Writing product requirements with AI: Part 1 - The Fundamentals
Are you still writing product requirements the old-fashioned way? Discover how AI can automate your task in this step-by-step guide.
“I hate writing requirements.”
I hear that from my PMs all the time.
And for good reason. Writing good-quality requirements takes a lot of time and effort to get it just right. Plus it’s boring. I mean who likes documenting?
So when Phil (a PM I coach), asked me to help him use AI to write requirements. I jumped at the opportunity.
In today’s newsletter, we are going to deep dive into:
How most PMs use AI to write requirements and why it does not work
Five steps Phil uses to write requirements
How Phil works with AI (he is always in the driver’s seat)
Benefits of using AI
Exact requirements writing prompts (part 2)
This article will cover five essential parts of requirements: features, user stories, acceptance criteria, negative acceptance criteria, and non-functional features.
So let’s dive in.
How most PMs use AI to write requirements and why it never works.
Here is how most PMs try to write requirements with AI:
We want to improve the digital application experience for our store loyalty program sign up process. Can you generate requirements?
The problem with this type of prompt is that it does not provide the LLM enough context - role, input, instructions, etc.
As a result, the LLM has to guess. It has to fill in the blanks or make things up, to answer the question.
And you end up with low-quality, irrelevant responses.
Five steps Phil uses to write requirements
So when Phil approached me, we wanted to build an AI process that would
Provide high-quality responses in the right format every time.
Deliver the relevant results in the first shot. No time wasted sparring.
It was repeatable. Build it once and run it every time for different scenarios. And,
Make sure that Phil was in the loop - directly managing the AI every step of the way.
Here are the steps we came up with for Phil to write clear, detailed, and holistic requirements.
Leverages AI to brainstorm innovative feature ideas.
Select a few relevant features. And for each selected feature, he uses AI to write user stories.
Similarly, for every selected user story. He leverages AI to generate acceptance criteria.
Next, he generates negative acceptance criteria and edge cases.
Lastly, he feeds in all the above details and generates non-functional features.
AI works best when Phil is in the driver’s seat.
In the process above, within every step:
Phil tells AI what to do (i.e. prompts).
AI generates lots of options.
Phil reviews, edits, and selects the best option.
Phil moves on to the next step.
It’s a lot like managing an intern. You provide the AI with just enough information and the AI does the grunt work.
What used to take Phil weeks now only takes a few hours.
The benefits speak for themselves. Here is the break down when Phil wrote this first set of requirements with AI,
Time Saved: 35+ “hands on keyboard” hours saved (40 hrs → 5 hrs).
Velocity Improved: 4 weeks to 5 hours.
Free Up Time: Phil can now spend more time on strategic planning, user engagement, and innovation.
Reduced Rework: Better requirements mean more holistic, fewer gaps, fewer misunderstandings, and ultimately reduced rework.
Leveling Up: Phil does not have to be a technical wiz to write requirements that will make the engineers happy.
Next week, in part 2 of this article, I will detail the specific steps and prompts that Phil employed to write the requirements.
Stay tuned.
Happy building!!