Picture by Editor
# Introduction
Making a product necessities doc (PRD) is a standard course of in product administration and a commonplace job in sectors like software program growth and the tech business as an entire. A number of the sometimes discovered difficulties and arduous necessities in making a PRD embody guaranteeing readability, stopping scope creep, and preserving stakeholder alignment.
Fortunately, AI instruments have risen to assist navigate these challenges extra successfully, with out utterly delegating the strategic decision-making underlying the PRD creation course of — in different phrases, with the human nonetheless within the loop. One instance is Google’s NotebookLM, which synthesizes grounded uncooked information or supplies to reply questions, thereby turbocharging the workflow for creating grounded, helpful PRDs.
This text will navigate you, based mostly on a beginner-friendly use case, via the method of utilizing NotebookLM’s options to show uncooked, generally chaotic data right into a grounded PRD in a matter of minutes. Spoiler: it will not be nearly chatting with an AI assistant.
# From Messy Notes to a Structured PRD Draft
Let’s contemplate the next state of affairs. You’re the newly employed product supervisor for a startup that desires to develop a brand new cell app referred to as FloraFriend. The purpose of the app is to assist individuals cease by accident killing their houseplants.
The crew, together with you, has collected a set of three “messy” paperwork that comprise descriptions for what the potential app must be like:
interview_transcript_matt.txt: a 30-minute interview with a consumer referred to as Matt, who’s the proprietor of over 50 crops. In these interview notes, Matt says current apps are “overly sophisticated” and make it troublesome to retain in thoughts elements like “which fertilizer to make use of.”competitor_research_notes.txt: a tough listing of bullet factors made after analyzing competitor apps like “PictureThis” and “Planta”, highlighting their paywalls and interface drawbacks.brainstorming_whiteboard.jpg: random however considerably “cool” concepts which were talked about by the crew throughout lunch breaks and different informal conversations, e.g. “spotify playlists for crops”, “watering reminders”, and so forth.
Think about full paperwork containing the entire content material described above. Manually turning these right into a clear PRD that properly brings all of it collectively could sound like a ache, proper? Enter NotebookLM!
Log in to NotebookLM together with your Google Account and click on “Create New Pocket book“. Give your new pocket book a reputation, one thing like “FloraFriend PRD.”
As soon as the brand new pocket book has been created, you will be welcomed to the primary NotebookLM interface, which seems like this:
NotebookLM Interface
A phrase of warning: this newly created pocket book isn’t clever per se. It isn’t a daily giant language mannequin (LLM); it doesn’t know plant care or every other particular matters. However we’re about to show it an “specific” Grasp’s diploma about it with our messy — but enlightening for the instrument — notes.
Suppose you may have the three above talked about information with some content material associated to the plant care app, or every other uncooked data information of your individual. You possibly can add them to the NotebookLM canvas by utilizing the add button in the primary, central part.
As soon as uploaded, you may consider your pocket book as one thing just like a tiny, toy-sized retrieval-augmented technology (RAG) system that may begin considering and behaving AI-like based mostly on the knowledge it has entry to. In truth, with out asking it, by clicking on both one of many uploaded information on the left-hand facet, NotebookLM generates a concise, well-organized abstract of the contents in that file: that is referred to as a file’s Supply information.
Now comes the important thing half. We may merely ask within the chat field on the backside one thing like “Write a PRD”, and that is it. However we need to do that correctly and supply clear, particular directions, and that entails some immediate engineering, specifically to power the newly born AI to prioritize what we wish our PRD to mirror: prioritizing the consumer issues over the random concepts generated by the crew (with out completely neglecting them). Here’s a well-crafted immediate that works:
I’m the product supervisor for FloraFriend. Primarily based solely on these sources, draft a PRD.
Essential constraints:
1. Prioritize options that clear up the ache factors talked about in interview_transcript_matt.txt.
2. Exclude any ‘brainstorming’ concepts that do not immediately handle a consumer downside.
3. Construction the output with these headers: Drawback Assertion, Core Options, Non-Purposeful Necessities (UI/UX), and Success Metrics.
Strive adapting this immediate to your individual enterprise downside or use case. As soon as despatched, likelihood is you’ll get a pleasant and clear PRD with key sections like Drawback Assertion, Core Options, Non-Purposeful (UI/UX) Necessities, Success Metrics, and so forth.
Apparently, the PRD comprises one thing that appears like numerical citations you may hover on. When you accomplish that, you will notice the supply (one of many supply information) pop up:
Earlier than accepting this primary PRD as it’s, do not forget that a primary draft isn’t good. Maintain partaking in dialog to step by step refine it, e.g. in the event you discover there’s a lacking monetizing part, ask: “Primarily based on the competitor_research_notes.txt, what monetization fashions are our rivals utilizing, and what ought to we keep away from?“. After that, manually verify the outputs, make sure that they’re according to the remainder of the primary PRD draft, and incorporate the primary monetization insights into it, both manually or by asking NotebookLM’s AI to take action — in the event you go for the latter, all the time verify what you get earlier than blindly approving it. Keep in mind: AI could make errors!
The icing on the cake is the Audio Overview part on the right-hand panel (Studio). By simply clicking on it, you’ll generate an audio overview of the knowledge contained within the supply information. This is a wonderful technique to take in data when studying is perhaps much less interesting, e.g. while you’re in your day by day commute.
# Subsequent Steps
This text introduces NotebookLM’s capabilities to generate grounded PRD specs from uncooked, messy paperwork in a matter of minutes, taking very simple steps. From right here, a worthwhile subsequent step could possibly be resorting to Google’s Antigravity to show your PRD specification right into a useful software program prototype.
Iván Palomares Carrascosa is a pacesetter, author, speaker, and adviser in AI, machine studying, deep studying & LLMs. He trains and guides others in harnessing AI in the true world.
