Defensibility in the Age of AI: 5 Questions to Ask Before You Build

The AI revolution is here, and every founder, strategist, and business leader is rushing to build. The promise is intoxicating: create powerful tools fast, cheaply, and with unprecedented intelligence.

I've been in that rush. I've built, refined, and pivoted an ambitious AI-powered venture, only to confront the stark reality: the biggest risk in AI venture building isn't technical failure, it's strategic fragility. Most new products are built on a house of cards: high token burn, easily replicated features, and zero defensible intellectual property.

🤔 You can build something fast, but can you build something that lasts?

Before you invest another hour, another cent, or another ounce of emotional energy, you should ask yourself:

Is my product a powerful feature waiting to be absorbed by GPT-6, or a durable business built on a lasting and defensible moat?

This article provides a five-question framework designed to help you uncover your product's defensibility angle in the era of AI hyper-acceleration. And for each question, l offer practical suggestions on how to fix your idea if your answer is “no”.

🔦 The Utility Audit 

This is where true value engineering begins. Many AI products are "solutions looking for a problem"—a clever use of tech that doesn't genuinely alleviate a user's pain point beyond what a well-crafted prompt in a general LLM could do. You're building convenience, not a necessity.

Q1. Is this a problem AI solves, or a problem AI creates?

Ask yourself:

• Does my solution deliver 10x the utility (benefit) of simply using a general LLM with a good prompt?

• Am I building an expensive feature that a single Co-pilot update will render irrelevant?

• Is the value proposition human-centred (solving a validated Job-to-be-Done) or technology-centred (just showing off the latest AI capability)?

Did your AI business idea fail against one of these questions? It’s okay. Here’s how to fix it:

• Go back to Job-to-be-Done (the tasks your users need to complete): Interview/ reinterview your target users. What actual problem are they trying to solve? And how are they currently attempting to solve them? Are they already using AI against this problem? And if so, what prevents them from succeeding? Does your AI truly accelerate, simplify, or enable a task that was previously impossible or prohibitively difficult, or can your audience simply educate themselves to use existing AI solutions better? 

• Focus on the "Last Mile": Instead of replacing human intelligence, use AI to automate the laborious parts of a task, freeing up humans for higher-value decision-making or creative work. The value is in the human outcome, not the raw AI output.

• Don't Compete with Free: If a powerful LLM can do 80% of what your product does for a fraction of the cost (or for free, if bundled with other software), your AI is a feature, not a business.

🧮 The Token Economics Test

The illusion of cheap AI is a silent killer. While initial API costs appear negligible, the reality of complex, high-context interactions will rapidly erode your margins. For a service requiring deep understanding and ongoing conversation (like advisory, legal, or complex customer support), the AI must repeatedly process large volumes of information—a phenomenon I call "Context Creep."

 Q2. Does your cost structure survive Context Creep?

Ask yourself:

• For complex user needs (e.g., high-context advisory or analysis), what is the true cost when the model must re-ingest 5-10 million tokens per month per user?

• Can the service model still achieve a positive margin if the cost of the base LLM drops by 50% and a competitor integrates it for free?

• Does the cheap starting price hide a very high running cost later on?

Did your AI business idea fail against one of these questions? It’s okay. Here’s how to fix it:

• Redesign for Transactional AI: Can you break down the user's problem into smaller, less context-heavy queries? Only use AI where it's most impactful, supplementing with static content or human-curated resources where context is needed.

• Build a Hybrid Model: Shift the core value away from raw AI output. Perhaps AI generates draft responses that are then reviewed by humans, or AI summarises data that humans then interpret. The customer pays for the human touch, making the token cost a controllable expense.

• Scope Down: Drastically reduce the scope of your AI's responsibility to only the most impactful, low-context tasks, where token burn is predictable and low.

⛫ The Proprietary Data Moat

In the age of AI, data is the new oil. But not all data is created equal. A simple knowledge base scraped from public sources will NOT offer defensibility. Your data moat must be a genuine barrier to entry for competitors.

Q3. Is your data truly gated, unique, and irreplaceable?

Ask yourself:

• Can a competitor license, scrape, or buy the data you’re using to train your RAG model? If so, it's not a moat.

• Is your data defensible because it's Proprietary (only you have it), Exclusive (governed by a unique partnership), or Dynamic (constantly updating in real-time, making static scrapes useless)?

• If your entire dataset were stolen, would your AI product still be worth its subscription price?

Did your AI business idea fail against one of these questions? It’s okay. Here’s how to fix it:

• Focus on Unique Data Generation: Can your product itself generate unique, opt-in user data that improves its service (e.g., through user interactions, feedback loops, or anonymised usage patterns)? This creates a network effect.

• Forge Exclusive Partnerships: Seek out partnerships with organisations that hold unique, gated datasets (e.g., industry associations, research bodies, large enterprises) that you can license exclusively.

• Curate and Structure for Superiority: Even with public data, superior curation, annotation, and structural indexing can create a temporary advantage. However, be aware that this is short-lived because it can be replicated.

👥 The Distribution Moat

A brilliant AI solution is useless if no one knows it exists, or if acquiring customers costs more than they'll ever pay. Your distribution strategy needs its own moat, especially as AI tools become more and more commoditised.

Q4. How will you acquire users without a price war?

Ask yourself:

• Do you have access to a specific niche community or channel partnership that large players cannot easily access or purchase?

• Is it too difficult or expensive for customers to stop using your product? (The Workflow Moat).

• Are you selling a small AI tool, or a whole new, much easier way to finish a daily job?

Did your AI business idea fail against one of these questions? It’s okay. Here’s how to fix it:

• Build a Niche Community: Instead of targeting broad markets, focus intensely on a specific, underserved niche. Build trust and provide unique value within that community. Distribution becomes word-of-mouth, and word-of-mouth becomes a loyal customer base.

• Integrate Deeply into Workflows: Position your AI as an indispensable component of existing, high-friction workflows. This creates switching costs. Think less "app" and more "invisible intelligence layer."

• Strategic Partnerships for Go-to-Market: Partner with established players (e.g., SaaS platforms, consulting firms) who already have access to your target audience and can distribute your solution.

🧐 The Irreplaceable Human Moat:

A "human moat" can be a certified expert, a trusted brand, or a vibrant community. It provides the warmth, judgment, and emotional connection that AI cannot (yet) replicate. 

Q5. What is the irreplaceable non-AI layer of your value?

• Is the final output certified, regulated, or validated by a human expert (e.g., a doctor, lawyer, certified coach)?

• Does AI help people do their job much faster (which they pay for), or are you just selling the AI's words?

• Is your brand positioned as a community, identity, or status symbol (like Alo or Lululemon) rather than just a functional tool? (The Brand Moat).

Did your AI business idea fail against one of these questions? It’s okay. Here’s how to fix it:

• Blend Human + AI: Use AI to augment human capabilities, not replace them. For example, AI generates the first draft, and a human expert refines it. The customer pays for the human-grade quality and accountability.

• Build a Brand for Trust and Belonging: Focus on creating a brand identity that evokes trust, community, or a sense of belonging. People buy into brands, not algorithms.

• Develop a Unique Methodology/Framework: If your AI is built around a proprietary methodology (e.g., a unique coaching framework, a specific strategic planning model), the value is in that validated approach, not just AI executing it.

⛲️ The Path Forward: From Idea to Enduring Value

The Age of AI demands a new level of strategic rigour. It's easy to get caught up in the excitement of what AI can do, but the true challenge is building what AI can’t easily replicate. By asking yourself these five questions, you're not just auditing your business idea; you're future-proofing your business. 

Let's move beyond the hype and build ventures that provide lasting value, foster true creative independence, and withstand the relentless pace of innovation!

---

Which of these questions made you rethink your current business idea? Share your insights—I'd love to hear how you're building defensibility.