AI Reality vs. AI-Washing

Hello! Welcome to the second lesson in our special topic module on AI startup diligence.

In our last lesson, we focused on what makes a strong AI business, exploring the concepts of technical feasibility and defensibility. You learned to assess if a team can actually build their claimed solution and, if so, whether they can protect it from competition using moats like process power, data network effects, and counter-positioning.

This lesson tackles the flip side of that coin and addresses the learning outcome: Distinguish between genuine AI innovation and superficial 'AI-washing'.

AI washing is the practice of overstating or fabricating the use of AI to capitalize on the hype. As you prepare to launch an AI-focused incubator, the ability to quickly and accurately identify these superficial claims is a fundamental screening skill. It will save you immense time and, more importantly, prevent you from investing in businesses built on a fragile foundation of marketing buzz rather than genuine innovation.

1. What is AI Washing?

Just as "greenwashing" emerged to describe false environmental claims, "AI washing" has become the term for exaggerated or misleading claims about a company's use of artificial intelligence.

To ground our understanding, let's start with a formal definition and context.

As the report highlights, AI washing can range from using buzzwords without substance to making outright false claims about a product's capabilities. A study cited by Mercanis found that 40% of European startups presenting themselves as "AI startups" didn't actually use real AI in a meaningful way.

The most common form of AI washing today involves building a thin "wrapper" around a publicly available large language model (LLM) from a provider like OpenAI or Anthropic. The startup's product is often just a user interface that passes prompts to the underlying model's API. While these can be built quickly and look impressive, they often represent weak and unsustainable businesses.

2. The "AI Wrapper" and its Unsustainable Economics

A key indicator of potential AI washing is a business model that lacks a path to profitability. Many simple "AI wrappers" fall into this category. Unlike traditional Software-as-a-Service (SaaS), where the cost to serve an additional customer is near zero, AI-native products have a real, ongoing computational cost for every user action (inference). This fundamentally changes the unit economics.

This video provides an excellent, numbers-focused breakdown of why many of these businesses are economically unviable.

The video makes a critical point that serves as a powerful first-pass filter: Would this product solve a real problem without AI? If the answer is no, and the entire value proposition is "we use AI," you should be highly skeptical. A genuine innovation uses AI to solve an existing problem 10x better, faster, or cheaper, rather than using AI for its own sake.

3. A Practical Framework for Spotting AI Washing

As an investor, you need a systematic way to probe a startup's claims. Your due diligence process should be designed to separate hype from reality. Here is a framework combining key signals.

Signal 1: Scrutinize the Claims and the Team

Genuine AI companies are typically transparent about their technology. Vague, evasive language is a major red flag.

The Mercanis blog provides a clear, actionable checklist for this.

Going a step further, the CFA Institute report suggests that one of the easiest ways to verify a firm's AI claims is to investigate the people behind them.

A startup claiming to build a novel AI system should have founders or key technical leaders with demonstrable experience in machine learning, data science, or a related field. If the "Head of AI" is a long-time marketing executive with a recently completed online certificate, it's a sign that the company's AI efforts may be more for show than for substance.

Signal 2: Demand Proof of Value (Traction)

In the current environment, a slick demo is easy to create. An investor's most reliable signal against AI washing is not the demo, but customer validation. A startup that has convinced a customer to pay for their product, even an early version, has demonstrated it solves a real problem worth solving.

This concept is brilliantly illustrated in the following pitch deck review, which analyzes a startup that is successfully securing VC meetings precisely because it has this proof.

As Kang emphasizes, development is now the easier half of the equation; distribution and proving value are what investors are looking for. Your diligence process should prioritize evidence of market pull—letters of intent, pilot customers, and especially paying customers—over technical wizardry alone.

Signal 3: Ask the Tough Technical Questions

Finally, for startups that pass the initial screens, your CS background allows you to go deeper. You don't need to be a research scientist, but you should be comfortable asking pointed questions to test the depth of their technical claims. The goal is to see if they can move beyond buzzwords to specifics.

The CFA Institute report provides an excellent list of such questions. For an early-stage investor, the most relevant ones are those that probe the core of their claimed innovation.

Here is a curated set of questions, adapted from the CFA report and tailored for pre-seed AI diligence:

• Model & Data:
  • "Can you specify what type of model you're using? Are you fine-tuning a base model or building from scratch?"
  • "What data sources are you using to train or fine-tune your model? Is this data proprietary?" (We will dive much deeper into this in the next lesson).
  • "How does your AI-driven model outperform simpler, non-AI heuristics? Can you show a comparison?"
• Robustness & Overfitting:
  • "What precautions are you taking to guard against overfitting?"
  • "How do you validate the model's performance? Can you walk me through your evaluation framework?" (This connects back to our previous lesson).
• Interpretability:
  • "Can you provide an example of a recent decision influenced by the model’s output? How was the rationale explained to the team?"

A team with genuine expertise will welcome these questions and provide specific, confident answers. A team engaged in AI washing will likely respond with vague marketing-speak.

Test your understanding!

You are evaluating a startup called "MarketAI". Their pitch is to "use the power of generative AI to create high-conversion marketing copy for e-commerce stores."

Here's what you learn during diligence:

• Product: The product is a web app where users input a product description, and it generates ad copy, emails, and social media posts.
• Technology: When asked about their model, the CEO says they use "a proprietary blend of advanced intelligent algorithms, including GPT-4." They refuse to give more detail, citing their "secret sauce."
• Team: The founding team consists of two experienced marketing executives. Their CTO is a recent computer science graduate whose main experience is in web development.
• Traction: They have a beautiful, functional demo. They have 1,000 free-trial sign-ups but have not yet converted any to a paid plan. They say they are "focused on growth before monetization."

Based on the framework from this lesson, identify at least three red flags that suggest MarketAI might be an example of AI washing.

Show answer

Here are several red flags indicating AI washing:

1. Vague Technical Claims & "Secret Sauce" Defense: Their response about using a "proprietary blend of advanced intelligent algorithms" is a classic example of vague marketing-speak. A genuine team would be able to articulate their specific approach (e.g., fine-tuning, retrieval-augmented generation, prompt engineering chains) without giving away proprietary code. Hiding behind the "secret sauce" defense for high-level architectural questions is a major red flag.
2. Lack of Deep Technical Expertise on the Team: While the marketing experience is relevant, the lack of a seasoned AI/ML leader is concerning. A CTO with only web development experience is likely not equipped to build a defensible, novel AI system beyond a simple API wrapper around GPT-4.
3. No Customer Validation (No Paying Customers): 1,000 free-trial users demonstrate good top-of-funnel marketing, but the inability to convert any to paid customers is a critical failure. It suggests the product is a "nice-to-have" toy, not a must-have tool that solves a real business problem. They haven't proven value.
4. Likely a Thin "Wrapper" with Poor Economics: The description strongly suggests the product is a simple interface for the GPT-4 API. This means they have no technical moat (anyone can build this) and are subject to OpenAI's pricing, likely resulting in very thin or negative margins, especially if they offer a low-priced subscription.

Conclusion

You now have a robust framework for looking past the hype and identifying superficial AI claims. Your ability to apply this filter will be invaluable as you build your deal flow and reputation as a savvy AI investor.

Key Takeaways:

• AI Washing is Widespread: Be skeptical by default. Many companies claiming to use AI are simply leveraging buzzwords for marketing.
• Follow the Economics: A thin "AI wrapper" is often an unsustainable business with no moat and poor margins. Ask if the product would be valuable even without AI.
• The Ultimate Test is Traction: A slick demo is not enough. The most powerful signal against AI washing is evidence that a customer finds the product valuable enough to pay for it.
• Trust, but Verify: Scrutinize the team's expertise and don't be afraid to ask specific technical questions. Evasiveness is a clear red flag.

Preview of the Next Lesson

In this lesson, we identified "proprietary data" as a key differentiator between a genuine AI innovation and a simple wrapper. But what makes a data strategy strong and defensible? In our next lesson, we will dive deep into this topic to address the learning outcome: Analyze the components of a startup's data acquisition and annotation strategy. You will learn how to evaluate the quality, defensibility, and strategic value of a startup's most critical asset: its data.