Drilling Down on AI Depth — The 31 Key Questions to Answer Before Making an Investment Decision // or: how VCs look at your AI venture

|

Content Piece by Tim Winter


 

This is an article I wish I had written one year ago — one that would have been incredibly helpful when evaluating the numerous AI ventures we reviewed at Peak over the last 12 months.

During our due diligence process at Peak, we use a list of 100 critical questions to guide us in deciding whether or not to issue a term sheet. The goal is to work through and answer these 100 questions as fast as possible to make an informed decision.

These questions span our four core pillars, the 4 T’s — Team, Thesis, Traction, and Timing — and cover distinctions between various business models, like SaaS or marketplaces, across both B2B and B2C sectors.

Now, with category-defining shifts in AI, I am expanding this list to focus on a vital new factor: AI depth. This list is designed to help you evaluate AI’s technological depth and impact within a venture.

Let me be clear upfront: AI depth alone does not equal a great or poor investment case. Instead, the list of questions is meant to elevate your assessment beyond the obvious. When it comes to AI, relying solely on your gut feeling isn’t enough.

Initial Distinguishment: Before diving deeper, it’s crucial to distinguish between companies developing their own AI technology versus those merely utilizing AI tools or databases. Simply using AI is not a selling point by itself — especially when the real question is whether it contributes to solving the customer’s pain point. Ultimately, clients care about the solution, not whether AI is involved.

First, let’s define the 5 dimensions of AI Depth:

1. Assessing the Depth of AI Integration

Criteria:

  • AI’s Role in Core Operations: Determine how central AI is to the venture’s core operations and value proposition. Is AI an essential component of the product or service, or merely a supplementary tool? Determine whether the business would be feasible without AI.
  • Technical Architecture: Analyze the tech stack to understand the complexity and sophistication of the AI models being used. Look for custom-built models over off-the-shelf solutions.

Evaluation Steps:

  • Product Demos and Technical Deep Dives: Request comprehensive demos and technical breakdowns. For example, a venture could show advanced proprietary algorithms and custom solutions, as opposed to ventures that implement a basic application layer.
  • Technical Due Diligence: Engage with technical advisors or experts to assess the quality and uniqueness of the AI technology being utilized.

2. Talent and Innovation Pipeline

Criteria:

  • Team Expertise: Evaluate the experience and expertise of the founding team and key hires in AI and machine learning.
  • R&D Commitment: Assess the company’s commitment to ongoing research and development in AI.

Evaluation Steps:

  • Founders and Key Hires: Conduct background checks and interviews with the founding team and key technical personnel. For example, ventures like Anthropic are founded by recognized AI experts.
  • Research Output: Review published papers, research backgrounds, patents, and participation in AI conferences to gauge the company’s active contribution to AI innovation.

3. Evaluating Proprietary Data

Criteria:

  • Data Uniqueness: Evaluate the exclusivity and quality of the data the venture uses. Proprietary datasets are a significant differentiator.
  • Data Acquisition Strategy: Understand the company’s strategy for acquiring and maintaining data. Look for sustainable and scalable data collection methods. Assess whether the product is enhanced by a data flywheel structure. Next to hosting costs and inference costs, no more direct data costs in acquisition.
  • Data Quality: Evaluate the company’s data quality assurance processes and the safeguards in place to ensure the reliability of their AI models.

Evaluation Steps:

  • Data Audits: Conduct a detailed audit of the data sources to assess how much of the data is proprietary versus publicly available.
  • Data Utilization: Analyze how effectively the company uses its data to train and refine its AI models.

4. Scalability and Network Effects

Criteria:

  • Solution Scalability: Assess the scalability of the AI solutions. Scalable solutions should be able to handle increased data and user loads efficiently.
  • Potential for Network Effects: Evaluate whether the venture’s business model benefits from network effects, where the value increases as more users or data are added.

Evaluation Steps:

  • Scalability Tests: Review performance metrics and scalability plans with regards to GPU or cloud provider capacities needed. Companies should demonstrate robust scalability in their data platforms.
  • Growth Projections: Analyze user growth projections and how this growth will enhance the AI’s effectiveness and strengthen the venture’s competitive edge.

5. Level of Innovation and Business Impact

Criteria:

  • Core Technology vs. Thin AI Wrappers: Differentiate whether the company is developing core AI technology or simply applying existing AI (e.g., ChatGPT wrappers). Assess how closely the venture is tied to large language models (LLMs) — how many layers exist between them, and how dependent they are on third-party AI. Which specific areas of AI technology does the venture build or touch that are being assessed?
  • Value Proposition: Understand the unique value the company brings. Is it creating a fundamentally new AI capability, or is it simply an interface over existing tech?

Evaluation Steps:

  • Technology Analysis: Conduct a thorough analysis of the underlying technology. Core AI companies will have significant innovation in their models and algorithms, while application layer companies might primarily focus on user interfaces and integrations.
  • Market Differentiation: Look at the competitive landscape to determine whether the venture is offering something truly unique or merely repackaging existing solutions. For example, a company solely building a new chatbot using GPT-4 may not have the same depth as one developing an entirely new AI model from scratch.

The 31 Questions to Assess AI Depth:

1. Depth of AI Integration

  • Core Technology:
  1. Is AI central to the venture’s product/service? (Yes/No)
  2. Is the company developing core AI technology? (Yes/No)
  3. How much of their AI is built in-house versus using existing models?
  4. Are the AI models custom-built or off-the-shelf?
  5. Is the company leveraging AI-native UI (text, video, voice input)?
  • Technical Complexity:

6. Does the company demonstrate sophisticated AI usage? (Yes/No)

7. How would you rate the level of sophistication from 1–10?

8. Are there clear differentiators in their AI approach that give them a competitive edge?

2. Talent and Innovation Pipeline

  • Team Expertise:

9. Does the team have a strong AI and machine learning background? (Yes/No)

10. Are there notable AI researchers or experts on the team?

  • R&D Commitment:

11. Is the company investing in ongoing AI research and development? (Yes/No)

12. What percentage of their budget is allocated to AI R&D?

13. How much R&D budget is needed vs attributed?

14. Are there partnerships with academic institutions, research organizations, or industry leaders?

3. Proprietary Data Evaluation

  • Data Uniqueness and Quality:

15. Does the venture have access to unique, proprietary data? (Yes/No)

16. What are the sources of their data?

17. How does the company ensure high data quality?

18. Does the company have processes in place to prevent data biases?

  • Data Strategy:

19. Is there a sustainable and scalable data acquisition strategy? (Yes/No)

20. How much are they paying for data collection and how will the costs develop over time?

21. How well is the generated data collected and utilized to enhance AI models over time?

22. Does the company have mechanisms in place to secure data privacy and comply with data regulations

4. Scalability and Network Effects

  • Solution Scalability:

23. Can the AI solutions scale effectively with increased data and user loads? (Yes/No)

24. Is there a clear plan for managing scalability?

25. How does the cost structure scale with growth in AI processing (e.g., inference, training) in terms of P&L impact?

26. Does the company have a clear understanding how their economics will evolve over time? (different economics in AI vs classic SaaS)

  • Network Effects:

27. Does the business model benefit from network effects? (Yes/No)

28. How do additional users or data improve the AI’s performance?

5. Level of Innovation and Business Impact

  • Value Proposition:

29. Does the company’s AI solution lead to a tenfold improvement compared to existing solutions in terms of P&L impact? (read my other article going in detail here)

30. Is the value proposition unique and difficult for competitors to replicate (AI moat)? (Yes/No)

31. Does the company offer something fundamentally new or improved?