AANZ February Expert Session with Saba Samiei, Mitali Purohit & James Palmer | AI Frontier

This podcast link and summary are from the AI Frontier Expert Session held online in February 2024.
Podcast: Navigating the AI Frontier: Essential Due Diligence for Early-Stage Ventures

Unpacking the Nuances of AI Investment and Growth in a Rapidly Evolving Landscape

The venture capital landscape is continually reshaped by technological advancements, with Artificial Intelligence (AI) currently at the forefront. A recent expert panel session hosted by the Angel Association of New Zealand brought together leading voices: Vitali (General Partner at Nuance VC and session moderator), James Palmer (Associate at Blackbird), and Sabah (Founder and CEO of Maxo Technologies and Comfort AI), to demystify AI investments and illuminate the critical aspects of due diligence for early-stage AI ventures. This discussion, driven by growing investor demand, provided invaluable insights for business leaders, analysts, and investors looking to navigate the complexities and opportunities within the AI sector.

The core purpose of the session was to provide perspectives from both founders on preparing for fundraising and investors on assessing AI opportunities. A key challenge highlighted was differentiating between an actual AI company and merely utilising AI, underscoring the need for specialised due diligence. The insights reveal that successful AI investment hinges on understanding the technical underpinnings, strategic data management, robust IP protection, and adaptable market distribution models, alongside a foundational belief in the entrepreneurial team.

Understanding the AI Ecosystem: Beyond the Hype

A crucial starting point for any investor is understanding what constitutes AI. Sabah, an expert in the field, provided a foundational overview, defining AI as any technology replicating human intelligent behaviour, such as navigating, talking, understanding, or problem-solving. She meticulously broke down different AI paradigms, including:

• Supervised Learning: Used when input and output are known (e.g., Google Images categorising objects), requiring high data labelling effort.
• Unsupervised Learning: Input is known, but the desired output is unknown, allowing the model to find relations and trends (e.g., Amazon for marketing trends), requiring the least labelling effort.
• Semi-supervised Learning: A hybrid approach, combining labelled and unlabeled data (e.g., Google speech recognition).
• Reinforcement Learning: The model learns through consequences, receiving positive or negative points (e.g., Boston Dynamics Spot Robots, AlphaGo).
• Deep Learning: Often a subset of supervised learning, involving large models with many layers, exemplified by Tesla’s autopilot system.
• Transfer Learning & Ensemble Learning: More advanced methods like ChatGPT (transfer learning) and Netflix’s recommendation system (ensemble learning) leverage knowledge transfer or combined models for personalised outcomes.

Sabah also clarified what AI is not: traditional algorithms (where a programmer defines the exact path), search algorithms, threshold-based algorithms, encryption algorithms, and standalone data analytics/visualisation. In her opinion, the unique power of AI lies in its ability to create “customisation at scale”. The end-to-end process of creating an AI model involves data collection, engineering (cleaning, noise reduction), model generation, deployment, and continuous feedback for improvement. Importantly, this process is rarely fully automated, with significant manual intervention, particularly in data engineering and model creation.

Investor Playbook: Assessing AI Opportunities

James Palmer outlined Blackbird’s investment philosophy, emphasising that they don’t treat AI opportunities fundamentally differently from other technology ventures. Their primary focus remains on “outlier founders who capture our imagination,” possess “unfair access to talent,” and have “hard-earned customer and problem insights”. Due diligence at the earliest stages is heavily oriented around the people behind the venture, especially when little customer revenue data is available.

James categorises the AI market into three areas:

1. Foundational Models: Capital-intensive, building underlying API and model capabilities (e.g., OpenAI, Google Bard). These are generally less investable for early-stage funds unless there’s a highly differentiated approach to data capture or a vertical application.
2. Picks and Shovels (Developer Tooling): Companies providing infrastructure for AI development and deployment (e.g., vector databases, ML ops, security layers). Success often comes from winning developers’ hearts and minds.
3. Applications: End-to-end applications leveraging AI. AI should be viewed as a tool to create customer value, similar to any other software business. This is generally where Blackbird focuses its investments.

For investors, validating an AI venture involves:

• Customer Insights: Talking to customers and observing beta testing results to validate problem veracity and potential market value.
• Expert Opinion: Leveraging external data sources and portfolio companies for feedback on developer tools.
• Team Credibility: Trusting the founding team’s expertise, as directly assessing technical details like data quality can be challenging for non-experts.
• Product-Market Fit: Evidenced by “explosive product uptake,” post-founder sales, and exploding organic inbound referrals. Revenue growth rate and, for developer tools, metrics like GitHub stars, are key indicators.

The Founder’s Journey: Building and Protecting an AI Venture

Sabah shared her journey with Maxo Technologies, which aims to process various sensory inputs for environmental monitoring and risk detection. Her decision to pursue Maxo was driven by the initial idea’s application-based nature and her partners’ ambitious vision at Bridgewest Ventures. A significant learning curve for Sabah was simplifying her complex vision for investors, many of whom lacked a technical background. She found it challenging to explain concepts like patentability, as AI models are not patentable (they are mathematical algorithms), but their applications and commercialisation methods are.

Key founder strategies and challenges include:

• Simplifying the Vision: Articulating a big vision while focusing clearly on the immediate application.
• Educating Investors: Taking investors on a journey to understand new AI paradigms, especially those with traditional software backgrounds.
• Data Quality Validation (from a founder’s perspective): Instead of investors directly testing data quality (which is highly specialised), founders should demonstrate solutions through free deployments and rigorous customer feedback loops. The goal is to reach a point with minimal false positives or negatives before commercialisation. Investors can validate this by speaking to early customers.
• IP and Moat Building:
  • Trade Secrets: Unique approaches developed internally.
  • Application Patents: Patenting the unique commercialisation of AI models, not the models themselves.
  • Team Structure: Designing the company so that no single employee can replicate the entire application, making technical teams interdependent.
  • Aggressive Commercialisation: Being first or second to market and rapidly expanding to outcompete potential rivals.
  • Proprietary Data Ownership: Critical for AI companies, especially for environmental or unique data sets. Contracts should clearly define data ownership (e.g., Maxo owns raw environmental data, while aggregated dashboard data may be shared with customers).
  • Data Currency: Keeping data sets current is vital, especially for human-related applications where language evolves. Models may plateau for consistent environments (like farms), requiring only outlier data for updates.
  • Cost vs. Value of Data: The cost of capturing new, unique data must be significantly less than the value derived from the ultimate model.
• Market Entry and Distribution:
  • Broad Net Strategy: This strategy is mainly for startups trying diverse channels and analysing effectiveness (e.g., cold outreach, conferences, partnerships).
  • Strategic Partnerships: Collaborating with hardware creators or larger enterprises (e.g., Avnet, Halo AI, Verbeek) for broader market access, particularly in non-tech-savvy industries.
  • Targeting the Top of the Food Chain: For B2B, focusing on large distributors or pharmaceutical companies (e.g., Merck, Zoetis) rather than individual end-users.
  • Leveraging Networks: Actively seeking customer introductions and connections from board members and investors.
  • Navigating Regulated Markets: Recognising that highly regulated industries (healthcare or government) add significant complexity and slowness. Maxo strategically avoided these for its initial go-to-market to focus on tech development without regulatory hurdles.
  • Addressing Regulatory Risks (Copyright): A significant concern, especially for generative AI. Investors must ask about data ownership: was the data purchased, are profit-sharing agreements in place with artists, or is the data free to use? Companies relying on existing internet data without proper licensing face high legal risks. Incumbents with proprietary, decades-long data sets (e.g., Adobe, Canva) have a significant advantage here.

Scaling and Sustaining Innovation in AI

The rapid pace of AI innovation requires companies to adapt and scale continuously. James emphasised looking for evidence that a founding team can build and operate “with velocity” and has an “unfair advantage in magnetising great talent”. Access to top-tier talent is a significant productivity lever in early stages.

Furthermore, companies need “nuance around mode of distribution” around how they acquire customers more cheaply at scale and demonstrate customer love faster. This could involve novel, viral distribution methods rather than traditional marketing spend. The distribution model must align with the product’s pricing and Annual Contract Value (ACV); high-ticket enterprise sales can justify complex, in-person distribution, while lower-priced consumer products require rapid, viral adoption.

Conclusion: Cultivating Informed AI Investment

The discussion underscored that successful AI investment is not just about understanding the technology but deeply understanding the founders, their strategic approach to data, IP, and market entry, and their ability to execute with speed and attract top talent. While technical due diligence can be complex, investors can rely on expert advisors like Sabah to cut through the jargon and obtain accurate answers. Resources like Coursera, YouTube, podcasts, and even generative AI tools like ChatGPT can provide foundational knowledge for those looking to educate themselves. Ultimately, investing in AI means investing in visionaries who can navigate a dynamic, data-intensive, and often unpredictable technological frontier.