5 Questions To Ask When Someone Is Trying To Sell You A New AI Product?

5 Questions To Ask When Someone Is Trying To Sell You A New AI Product?

Artificial Intelligence breakthroughs, new user interfaces, APIs, and use cases are propelling a new wave of startups and blue chip companies to embrace AI in their products and services.

At Penta, we have been building AI-powered products for over a decade and have accumulated a wealth of knowledge in discerning what makes an effective product and what puts a business at risk. We are constantly balancing staying on the cutting edge with over-investing in products that introduce far more risk into our businesses than the problems they claim to solve.

When we consider any new AI product, we always ask five simple questions to decipher the difference between an AI product that can transform a business and an illusion wrapped in trendy marketing language. These five questions can serve as a good litmus test for any leaders responsible for AI investments.

1.How will my data be used, and what safeguards are in place?

You need to have a clear and documented explanation for how an AI vendor stores, repackages, and otherwise uses your inputs into their products. Vendors can use those inputs for other purposes like training their model, research and development, marketing, or other means that may not even be known when you contract with them. Without proper security and segmentation, your confidential or personal identifiable information (PII) can be exposed to outside entities and result in you losing control over your data and exposing you to potential legal and regulatory liability.

Three essential safeguards that must be incorporated into every enterprise contract include legal protections, audit capabilities, and regulatory compliance. Legal protections offer clear terms for data handling and dispute resolution to enforce the appropriate management of customer information. The ability to audit the vendor allows for transparency and oversight, confirming the contract is followed. Adherence to relevant regulations is crucial to ensure customers are not implicated in malpractice lawsuits. If a vendor is unable to articulate how they will meet these requirements, their product is not ready for deployment in an enterprise environment.

2. What is the model that is being used, and what data is it trained on?

Many vendors are rebranding conventional or off-the-shelf solutions as “artificial intelligence” to capture attention and engage with the prevailing trend. Approaching these claims with skepticism is imperative to integrate the right technologies into an organization.

A more thorough investigation into the foundational model should uncover whether it was internally developed, sourced from a major provider like OpenAI, Google, or Anthropic, and the extent of specialized fine-tuning that has been applied. This deeper comprehension allows for a more informed understanding of the tradeoffs between different vendors.

The phrase “garbage in, garbage out” emphasizes that a model’s performance is directly linked to the underlying training data. Biased inputs, uniform outputs, and the potential for model collapse (an AI echo chamber where outputs become inputs and novel value is lost) must be thoroughly evaluated when assessing the quality and reliability of AI products. Training data should be representative in an attempt to avoid underlying biases being reflected in the model’s output. Similarly, data should be high quality and sourced from reputable suppliers to minimize the risk of copyright concerns and repetitive or inaccurate results.

3. How do you measure accuracy and make improvements?

Accurate and trustworthy outputs elevate an AI product from being a novel playground to an indispensable instrument. A clear understanding of how the vendor measures response accuracy, guards against hallucinations, and continuously updates their algorithms is crucial in evaluating the long-term reliability and effectiveness of their AI solution.

Product development is iterative. Reputable vendors are able to articulate their strategy for rolling out enhancements and the impact to users. This involves not only a transparent roadmap of future updates, but also a clear communication of how these changes will improve functionality, user experience, and overall performance. A vendor’s commitment to continuous improvement reflects their dedication to customer satisfaction and adaptability in an ever-evolving tech landscape. Organizations must think through the products they plan to implement and attempt to invest in a solution that scales alongside their changing needs and technological advancements.

4. How can I implement the product into my workflow and what level of support can I expect?

Even the most versatile models excel in certain tasks while falling short in others, emphasizing the need for task-specific evaluation. A clear understanding of the intended use case is crucial to determine if a product aligns with organizational requirements. Additionally, some products integrate with an existing tech suite through the use of APIs, while others have a more intuitive and user-friendly graphical interface. Both options have unique advantages allowing the intended use case and the users interacting with the tools to determine the correct offering.

Reputable vendors will offer support in the event of system malfunctions or inaccurate outputs, ensuring timely and effective resolutions to maintain optimal performance. This support must include dedicated account managers that understand your industry-specific requirements and are more responsive in addressing issues. The rapid evolution of AI products invariably means updates that can alter a tool’s functionality. A dedicated point of contact provides reassurance and a reliable resource for assistance.

5. How is this offering differentiated from others on the market?

New AI products are being launched daily. Vendors must display the value they deliver over competing products and show a clear path forward for maintaining their edge. Candidly asking how a vendor’s product differs from specific competitors can show how knowledgeable the vendor is about the domain they operate in. Press on questions around how AI will add tangible value over current processes to reveal whether the vendor is capitalizing on a competitive advantage or is merely riding the wave of the latest trend. Furthermore, inquiring about the human elements of an AI product (HITL) can help to gauge the safeguards in place to ensure oversight on accuracy, adaptability, and compliance.

The integration of AI products in an enterprise environment is a balancing act between harnessing cutting-edge technology for efficiency and ensuring the tools align with the organization’s existing infrastructure, ethical standards, and operational goals. Improving your understanding and skepticism is instrumental in seeing through the hype.

Contact Us

Get in touch to learn more about our products, services, & solutions.

Subscribe to the
Penta newsletter