building an AI assistant integrated with the existing customer support chat.
It answers to an identified market opportunity (25% CAGR), is aligned with our company strategy, and leverages our unique competitive edge (proprietary data) with the pilot phase planned for the end of Q2 2025.
Success criteria:
An 80% satisfaction (CSAT) rating.
50% of user requests are solved automatically.
Generate accurate responses to the 25 most frequently asked FAQs.”
2. Market Opportunity
Purpose: Clarify if the opportunity is strategically compelling, timely, and positioned to deliver meaningful value.
Key Questions:
Is this the right market growth stage (e.g., emerging, maturing, declining)?
What is the market’s current growth rate (e.g., CAGR), and what data supports this?
What’s the potential for this opportunity in the future? Can it create enough business value?
Recommended Tools:
Industry Lifecycle Model (related to market and business growth)
Technology Adoption Lifecycle (how users adopt innovative products or technologies)
Example of Good Response:
"Market X has shown a consistent growth rate of 25% CAGR due to recent regulatory changes and customer adoption increasing exponentially. Currently, it's at an early growth stage (10-15% market penetration), providing ample room for capturing substantial value. TAM projected at $5B by 2030."
Example of Bad Response:
"Our competitors are entering this market, so we should too."
Common Traps to Avoid:
Misinterpreting temporary trends as long-term market shifts.
Failing to validate growth assumptions.
Ignoring external macroeconomic or regulatory factors.
Advice: Hard data beats gut feelings. Quantify the opportunity with market reports or customer surveys.
3. Strategic Alignment
Purpose: Articulate how your AI product or feature aligns with your vision, strategy, and objectives.
Key Questions:
For an AI product:
Does this align with our company’s vision and long-term strategy?
How does this support our company objectives?
Does this play to our strengths and competencies?
For an AI feature:
Does this align with our company’s vision and long-term strategy?
Does this align with our product vision and strategy?
How does this support our team objectives?
Recommended Tools:
Lean Canvas (unlike Business Model Canvas, it defines “unfair advantage”)
Product Strategy Canvas and Startup Canvas (add vision, tradeoffs, and capabilities)
Product Mix Strategy (for managing and optimizing product portfolio)
Example of Good Response:
"Our AI-driven risk assessment product aligns with our company’s focus on the finance sector. It directly supports our mid-term objective of increasing adoption of 20% mid-market bank adoption by automating regulatory compliance workflows."
Example of Bad Response:
"We’ll use AI because it’s innovative and trending."
Common Traps to Avoid:
Not tying the initiative to your company/product vision, strategy, and objectives.
Ignoring the company’s competencies and resources.
Advice: Not every problem is the right one for your team or organization.
4. Customer & User Needs
Purpose: Prioritize solving the highest-value user problems clearly and precisely.
Key Questions:
What are the key market segments and user personas?
What are their primary jobs-to-be-done and pain points?
Are there any constraints (e.g., geographic, language, regulatory)?
Which user problems, if solved, would generate the most frequent, severe, and widely experienced value?
Recommended Tools:
Opportunity Score by Dan Olsen (importance vs. satisfaction)
Jobs-to-be-Done Framework (more time-consuming if performed rigorously)
Example of Good Response:
"Users spend two hours daily manually reviewing data reports, resulting in frequent errors. 80% of users report this as their top pain (survey, n=200, extremely high importance, extremely low satisfaction)"
Example of Bad Response:
"Users have said our current solution is not user-friendly."
Common Traps to Avoid:
Choosing low-frequency or low-severity user pains.
Assuming user pain without direct evidence or validation.
Over-generalizing user needs without specifics.
Advice: Use interviews, surveys, or (for an existing product) analytics to understand and build empathy with the users rather than assuming their jobs and pain points.
5. Value Proposition & Messaging
Purpose: Communicate the unique value proposition clearly and compellingly.
Key Questions:
Which problems for each market segment do we address?
What are the key capabilities and features (high-level) that solve those problems?
What are the benefits and customer outcomes?
How is this different from what others offer?
How can we clearly and compellingly communicate our product’s unique value to resonate with each segment?
Recommended Tools:
Value Proposition Template (in their post, How to Design a Value Proposition,
and explain how they added focus and structure to the Value Proposition Canvas)Value Curve (to compare your value proposition with competitors’).
Example of Good Response:
"Our product reduces customer support workload by 40%, saving 10 hours weekly, clearly addressing frequent and painful user bottlenecks with AI assistants available 24/7."
Example of Bad Response:
"Our product is powerful and innovative."
Common Traps to Avoid:
Generic messaging without specific, measurable user outcomes.
Listing features and capabilities without benefits.
Listing benefits without explaining how they will be achieved.
Failing to align messaging with specific market segments.
Advice: Adjust messaging and format to each segment and channel (e.g., cost savings for budget users and short videos on Instagram).
6. Competitive Advantage
Purpose: Clarify your product’s defensibility and ability to sustain competitive advantage.
Key Questions:
What makes us think competitors can’t/won’t copy our strategy?
How defensible and durable is our advantage in the long-term?
Recommended Tools:
Competitive Advantage (MOATs)
Example of Good Response:
"The product leverages our proprietary dataset of over 10 M+ transactions and integrates seamlessly with our existing fraud detection system, giving us a 3-year competitive lead.
While other solutions rely on generic machine learning models, our domain-specific AI is trained on industry-specific workflows, making it even harder to replicate.”
Example of Bad Response:
"Our UI is more user-friendly than competitors."
Common Traps to Avoid:
Vague or unclear differentiation statements.
Relying on easily replicable advantages like superficial UI features.
Underestimating competitor agility or resources.
Choosing differentiation based solely on technology.
Neglecting reassessment of competitive advantage. Competitive edges erode.
Advice: Consider hard-to-replicate edges like data, partnerships, or integrations.
7. Product Scope and Use Cases
Purpose: Define the key capabilities and features with tasks our product (or a feature set) must perform exceptionally well.
Key Questions:
What are the key capabilities and features?
Can we link designs or prototypes for better alignment?
What are the desired customer outcomes?
What are the high-risk assumptions? How can we test them with minimal effort?
Recommended Tools:
Low-fidelity and high-fidelity user prototypes.
Market engagement experiments, usability testing, spikes, and experiments on production. A comprehensive overview: Testing Product Ideas.
Example of Good Response:
An AI assistant integrated with the existing customer support chat. It must generate accurate responses to the 25 most FAQs, achieving an 80% satisfaction (CSAT) rating.
Design: [Figma prototype]"
Example of Bad Response:
"The AI assistant will answer questions users might have."
Common Traps to Avoid:
Trying to include too much information (e.g., including all user stories or describing every corner case) - your PRD is a tool for building alignment, not explaining every detail.
Not specifying measurable customer outcomes.
Trying to address all user needs simultaneously rather than starting simple and iterating based on customer feedback.
Neglecting to test high-risk assumptions before implementation.
Advice: A picture is worth a thousand words. Show, don’t just tell.
8. Non-Functional Requirements
8.1 General Requirements
Purpose: Define the essential system attributes (e.g., performance, scalability, security) that ensure the product operates reliably under expected conditions.
Key Questions:
What technical, performance, scalability, security, and reliability attributes must our product achieve to deliver and sustain its intended value?
What are the peak load and user volume expectations?
Recommended Tools:
MoSCoW Method (Must, Should, Could, Won’t) for prioritization.
Performance benchmarking tools (e.g., JMeter, LoadRunner).
Security compliance checklists (e.g., OWASP Top 10, GDPR).
Example of Good Response:
"The AI assistant: delivers responses in <500ms for 95% of queries, scales to 50K users with 99.9% uptime, encrypts all user data per GDPR standards."
Example of Bad Response:
"The system should be quick and secure enough for users."
Common Traps to Avoid:
Overlooking scalability needs until late in development, leading to costly rework.
Writing vague requirements (e.g., "fast" or "reliable") without measurable targets.
Failing to balance trade-offs between attributes (e.g., performance vs. cost).
Advice: Define specific metrics early. Remember that “fast” isn’t a target.
8.2 AI-Specific Requirements (LLMs)
Purpose: Ensure the AI consistently delivers accurate, reliable, ethical, and user-aligned outputs.
Key Questions:
What are the key AI architectural choices?
What accuracy, reliability, and ethical standards must our AI meet (e.g., best practices, brand voice, and legal compliance guiderails)?
How will we measure these qualities?
How will we maintain them over time?
Recommended Tools:
AI evaluation metrics.
Bias and fairness audits.
Prompt engineering frameworks (e.g., chain-of-thought prompting).
Retrieval-augmented generation (RAG) for grounding outputs in verified data.
Example of Good Response:
"The AI must achieve ≥90% accuracy on a labeled test set of 10,000 queries, limit hallucination rates to <2% via RAG integration, and flag inappropriate outputs with 98% precision, validated monthly through human review."
Example of Bad Response:
"We’ll use GPT and expect good accuracy because it’s popular."
Common Traps to Avoid:
Using pre-trained models without validation or fine-tuning.
Neglecting ongoing monitoring for drift in accuracy or reliability.
Ignoring ethical risks (e.g., bias, inappropriate outputs) until user complaints arise.
Advice: Monitor drift. Without proper oversight, retraining, and adaptation, an AI system can become outdated and unreliable (e.g., due to changes in the relationship between input data and the expected outputs). Regular monitoring, validation, and feedback loops are essential.
9. Go-to-Market Approach
Purpose: Define how you rapidly demonstrate measurable value and grow user adoption.
Key Question:
What are the build and release phases (e.g., an MVP addressed to early adopters)?
What are the first market segment(s) we want to focus on?
How can we win those customers, rapidly demonstrate the value, and get evidence to accelerate further growth?
Recommended Tools:
Crossing the Chasm Framework (e.g., to target early adopters)
AARRR (Pirate) Metrics (e.g., to scale while tracking Acquisition, Activation, etc.)
Example of Good Response:
"We’ll launch an MVP targeting small e-commerce businesses in North America, offering 20% cost savings on customer support within 30 days. Success will be measured by a 50% activation rate and 10 case studies in 90 days, driving expansion to mid-sized companies.
We believe that the next step will be…"
Example of Bad Response:
"We’ll launch on Product Hunt and see what happens."
Common Traps to Avoid:
Targeting an overly broad or poorly defined initial segment(s), diluting focus.
Skipping measurable success metrics.
Confusing Early Customer Profiler (ECP) with Ideal Customer Profile (ICP).
Advice: Start small, measure the results, and learn fast rather than trying to make it perfect. At the same time, test responsibly (e.g., company reputation). Use early wins to fuel growth.
🔒 CHAPTER III: AI PRD Template to Download (Google Docs)
A Google Docs version of the PRD template below 👇
In CHAPTER IV, we discuss AI PRD Case Study for Shopify Auto Write.
Keep reading with a 7-day free trial
Subscribe to The Product Compass to keep reading this post and get 7 days of free access to the full post archives.