AI Product Management Frameworks
Jacob Duval
•
May 16, 2025
Do Frameworks Matter?
We often get asked about Frameworks when it comes to AI development. If you're happy with your existing Product Development framework, don't change it!
However, sometimes jumping into a new product line (like needing to add AI to your product), gives you the chance to shake things up. Here are some of our favourites.
Shape Up
Here's what makes Shape Up different:
First, it uses six-week cycles instead of short sprints. This gives teams enough time to build something substantial. For AI products, this means you can actually complete a feature that delivers real value instead of just setting up the infrastructure.
Second, work gets "shaped" before it's assigned to a team. Shaping means outlining the solution in rough terms - clear enough to guide the team but abstract enough to leave room for creativity. With AI features, this means defining what the feature will do for users without specifying exactly how the AI will work under the hood.
Third, there's the "betting table" where leaders decide what shaped work to bet on for the next cycle. Not everything makes the cut. This forces tough decisions about what AI features are truly worth building now versus later.
A product manager at a mid-sized software company recently told me how Shape Up transformed their AI assistant feature. "We kept getting stuck in the details of how the AI should work. When we switched to Shape Up, we focused first on shaping what the assistant would help users accomplish. The six-week timeframe forced us to narrow our scope to something we could actually deliver."
For AI products specifically, Shape Up helps in three key ways:
It helps you avoid the AI perfectionism trap. Instead of endlessly tweaking prompts to get that extra 2% accuracy, you focus on shipping features that are good enough to provide value now.
It gives teams autonomy to solve problems. Since AI development often involves exploration and experimentation, Shape Up's approach of defining outcomes rather than specific implementations gives technical teams the freedom they need.
It creates natural stopping points. The six-week cycle with a clear deadline helps prevent AI features from expanding indefinitely as teams chase the latest capabilities.
To get started with Shape Up for your AI product decisions, begin by identifying a meaningful feature you can shape and build in six weeks. Focus on the user benefit rather than the AI technology, and be ruthless about setting boundaries on what's in and out of scope.
Learn more about Shape Up directly from Basecamp
Remember - the goal isn't to build the most advanced AI. It's to ship products that solve real problems for your users.
Jobs to be Done
When building AI products, it's easy to get caught up in what the technology can do rather than what your users need it to do. This is where the Jobs to be Done (JTBD) framework comes in - it shifts your focus from the AI itself to the progress your customers are trying to make.
The core idea is simple: people don't buy products; they hire them to do specific jobs in their lives. For AI products, this perspective is game-changing.
Take ChatGPT, for example. Users aren't hiring it because they want a large language model. They're hiring it to write emails faster, brainstorm ideas, or explain complex topics in simple terms. Understanding these jobs helps you build AI that people actually want to use.
Here's how to apply JTBD to your AI product decisions:
Start by identifying the struggles. Talk to potential users about what they're trying to accomplish and where they get stuck. For a company building an AI-powered design tool, they might discover that users struggle with creating consistent brand assets across different formats.
Frame these struggles as jobs. A job statement follows this format: "When I [situation], I want to [motivation], so I can [outcome]." For our design tool example: "When I need to create marketing materials, I want to quickly generate on-brand designs for different platforms, so I can maintain brand consistency without hiring multiple designers."
Prioritize jobs based on importance and satisfaction. Which jobs matter most to users, and how satisfied are they with current solutions? This helps you focus your AI capabilities on areas where you can make the biggest difference.
A product manager at a financial services company shared how JTBD transformed their approach to an AI-powered financial assistant. "We initially built what we thought was cool - an AI that could predict market trends. But when we applied JTBD, we realized our customers weren't trying to predict the market. They were trying to avoid running out of money before payday. So we refocused our AI on helping them manage cash flow and avoid overdraft fees."
For AI products specifically, JTBD helps in three key ways:
It prevents technology-first thinking. Instead of starting with "What can our AI do?", you start with "What job needs doing?"
It clarifies success metrics. When you know the job, you know when your AI has done it well.
It guides ethical decisions. Understanding the real jobs helps you anticipate how your AI might be misused or cause unintended consequences.
To get started with JTBD for your AI product, conduct interviews with potential users focused on understanding their goals rather than their feature requests. Ask about the last time they tried to accomplish something related to your product area. What were they really trying to achieve? What got in their way?
Learn more about applying JTBD to AI product strategy
Remember - the most sophisticated AI in the world is worthless if it doesn't help people make progress in their lives.
CRISP-DM
Building AI products involves more uncertainty than traditional software. You're not just implementing known requirements - you're exploring what's possible with data and models that might not behave as expected. This is where CRISP-DM shines.
CRISP-DM (Cross-Industry Standard Process for Data Mining) provides a structured approach to data projects that works remarkably well for AI product development. While it was created back in the 1990s, its principles are perfect for navigating the complexity of modern AI.
The framework breaks down AI development into six phases that help product teams stay focused on business outcomes rather than getting lost in technical details:
Business Understanding: Start by clarifying what problem you're actually solving. A healthcare company building an AI diagnostic tool might define their goal as "reducing misdiagnosis rates for rare conditions" rather than just "implementing a classification model."
Data Understanding: Explore what data you have and what it can tell you. This phase helps you set realistic expectations about what your AI can actually do given your data limitations.
Data Preparation: Clean and structure your data for modeling. For AI products, this often consumes 60-80% of the total project time but is crucial for success.
Modeling: Build and train your AI models, testing different approaches to find what works best for your specific problem.
Evaluation: Test your models against business objectives, not just technical metrics. A recommendation system with 99% accuracy might still fail if it doesn't increase user engagement or sales.
Deployment: Integrate your AI into products and processes where it can deliver value to users.
A product leader at a retail tech company told me how CRISP-DM saved their inventory prediction project. "We were six months into development when we realized our model wasn't going to work in production. When we adopted CRISP-DM, we discovered in the data understanding phase that our historical inventory data had major gaps. We could have saved months of work if we'd followed this process from the start."
For AI products specifically, CRISP-DM helps in three key ways:
It forces business alignment early. By starting with business understanding, you ensure your AI solves problems that matter.
It acknowledges the iterative nature of AI. The framework is designed to loop back to earlier phases as you learn, which matches how AI development actually works.
It balances technical and business concerns. Unlike purely technical approaches, CRISP-DM keeps the focus on delivering business value.
To apply CRISP-DM to your AI product decisions, start by documenting your business objectives clearly before any technical work begins. Then assess your data assets honestly - what do you have, what's missing, and what limitations might that create? This upfront work will save you from pursuing AI features that sound exciting but can't be supported by your data.
Learn more about applying CRISP-DM to AI development
Remember - successful AI products don't come from jumping straight to the latest algorithms. They come from methodically working through business needs, data realities, and technical possibilities.
Lean AI
AI products are expensive to build, risky to launch, and often fail to deliver the value everyone hoped for. This is where Lean AI comes in - it applies the principles of Lean Startup to AI product development, helping teams learn quickly and avoid wasting resources on features users don't want.
The core idea is simple: treat your AI features as experiments designed to test hypotheses about what will create value for users. Instead of building a complete AI solution upfront, you build the minimum version needed to test your riskiest assumptions.
Here's how Lean AI works in practice:
Start with a clear hypothesis. "We believe that adding AI-powered product recommendations will increase average order value by 15%." This gives you a specific prediction to test.
Build the smallest version possible. Instead of a fully automated recommendation engine, you might start with recommendations selected by humans but presented to users as if they were AI-generated.
Measure the results. Did order values increase as predicted? If not, why not? This data helps you decide whether to invest in a full AI implementation.
Learn and iterate. Use what you learned to refine your approach before investing more resources.
A founder of an AI document analysis startup shared how Lean AI saved their company. "We were ready to spend nine months building an AI that could extract data from any document type. Before writing a line of code, we tested the concept by manually extracting data for customers but presenting it as automated. We discovered that customers only cared about three specific document types. We built AI for those three types in two months and started generating revenue immediately."
For AI products specifically, Lean AI helps in three key ways:
It reduces technical risk. By testing concepts before full implementation, you avoid building complex AI systems that don't deliver value.
It accelerates learning. Instead of waiting months to get user feedback on your AI, you can start learning in days or weeks.
It focuses on outcomes over technology. The emphasis stays on the value you're creating for users rather than the sophistication of your AI.
To apply Lean AI to your product decisions, start by identifying your riskiest assumptions. What needs to be true for your AI product to succeed? Then design small experiments to test these assumptions before committing to full development.
One powerful technique is the "Wizard of Oz" test, where humans perform functions that will eventually be automated. This lets you validate that users want the outcome before investing in the AI to deliver it.
Learn more about Lean AI product development
Remember - the goal isn't to build AI for its own sake. It's to create value for users and the business as efficiently as possible.