APattern A
The Meeting You Forgot, Surfaced by Your AI
As a skilled product manager, you need to access and use a large amount of information.
But what about the user interview four months ago that you’ve forgotten - even though it’s directly relevant to a feature you’re working on? Here, we’ll focus on the Product Requirements Document and how to use AI to pull in context you may have forgotten without overstuffing the prompt.
Good PRDs focus on a discrete piece of development that solves a specific need. Say, a user is having difficulty viewing a key piece of data in an actionable way. In this case, the data is an appointment change request. In the current state, the user has difficulty being aware of the request. And if they are aware, they don’t have a full understanding of what action they need to take.
You’ve done extensive research. You’ve built and iterated on prototypes. You have a solid solution, and it’s time to put together the PRD to document the solution, get stakeholder sign-off and pass it off to engineering.
You point your AI to all of your relevant assets and ask it (saying please, of course) to generate the PRD.
You know you’ll get a good PRD based on the assets you have related to this need.
But you know you’ll get a stronger PRD if the AI has even more context. So you feed it:
- All prior Jira tickets for this product
- Confluence documentation across the product organization
- Company strategy presentations
- Meeting notes / transcripts from the past several months
- User documentation
- User journeys
- User workflows
Not all of this is relevant, but you’ve been incorporating this information into your AI all along, so the AI has the information indexed across your internal documents. Every meeting you attend, every prototype you build, every observation or interview, you’ve been feeding the output into your AI - even if it wasn’t relevant at the time.
Your AI has access to much of what you know, and close to a photographic memory of it.
Back to the PRD. It’s created the document, and as you’re reviewing it you notice a few things.
First, it’s a solid PRD. But you expected that. Your AI had all the relevant information and access to a template of what the PRD should look like.
The second thing you notice, though, is a surprise.
The AI remembered a 30-minute meeting from four months ago where an end user mentioned that they don’t typically work out of the appointment list view. Rather, they almost always work from a view that shows appointments for a particular person.
You validate what the AI found against user monitoring logs. (The AI probably already did this, but it’s still good to validate.) The logs confirm that, yes, about 80% of the time users look at appointments on an individual level rather than across all appointments. You’re embarrassed, but grateful to the AI.
Users are still going to want to act on the change request within an individual appointment. They just need to have access to broader context at the time.
It’s good the AI caught that gap. Even better, the AI expanded the PRD so users can view and act on change requests in any relevant view. Nice job, AI. And nice job, you, for having the foresight to continuously load context into your AI.
A standalone AI works well, but an AI that has the context of your work and your company is much stronger. To help your AI stay strong and relevant, schedule jobs to regularly pull in meeting notes / transcripts, new / updated Jira tickets, changes to Confluence spaces and code updates. As part of the jobs, make sure information is indexed and accessible - you don’t want the AI to load all your information for every query. But you do want it to know where to look when it needs information.
The AI is only as strong as the data it can actually find. And by giving it all the data you have access to in an organized format, the AI becomes a surprising partner. As a first step, load indexed and tagged meeting transcripts into your AI to provide context for your prompts. Transcripts have lots of great data, but are often hard to summarize and parse manually - making them a great place to incorporate AI.
Comments (0)
No comments yet. Be the first to share a thought.
Leave a comment
Comments are reviewed before they appear. I store your name and comment, not your email.