How to Prioritize as an AI Product Manager
How to Prioritize as an AI Product Manager
Why You Should Prioritize for Assessment and Not for Development
#beyondAI
Imagine launching an AI feature only to realize it solves a problem no one really cared about. That’s what happens when prioritization goes wrong.
Whether you're just beginning your journey with a new product vision or deep into its development, you will inevitably face the challenge of prioritization. Which product or feature idea should take center stage?
The faster you learn to solve the prioritization dilemma within your context as an AI Product Manager, the more likely you and your team are to deliver real value. AI products are highly complex and resource-intensive, and getting the sequence right can mean the difference between breakthrough success and wasted effort.
And let’s be honest—the primary reason the role of a Product Manager exists isn’t simply to build things. It’s to create value. A Product Manager wouldn’t even be necessary if our goal were purely to build. Our mission goes far beyond that (which is also the reason why this publication is named beyondAI)—it’s about making sure that what we build has a meaningful, lasting impact.
It’s about solving real problems, along with all the other major and minor challenges that come along the way to solving that core issue.
If you’ve followed my earlier discussions on the core responsibilities of a Product Manager, you’ll know where I stand: problem-solving is at the heart of what we do. And, yes, prioritization is one of those critical problems every Product Manager must master.
This topic is rooted in my experiences with prioritizing ideas that eventually evolve into AI products or features. While I primarily develop AI products for large organizations as an internal AI Product Manager, the principles I discuss apply universally—to those building for companies (B2B) and those building for end-consumers (B2C) alike.
You will learn about
My Process of AI Idea Assessment
The Quick Assessment
The Most Important Step: Capture the Data
How Occam’s Razor will help
Key Parameters for a Quick Assessment
Final Thoughts
Happy reading 🛋️
If you’ve found my posts valuable, consider supporting my work. While I’m not accepting payments right now, you can help by sharing, liking, and commenting here or on my LinkedIn posts. This helps me reach more people on this journey, and your feedback is invaluable for improving the content. Thank you for being part of this community ❤️.I always maintain an idea backlog—my treasure trove of potential innovations. Every now and then, when I see that we're approaching the delivery of our latest feature, I dive back into that list, searching for the next most valuable idea to pursue. But this list isn’t just a random collection of scribbles or a pile of fragmented thoughts, though I must admit, for a long time it was just that—a messy list with high-level descriptions.
It wasn’t until I realized how essential it is to prepare this backlog well in advance of our development capacity freeing up. With this mindset shift, I began treating my backlog as a strategic asset, meticulously curating it so that when it’s time to start a new feature, I can quickly identify the most promising idea and begin deep diving into that problem. This simple yet crucial change transformed my backlog into what it is today—my treasure.
What Does This Process Actually Look Like?
Let me walk you through my process for generating and refining ideas for new AI products or features.
At the very beginning of any product ambition—whether AI or otherwise—there’s always a rough idea of a solution. And when I say "solution," I’m inherently referring to a problem that people have. After all, no solution exists in a vacuum. The challenge is often figuring out which problems are truly worth solving and, in our context, just as importantly, which problems AI can uniquely address.
There are a few ways I come across these problems. Networking deeply within the organization is one of them. I frequently connect with stakeholders across various roles—whether it’s operations, marketing, customer service, or engineering. When someone mentions a pain point, especially one that might align with AI’s capabilities, I capture it immediately. But I’m not simply jotting down. I’m recording all relevant information right away. Or at least as much as possible in that moment. This ensures that my prioritization framework remains structured and effective. Only this way can I build up a well-prepared idea list that seamlessly transitions into the next phase.
But my prioritization is a bit different.
It doesn’t prioritize the next thing to develop.
In my ten years working in AI product management, I’ve learned that prioritizing directly for development is highly inefficient. Each development phase needs to be preceded by a detailed assessment. And this process is resource-intensive. It requires experts from various domains to scrutinize the problem from multiple perspectives. You need input from engineering, data science, UX, legal, and compliance teams to fully understand the feasibility of a potential AI solution. Additionally, involving domain experts (or end customers) to ensure the solution's viability and desirability. This detailed assessment, sometimes referred to as idea due diligence, can be quite expensive. It's also known as the viability, desirability, and feasibility assessments.
Detailed Assessment for AI Solutions
The complexity of building AI solutions is defined by six dimensions: three elements focused on the technical aspects of an AI product—Data, IT, and AI itself—and three elements that ensure these technologies are effectively implemented within a business context—Governance, Business, and People. I call this the Double Trio of AI Product Development.
This duality emphasizes the balance between technology and the human-centric, strategic approaches needed for successful AI products. It also illustrates why the stakes are much higher compared to traditional software development.
During each detailed assessment, we gather sufficient detail to make educated guesses about our chances of successfully building the product. Yes, even after a detailed assessment, it remains an educated guess. There is no guarantee for success. You can only increase the likelihood.
With AI, poor assessment can lead to biased models, scalability issues, or products that fail to deliver on their promises because they are not secure, trustworthy, or compliant with governance standards.
The outcome of a Detailed Assessment is, simply put, a Go or No-Go for starting development.
If my backlog only contains rough ideas with no details, I can’t effectively prioritize which ones to put through this costly, resource-intensive process. This increases the risk of pushing ideas into Detailed Assessments that are likely to fail.
📌 Note: This might seem like a waterfall approach, but it’s not about understanding all the requirements to build a full-fledged AI product. Instead, it’s about deciding whether you should build at all.That’s why I’ve developed the quick assessment. It serves as an initial filter. It allows me to prioritize ideas for detailed assessment rather than randomly selecting what to assess in-depth.
The Quick Assessment
The essence of the quick assessment is to capture necessary information on the fly and by yourself. And if needed, get help from experts whenever you have the opportunity to ask. Over time, with more experience and technical understanding, you’ll find that you can quickly assess mostly independently without much help.
The Most Important Step: Capture the Data
Make it a habit to capture information. Whenever you see an opportunity, seize it. Ask the right questions to fill in the blanks in your idea list. This is the most critical step, but it doesn’t require much explanation. The idea of assessing on the fly and on your own boils down to this essential habit. No matter where you are or what you’re doing, always remember that you have a list of ideas to nurture. This list can become your treasure, but only if you take it seriously and prioritize it as a core part of your role.
Are you in a meeting with your marketing colleagues? Have your idea list open and ask about potential business impacts or data they have that could help refine your ideas.
Sitting together in a review meeting with another AI product team? Listen to the types of data they are using, and think about whether it could also help strengthen or validate one of your own ideas.
Attending a team workshop? Watch for discussions about pain points or upcoming trends that could inspire new product ideas, and capture those insights right away.
The key is to always keep your mind open and be prepared to capture relevant information whenever it presents itself. Opportunities to fill in your idea list come up more often than you think, but only if you’re ready to ask the right questions and gather useful data on the fly. This habit is what transforms a simple idea backlog into a well-informed, ready-to-prioritize asset.
Here are some practical ways to capture relevant data:
Quick Conversations: Arrange short discussions with sales, marketing, operations, or engineering leads to ask these core questions about potential impacts, data availability, or technical feasibility.
Internal Documentation: Check past reports, team dashboards, or existing KPIs to find rough estimates on business metrics like cost savings, cost avoidance, or revenue growth.
Existing Tools: Utilize internal analytics tools, project management software, or data dashboards that already capture business metrics. These tools often provide enough high-level data to support rough estimates for your quick assessment.
Experience-Based Estimates: When in doubt, ask team members for a gut-level estimate based on their experience. It doesn’t need to be precise at this stage—rough numbers are good enough to guide your decision-making process.
Maintaining this habit becomes much easier if you understand that it is all about estimation. And here, the principles of Occam’s Razor are invaluable.
What is Occam’s Razor?
Occam’s Razor is a principle that says the simplest solution is usually the best one. It means you should avoid adding unnecessary complexity to a problem when a simple explanation or solution is available. Instead of looking for more complicated answers, start with the simplest approach that solves the issue.
Think of it like this: If you're trying to figure out why your phone isn't charging, you wouldn't first assume the charger is broken, the outlet is faulty, and your phone’s battery needs replacing all at once. Instead, you'd start with the simplest idea—maybe the cable is just unplugged. Once you check that, you can move on to more complex ideas if needed. The same holds for the Quick Assessment.
How Does Occam’s Razor Help Us Capture Relevant Data for an Idea?
When assessing a new idea for a product or feature, there’s often a temptation to gather every possible detail. However, this can slow down the decision-making process and waste resources. What’s needed is the most important information—the core facts that allow you to make a decision quickly and move forward.
This is where Occam’s Razor becomes invaluable. It helps you focus on capturing data in the simplest, fastest way possible. Instead of waiting for perfect data or getting bogged down in complexities, you zero in on the core questions that matter and gather rough estimates. These are good enough for quick assessments and let you determine whether an idea is worth further investment.
⚠️ However, it’s important to remember that a Quick Assessment doesn’t replace a Detailed one. It simply increases the likelihood of a positive outcome after the Detailed Assessment.Example 1: Estimating Cost Savings
Instead of spending hours gathering historical cost data and calculating detailed financial models, you can simply ask the problem owner or relevant team:
Question to Ask: “Roughly, how much money do you think this solution could save?”
For example, if you're considering an AI-driven solution to automate customer support:
Opportunity: Speak with the head of customer service or review current staffing costs.
Estimated Response: “I’d estimate we could save around $50,000 annually by automating 30% of the inquiries.”
This rough estimate gives you a solid enough figure to decide whether the idea merits further investigation, without needing precise calculations upfront.
Example 2: Determining Data Availability
Occam’s Razor also guides you to quickly determine whether the data needed to support an AI solution is available. Instead of running detailed data audits, ask simple, direct questions:
Question to Ask: “Do we already have the necessary data to implement this solution?”
For an AI-driven logistics optimization feature:
Opportunity: Check with your data team or look into current databases. Often, a quick conversation or browsing internal dashboards can give you the answer.
Estimated Response: “Yes, we have delivery route data from the last 12 months. It’s stored in our system and ready to use.”
This straightforward check gives you enough confidence to assess whether the AI solution is technically feasible without needing an extensive review of the data's quality at this stage.
Example 3: Assessing Revenue Growth Potential
If the idea’s goal is to increase revenue, instead of diving into complex sales projections, you can ask sales or marketing for a ballpark estimate based on their knowledge:
Question to Ask: “How much additional revenue do you think this feature could generate?”
For a personalized product recommendation AI tool:
Opportunity: Reach out to the marketing team or review past performance data from similar personalization tools.
Estimated Response: “From our experience with similar tools, this could increase online sales by around 5%, which translates to about $100,000 annually.”
This gives you a rough idea of the potential revenue impact without having to build a full financial model.
Example 4: Evaluating Technical Feasibility
Rather than conducting a full technical feasibility study, Occam’s Razor suggests asking your technical team the simplest version of the question:
Question to Ask: “Can we build this with our current infrastructure?”
For example, if you're considering a machine learning model to predict customer churn:
Opportunity: Schedule a brief conversation with your engineering lead or check documentation on current tech stack capabilities.
Estimated Response: “Yes, we already have the necessary computing resources and frameworks in place. We could start working on a model in the next sprint.”
This quick check gives you enough confidence to move forward without a detailed infrastructure analysis at this stage.
Example 5: Determining Time to Market
Instead of building out a detailed project plan during a quick assessment, Occam’s Razor suggests asking:
Question to Ask: “How long do you think it would take to develop a prototype?”
For a new AI-powered recommendation engine:
Opportunity: Ask your product or engineering team for a rough timeline based on their experience.
Estimated Response: “I’d say we could have a working prototype in about 3 months.”
This rough estimate helps you decide whether the timeline aligns with business needs without needing an in-depth project plan.
How Does Occam’s Razor Help Us Select Which Ideas to Move to a Detailed Assessment?
Once you’ve captured the key information about your ideas using Occam’s Razor, you might have several ideas with potential. But you can’t assess everything at once, so you need to decide which idea should go to the next stage for a detailed assessment.
This is where Occam’s Razor helps once again: by guiding you to pick the simplest and most impactful ideas first, without getting caught up in overly complex options. Here’s how you can apply this principle to a list of ideas:
Focus on the Core Business Impact - Start by looking at the key information you gathered, particularly the potential business impact (cost savings, cost avoidance, and revenue growth). Instead of getting lost in every detail, ask yourself:
Which idea has the clearest, most direct impact on the business?
Which idea offers the highest potential benefit with the least amount of assumptions?
For example, if one idea could save $50,000 in costs based on straightforward assumptions, while another idea might generate $100,000 in revenue but requires several complex conditions to be met, Occam’s Razor suggests focusing on the $50,000 cost-saving idea first. It’s simpler, easier to validate, and doesn’t rely on multiple uncertain factors.
Eliminate Complex Dependencies - Occam’s Razor tells us to cut away unnecessary complexity. If an idea relies on many uncertain variables—such as needing new data, additional resources, or external factors—you might want to pause that idea and focus on the one that has fewer dependencies and can be executed more easily.
For instance:
Idea 1: Requires a new data set that might take weeks to collect.
Idea 2: Can work with existing data and doesn’t need new resources.
Even if Idea 1 has a higher potential benefit, Occam’s Razor would push you toward Idea 2 because it’s simpler and you can start assessing it right away.
Select the Idea That Can Be Validated Quickly - You don’t want to spend weeks in detailed assessments without seeing results. Occam’s Razor pushes you to select the idea that can be validated quickly. Ask:
Which idea can be prototyped or tested the fastest?
Which idea has the least uncertainty around it?
For example:
Idea A: Requires a month of data gathering before you can even start testing.
Idea B: Can be tested with existing tools and data, and you can have initial results within a week.
Occam’s Razor suggests choosing Idea B because it allows you to validate your quick assessment sooner, without needing extensive preparation.
Key Parameters to Capture in a Quick Assessment
The quick assessment process - to be simple and fast - needs to focus on key factors that help filter out less viable ideas early. Along with tactical considerations, we must account for strategic dimensions that ensure long-term value and alignment with broader business objectives.
📌 However, whatever you can’t capture qualitatively, always capture a rough estimate of the potential benefit. At the end of the day, the potential benefit will often be the main driver for your decision, especially in cases where other data is limited or ambiguous. While more detailed insights will certainly emerge in the detailed assessment, a rough estimation of the benefit in the quick assessment phase is essential.1. Problem Definition - Strip down the problem to its simplest form. Ask, "What’s the core issue we're trying to solve?" Avoid focusing on tangential issues or overcomplicating the problem.
✅ Tip: Focus only on the core pain point that will drive decision-making, ignoring secondary issues that can be assessed later.2. Potential Business Impact - Don’t overthink it. Ask the problem owner simple, direct questions about potential benefits —rough estimates are sufficient here, simplified to three KPIs: cost savings, cost avoidance, revenue growth
Cost Savings: "Roughly how much do you think we could save if we solve this?" Capture a quick estimate like, “Estimated $20,000 in labor cost savings annually.”
Cost Avoidance: "What future costs could we avoid if this problem is solved?" Get responses like, “Avoid approximately $50,000 in repair costs per year.”
Revenue Growth: "How much additional revenue could this solution generate?" Capture estimates like, “Potential $100,000 in new sales from improving product recommendations.”
✅ Tip: Don’t insist on precise numbers—ballpark figures provide enough insight to gauge potential impact without wasting time on excessive data collection. If you need more clarity on how an AI solution can bring value, give this article a read: https://www.jaserbk.com/p/how-ai-products-can-have-double-impact3. AI Applicability - Ask the simplest, most direct question—"Does this problem truly need AI?" If a non-AI solution could solve the problem, don’t assume AI is necessary. "Yes, AI is needed to optimize delivery routes" or "No, automation can solve this without AI."
✅ Tip: If you need more clarity on when AI is appropriate and when not, give this article a read: https://www.jaserbk.com/p/when-to-automate-and-when-to-augment4. Data Availability - Instead of conducting a detailed data audit, ask, "Do we already have data for this?" If yes, note it. If no, can it be easily gathered? “Delivery data from the past year is available.” Don’t worry about perfection; the point is to know whether the data exists.
✅ Tip: Don’t get bogged down in the quality or completeness of the data at this stage—just check whether data is available for a basic solution.5. Technical Feasibility - Instead of exploring all technical angles, ask the most direct question: "Can this be built with our current infrastructure and resources?" - “Yes, the current infrastructure supports building an AI model.”
✅ Tip: Don’t overanalyze the technical details at this stage—just assess whether it’s feasible with what you currently have.6. Time to Market - Simplify your estimate by asking, "How long will it take to build a prototype or solution?" You don’t need exact timelines—rough estimates will do. “Prototype possible in 2 months, full solution in 6 months.”
7. Cost of Development - Don’t overcomplicate by calculating exact budgets at this stage. Ask, "What’s the ballpark development cost?" - "Estimated development cost: $75,000–$100,000."
✅ Tip: Use rough estimates based on past projects or high-level input from engineers or finance. Avoid detailed cost analysis in a quick assessment. If you need more clarity on how you can figure out if an AI investment is worth is give this article a read: https://www.jaserbk.com/p/how-to-assess-financial-viabilityFinal Thoughts
Once an idea has passed through this quick yet effective filter, it’s no longer just an idea sitting in my backlog—it’s a qualified opportunity. That’s the real power of this approach. It’s the difference between a messy, unordered list of random thoughts and a strategically crafted backlog, ready for action. I’m not scrambling when development capacity opens up. I’ve already got a well-refined list of opportunities—each one vetted, understood, and ready to roll. So, when the time comes, I can act swiftly and decisively.
Of course, during the detailed assessment, it may turn out that some of the assumptions you made in the quick assessment were entirely wrong. But here’s the key: this will happen far less often than if you hadn’t done a quick assessment at all. By filtering ideas early, you significantly reduce the chances of pursuing low-value or poorly thought-out concepts, saving time and resources in the long run.
This approach isn’t magical.
It’s nothing new. It’s already being applied in so many other fields.
The only secret? You have to make it a habit.
It took me quite a while to build that habit—learning how to capture relevant information and ideas on the fly, by myself, no matter where I was. Once I nailed that, it became second nature.
Everything else just flowed.
Whether it was gathering key data for a quick assessment or deciding how to prioritize what came next. All fluid. All evolved. But the habit? That stayed solid.
And that’s the only constant you really need to focus on.
Care about building the right habits, and everything else is just straightforward from there.
JBK 🕊️
P.S. If you’ve found my posts valuable, consider supporting my work. While I’m not accepting payments right now, you can help by sharing, liking, and commenting here or on my LinkedIn posts. This helps me reach more people on this journey, and your feedback is invaluable for improving the content. Thank you for being part of this community ❤️.