Protecting Yourself From Fake AI Startups and Software With a Better RFP Process
Published on April 21, 2025 in Strategic
The transformative power of AI in both business and private life is seldom out of the news. AI experts are quick to extoll AI’s power to give businesses a competitive edge, whether it is through the automation of routine tasks or providing in-depth predictive analytics. However, the public is less enthusiastic. The latest data from Pew Research shows that most believe the potential negatives outweigh the positives, and almost three quarters are worried about the misuse of AI.
Those concerns are not without foundation. Like any fast-moving and high-profile technological innovation, the surge in AI adoption has attracted its fair share of opportunists who are only out to make a quick buck and are not afraid to misrepresent, overhype or completely fake their capabilities. The result is not just wasted money. It also leads to stress, project delays and, potentially worst of all, damage to personal reputation internally or your business’s reputation externally.
A robust RFP process is core to protecting yourself and your business. Here, we explore how and why traditional RFPs can fall short and how they can be redesigned to mitigate the emerging risk presented by fake AI startups and software.
The Rise of Fake AI Startups
Every new technology attracts scammers and fraudsters. We saw it with digital payments, we saw it with crypto and now we are seeing it with AI. Experian listed AI fraud as the fastest growing area of tech scams in 2024, and the FBI recently posted a public service announcement listing some of the methods criminals use to ensnare victims using generative AI.
Similarly, many purported AI startups consist of little more than smoke and mirrors. Back in 2019, a study by London-based venture capital firm MMC revealed that almost half of the 2,830 AI startups they examined did not use AI in a material way. These are known as pseudo-AI, as they use humans to simulate machines.
Other fake startups simply use off-the-shelf AI algorithms that they market as proprietary technology. There are even extreme cases of vendors offering capabilities that are entirely fictitious, relying on buyers simply taking them at face value and not having the time, inclination or capability to carry out the right checks or due diligence.
Typical red flags that are common to fake AI startups include the following:
- Pitches that include plenty of buzzwords but little substance or technical detail
- Lack of whitepapers or other publicly available technical documentation
- Lack of demos
- Reliance on manual intervention in processes that should be automated
That sounds fine in theory, but the rapidly evolving and specialist nature of AI makes it difficult for non-experts to ask the right questions or to confidently differentiate wheat from chaff – especially in the face of a sales pitch that is polished and persuasive. This is why the RFP process needs to be robust and fit for purpose.
Traditional RFPs Fall Short
In the past, RFP processes were designed for the acquisition of tangible goods and professional services. This might be sufficient for straightforward software solutions, but the process is unlikely to ask the right questions when it comes to evaluating more sophisticated technologies such as machine learning models or natural language processing systems. This leaves the business exposed to fraudulent vendors who can obfuscate with vague answers to generic questions.
Furthermore, traditional RFPs are likely to lack the necessary evaluation criteria for AI. Their scoring systems do not include key differentiators such as the following:
- Model accuracy
- Data lineage
- Explainability
- Proof of concept
- Integration with existing systems
- Ethical considerations
Designing a Robust RFP Process for the AI Age
Designing an RFP process that is fit for purpose in the AI age is about more than just asking tougher questions. It means a shift in mindset from buying software to validating innovation and is a multi-step process:
1. Establish a cross-functional evaluation team
There are AI touchpoints across your entire organization. These include IT, operations, legal, marketing and more. The RFP process needs to include stakeholders from every relevant part of the business to facilitate holistic evaluation. Be sure to include technical experts who can effectively vet AI models and can ask informed questions.
2. Insist on detailed technical disclosures
Fake AI startups are rich in vague promises and buzzwords but short on substance. A genuine service provider will willingly provide the following:
- A high-level architecture of their solution
- Details of the underlying AI models they use
- Model performance metrics such as accuracy, F1 score or recall
- Information on retraining processes and adaptability
3. Demand a Proof of Concept (PoC)
A PoC is a practical demonstration of a product used to demonstrate the real world feasibility of an idea by testing functionality in your specific business environment. It effectively separates genuine vendors from those built on smoke and mirrors. The PoC should use your own data, which can be anonymized if necessary, and must have clear success criteria that are agreed in advance. If vendors shy away from a PoC or demand payment without demonstrating results, it is a clear sign to be on your guard.
4. Consider compliance, ethics and data privacy
AI opens up new areas of risk that are probably not contemplated by your traditional RFP. Make sure it delves into such areas as how the AI handles personal or sensitive data (including compliance with relevant data protection legislation), bias mitigation, transparency and explainability.
5. Ask for real-world references and case studies
Anyone can concoct fake testimonials and recommendations. The RFP must include requests for case studies with quantifiable results, contactable references who have used the product and evidence of repeat customers or long-term deployments
6. Look beyond the tech
When you are working with a startup, the team is as important as the software. So make sure the RFP delves into the backgrounds and expertise of the founders and their team. This might include research papers or contributions to open source projects. Also, dig into their support capabilities, response times and so on.
7. Use a weighted scoring matrix
With weighted scoring, you can ensure the RFP reflects your priorities and that decisions are based on genuine value as opposed to marketing gloss. Naturally, the criteria and weighting will vary from one business to the next, but the following example demonstrates the principle:
- Technical performance (25%)
- Integration and scalability (20%)
- Vendor credibility and references (20%)
- Data handling and compliance (15%)
- Cost (10%)
- Ethical AI practices (10%)
In Conclusion…
AI solutions are no more complex or mysterious than other tech tools. However, the newness and rapid evolution of AI in combination with the media hype can make it seem like a dark art. In that environment, even a mature business with robust internal controls can be misled by a slick sales pitch and left counting the cost.
Ensuring the RFP process is fit for purpose does more than protect your business from fake AI vendors; It also facilitates smarter investments that can build internal and external trust, and deliver real value and competitive advantage by partnering with truly capable AI innovators.
Sources
https://www.experian.com/blogs/ask-experian/the-latest-scams-you-need-to-aware-of
https://www.ic3.gov/PSA/2024/PSA241203
https://www.mmcventures.com/wp-content/uploads/2019/02/The-State-of-AI-2019-Divergence.pdf
https://www.techtarget.com/searchcio/definition/proof-of-concept-POC