How Do Product Managers Classify AI Features Correctly?
For Product managers at tech companies · Based on Iseminger AI Landscape Explainer Framework
// TL;DR
Product managers use the Iseminger AI Landscape Explainer Framework to accurately place every AI feature — from fraud detection to chatbots — onto the correct layer of the nested AI hierarchy. This prevents mislabeling in roadmaps, ensures honest marketing claims, and gives stakeholders a clear mental model of what your product actually does. Use it whenever you're scoping AI features, briefing executives, evaluating vendor claims, or writing product positioning that references AI, ML, Deep Learning, or Generative AI.
Why do product managers need a framework for classifying AI?
As a product manager, you're constantly translating between engineering teams and business stakeholders. When your company ships an AI feature, the board wants to know: Is this real AI? Is it Generative AI? How does it compare to what competitors are doing?
Without a structured classification approach, you risk mislabeling a Machine Learning feature as Generative AI — or worse, failing to recognize that your chatbot sits on a Foundation Model while your competitor's doesn't. The Iseminger AI Landscape Explainer Framework gives you a repeatable 8-step workflow to classify any AI technology onto the correct layer of the Nested Venn Diagram: AI → Machine Learning → Deep Learning → Generative AI.
How do you classify your product's AI features step by step?
Start by listing every feature that your team calls "AI." For each one, run the Iseminger workflow:
1. Confirm it's AI — Does it simulate human intelligence (learning, inference, reasoning)?
2. Test for ML — Is it learning from data rather than following hardcoded rules? Does more training data improve its predictions?
3. Test for Deep Learning — Does it use multi-layer neural networks? Can you observe interpretive opacity (valid outputs whose reasoning is hard to decompose)?
4. Test for Generative AI — Is it built on a Foundation Model? Does it generate new content — text, images, audio, video?
5. Map the timeline — Is this feature using mature ML techniques from the 2010s, or is it leveraging the recent Foundation Model adoption explosion?
6. Label the capability — Pattern recognition? Prediction? Outlier detection? Content generation?
7. Flag dual-use risks — Especially for Generative AI features, name both legitimate applications and potential abuse vectors.
8. Calibrate for audience — When presenting to the board, use the autocomplete analogy for LLMs and acknowledge simplifications openly.
For example, your fraud-detection module learns from transaction data and flags outliers — that's Machine Learning. Your customer-service chatbot generates conversational responses using an LLM — that's Generative AI. Both are AI, but they sit at different depths of the hierarchy, have different maturity profiles, and carry different risk profiles.
How does accurate AI classification improve your roadmap and positioning?
When you label features correctly, three things happen:
- Roadmap accuracy — You can sequence features by technological maturity. ML-based features use proven 2010s-era approaches. Generative AI features ride the adoption explosion but carry more interpretive opacity and dual-use risk.
- Honest positioning — You avoid the credibility trap of calling everything "Generative AI." Stakeholders and customers increasingly understand these distinctions. Mislabeling erodes trust.
- Vendor evaluation — When evaluating third-party AI tools, the framework lets you test whether a vendor's "AI" is actually rule-based (expert system era), data-driven (ML), or Foundation Model-powered (Generative AI). This prevents overpaying for relabeled legacy technology.
What about communicating AI features to non-technical executives?
The framework's audience calibration step (Step 8) is designed for exactly this. Use the Nested Venn Diagram as a visual — four concentric circles with your products placed on the appropriate layers. Use the music analogy for Generative AI: all notes already exist, yet new songs are genuinely created. State your simplifications openly before delivering them. This builds trust with executives who may later consult technical advisors.
Next step: Pick one AI feature from your current roadmap and run it through the full 8-step Iseminger workflow. Document which layer it occupies, its timeline position, and its primary capability. Use that classification in your next stakeholder update.
// FREQUENTLY ASKED QUESTIONS
How do I tell if a vendor's product is really AI or just rule-based automation?
Apply the Iseminger framework's Step 2: test whether the system learns from data or follows explicitly programmed rules. If it uses hardcoded if-then logic, it may be an expert system (an older form of AI) but is not Machine Learning. If it improves with more training data and spots patterns autonomously, it qualifies as ML. This single test prevents you from overpaying for rebranded legacy software.
Should I call my product's ML feature 'AI' in marketing materials?
Yes — Machine Learning is a subset of AI, so calling it AI is technically accurate. However, the Iseminger framework warns against implying it is Generative AI if it doesn't generate new content via a Foundation Model. Be precise: say 'AI-powered fraud detection using Machine Learning' rather than just 'AI.' This honesty builds credibility and avoids regulatory scrutiny as AI marketing claims face increasing oversight.
How do I explain why our chatbot is different from our recommendation engine to the board?
Use the Nested Venn Diagram. Your recommendation engine sits in the Machine Learning layer — it learns from user behavior to predict preferences. Your chatbot sits in the Generative AI layer — it generates new conversational responses using a Large Language Model (a Foundation Model). Both are AI, but the chatbot represents the recent adoption explosion driven by Foundation Models while the recommendation engine uses more mature ML techniques.