Before you add AI, consider these three fundamentals

“AI adoption didn’t make sense for my product… and so I quit.” 

The product manager who told me this story is not alone. Product leaders worldwide are facing pressure from their executive teams and their market to incorporate artificial intelligence (AI) tech into their existing products. Often the executives are making these requests without sufficient investment into understanding the impacts to the business. 

This lack of consideration for existing roadmaps can be discouraging to product leaders, but worse is when the request is poorly aligned with the financial health and risk appetite of the organization. 

In the following article, we’ll examine the things you, your product and your product team need to be successful incorporating AI. 

Factors to consider

The promise of AI is thrilling. The promise of a more natural user experience that allows users to achieve their goals more quickly is attractive to customers and executives. And can sound like a magic bullet to any product manager. 

However, the work to understand that the costs and risks associated with the adoption of AI can be sobering. 

The following are three foundational elements that you should explore to find out if AI adoption is really in your product and company’s interest: 

• Operational Costs. A significant increase in operational costs come with the increased computational demand of AI. Evaluating these cost increases are critical to understanding product profitability. 
• Staffing Costs. AI expertise is scarce, in high demand and expensive. Evaluating the pros and cons of hiring expensive staff and building expertise in existing staff is critical in ensuring a profitable business.
• Business Risks. Like any new technology, AI comes with risks that pose significant reputational risk to the business. Examining the risk areas and ensuring your company has the tolerance for these risk is critical to protecting your company’s reputation. 

Artificial intelligence technologies are more expensive to run

For the product managers who are responsible for generating revenue (and profit!) understanding the impact that the compute-hungry nature of AI will have on bottom lines is critical to understanding the technology’s profitability. 

Increased cloud operations costs

There exists a real risk that introducing AI into an existing cloud offering may make your existing solution unprofitable. With researchers estimating that the amount of compute required to conduct a Google search with AI is “23–30 times the energy of a normal search”, you should consider the impact that a 10x increase in operational cost will have on your business.

Prior to investing in a solution expected to run in the cloud, product leaders should invest in understanding the impact on their operational costs. A variety of techniques exist to provide estimates of the increase that changes to solutions exist:

• Establishing core scenarios and baselining existing operational costs
• Invest in rapid prototypes that incorporate AI and can be used to evaluate changes to operational costs 
• Investing in your team’s understanding of various AI models and how they apply to your existing product 

You can expect these activities to take some time as your technical team experiments to find the right fit for your purpose. Success is clear: a solution that keeps your company profitable, your executives happy and your customers efficient. 

Environmental impact

As compute increases with AI adoption, so does the impact that your software has on the environment. Ensuring that you understand where your solution is hosted, how your compute consumption impacts your environmental footprint – including carbon emissions and wastewater - and what type of energy is powering the solution is key to understanding your environmental impact.

Here are three things to consider when evaluating environmental impact:

• Carbon footprint. Ensure you have a good understanding of the carbon emissions that the energy your solution produces. A clear understanding of the emissions you’re passing onto your customers (scope 3) can help you positively differentiate yourself in a highly competitive marketplace. 
• Wastewater produced. Most data centers use water to absorb the heat produced by computing device. This water either evaporates or is discharged to wastewater filtration facilities. Either way, the wastewater leaves the data center loaded with chemicals and heavy metals from the equipment it helped cool and reduces the water available in the local environment.
• “By 2027, global AI demand is expected to account for 1.1 to 1.7 trillion gallons (4.2 to 6.6 billion cubic metres) of water withdrawal”. Reference: World Economic Forum. Link.
• “With larger and new AI-focused data centers, water consumption is increasing alongside energy usage and carbon emissions.” Reference: Environmental and Energy Study Institute. Link.
• E-waste. With the frantic adoption of AI, data centers worldwide have begun massive projects to upgrade existing technologies. More efficient software slows the demand for upgrades and slows the production of associated e-waste. 
• “Generative AI could account for up to 5 million metric tons of e-waste by 2030.” Reference: MIT. Link  

Artificial intelligence requires technical expertise that is scarce

Setting aside the financial and environmental costs of incorporating AI, developing software that rationally and efficiently incorporates AI can be challenging. Just the research required to build your own understanding, as a product leader, of how your customers are most likely to benefit will take a significant investment of your time.

The real challenge is finding the expertise to realize your vision with quality and at scale. Researching, developing and training AI models can be a very expensive proposition from both a human and compute resource perspective.  A team that lacks experience can make architectural choices that waste development cycles and consume large amount of compute resources inefficiently. 

The following are the key capabilities your technical team needs to incorporate AI into an existing product line: 

• Ability to assess theoretical AI models for practicality against product goals
• Ability to plan, model, manipulate and migrate data for training
• Support verification process with robust technical scenarios at scale
• Establish governance practices that protect the data model from being stolen, corrupted or consuming information which exposes the company to legal and ethical risks

Build or hire, or buy

Once you’ve identified the skills you need to build these capabilities, you need to make the business case about building or buying the expertise. 

Build

With this option existing team members are given the opportunity to grow into AI experts. They’ll study models, approaches and the theory to help build products with AI.

• Pros: Your team is already familiar with your existing product and data.
• Cons: Your team can be easily distracted by day-to-day product escalations. It takes significant time to build knowledge and can require more iterations of trial and error.
• Cost: May take several years of time and salary.

Hire

With this option you recruit a leader or even an entire team who can integrate AI into your product

• Pros: Engineers with previous experience have expertise that can be used immediately.
• Cons: Salary costs for this skillset are high. The new team members will probably lack experience with the product’s domain.
• Cost: AI engineers are scarce and expensive, with some senior AI engineers sharing that they are paid ~$300k a year. (Reference: Levels.fyi. Link.)  

Buy

With this option, you contract a third party company that specializes in building AI models or modules that integrate with products like yours. 

• Pros: Can accelerate delivery
• Cons: May be difficult to perform care and feeding on solution, which could increase ongoing reliance on 3rd party and increased costs over time. 
• Cost: Varies significantly based on application. Simple models can cost $5,000 while more sophisticated models cost $500,000. (Reference: Future Processing. Link. )

Artificial intelligence poses new, and sometimes unexplored risks to organizations.  

After giving due consideration for the costs of adopting AI, another area to consider is the more qualitative business implications. A traditional exercise in risk management can help you identify how the new technology could positively or negatively impact your business or product’s reputation. 

Areas of risk that AI presents should be evaluated, just like any other risk, by the likelihood of occurring and the impact on the business if the risk does materialise.

Organizational risks and real examples 

Here are some areas of risk to consider when introducing AI to your product. As well, I’ve included some examples of how these risks have realized for specific companies. 

• Data breaches. As AI has grown in usage, well-intentioned users have been known to leak sensitive information through a lack of understanding or poorly governed applications. 
• Risk question: What sensitive information in your product may more easily leak due to your AI adoption?
• Ex: Samsung data leak via ChatGPT. Source: Bloomberg. Link
• Blind trust. Over time, some users begin to trust AI-enabled tools without question, resulting in errors.
• Risk question: If your users begin to blindly trust your application, what is the impact of a mistake to the user or their company on the part of the product?
• Google lose $1B value after providing incorrect info. Source: Reuters. Link
• Robustness of ecosystem. Your product may be ready for AI, but products it relies on may not. Without proper technical and logical controls it place, your product could have negative effects on other systems.
• Risk question: Are the systems that your product interacts with robust enough to handle disruptions that your AI could introduce?
• Ex: Zillow writes down $304M after AI unintentional purchases homes at higher prices. Source: CNN. Link 
• Legal & Regulatory. Your business is subject to additional legal and regulatory requirements when using AI. Ensuring your product complies fully with the regulations in the regions where you do business can help avoid fines and lawsuits.
• Risk question: What laws are you subject to when training or delivering AI solutions? 
• Ex: Italy’s data protection agency fined Chat GPT 15 million euros for breach of EU AI Act. Source: Reuters. Link.
• Ethical. The data that AI is trained on influences how it responds and, if trained incorrectly, these technologies can produce biased and unethical content, information and actions. 
• Risk question: What are the bias and discriminatory impacts that your product could experience when incorporating AI? 
• Ex: iTutor Group auto-rejects female candidates. Source: Greenbert Traurig Law. Link.
• Bad actors. Just like any new technology, AI presents new attack vectors that people with bad intentions can exploit.
• Risk question: What are the new attack surfaces in the product that incorporating AI creates?
• Chevrolet dealer offers car for $1. Source: Twitter.

Stay sane in the face of executive appetite for AI

Any change in roadmap direction will cause delays, frustrate your team and result in customer disappointment. While shouldering these impacts is a key responsibility of a product leader, ensuring that you understand the practicality of adopting AI for your product is a serious responsibility.

Properly evaluating whether you can be successful in incorporating AI can help you make the decision to either:

• Jump on the AI bandwagon wholeheartedly 
• Begin educating your executives on the impracticalities of AI for your product
• Start looking for a role where AI adoption makes sense