Practical Business Reasons to Resist the Allure of AI
xkcd.com
Using AI and machine learning to meet business needs has great potential business and customer value. AI can help you deliver value to your customers effectively, drive efficiencies in your operations, and even explore new avenues for growth. The potential for significant returns on investment (ROI) is high, making AI and ML attractive options for many organizations.
However, the key to realizing the benefits of AI/ML is to ensure you have the right drivers and vision. It’s not just about jumping on the latest tech trend; it’s about strategically aligning AI and ML initiatives with clear business objectives and customer needs. When implemented thoughtfully, AI and ML can transform businesses, delivering insights and capabilities. Without a solid foundation and purpose, however, even the most advanced AI and ML technologies can fall short. It’s crucial to understand the true value and applications of AI and ML, ensuring that their efforts are guided by strategic goals and a deep understanding of their potential impact.
In this blog post, we will explore flawed drivers for adopting AI/ML solutions. We’ll discuss the motivations for each and alternatives to try first.
Following the Hype
It's common for companies to adopt new technologies or trends simply because they are popular and shiny rather than based on a clear understanding of their benefits or strategic alignment with existing business goals. This often leads to misguided efforts, wasted resources, and unmet expectations.
Why It's Bad: Jumping on the AI/ML bandwagon without clear goals, a strategic approach for AI, and an understanding of its implications will lead to wasted time spent by expensive staff and investments in expensive ML technologies on projects that don’t deliver value. When companies adopt AI/ML based on an article like the Gartner Report espousing generative AI as the new “holy grail,” business value takes a back seat to the shiny new technology. The lack of a clearly defined value for the investment in developing AI/ML leads to floundering projects that meander in search of a solution.
What to Do Instead: Develop a Clear AI Strategy. Create a reusable framework that starts with assessing business needs and objectives. The framework must provide an organized value-based approach for developing machine learning models. Create an outcome-based roadmap for each effort, including goals, resources, and measurable outcomes. This approach ensures that AI/ML initiatives are purpose-driven and aligned with your company's vision.
Example: IBM Watson for Oncology aimed to transform cancer treatment with AI-driven recommendations but faced technical and integration issues. It was a laudable goal, but it was driven by a need to implement AI at the height of the Gartner hype cycle in 2012. The project struggled with providing accurate suggestions, aligning with medical practices, and managing high expectations. Financial losses, like MD Anderson Cancer Center's $62 million, and low trust among doctors led to its failure. IBM had to scale back its ambitions, focusing on administrative tasks. This case highlights the risks of AI hype and the need for realistic expectations and practical integration.
Impressing Stakeholders
Impressing stakeholders involves adopting AI/ML to appear innovative or cutting-edge rather than for strategic business reasons. It focuses on gaining short-term approval or excitement from investors, executives, or clients without a clear plan for practical implementation or long-term benefits.
Why It's Bad: This is very similar to “following the hype”. In some ways, this driver is worse than following the hype because the business is implementing AI/ML for the sake of appearance, with the hope that it can deliver tangible results. Showcasing cutting-edge technology might initially impress investors and stakeholders, but failing to produce measurable outcomes can damage credibility and trust. Stakeholders are more interested in the actual performance, efficiency gains, and ROI that AI/ML can bring. If a project is pursued without a clear understanding of its potential value and impact, it risks becoming a costly showcase with no real benefits. Genuine progress and value creation should take precedence over superficial technological displays.
What to Do Instead: Focus on Delivering Real Value. Hope is not a plan. The business leadership needs to engage stakeholders with transparent communication about the potential and limitations of AI/ML. Set realistic expectations and focus on projects that have clear, measurable benefits. Demonstrate progress through pilot projects and incremental improvements that showcase tangible value.
Example: Quibi, a short-form streaming platform, launched in April 2020 to revolutionize mobile content. The approach was to sell the idea to high-profile investors. Despite high-profile investors (and an investment of $1.75 billion) and talent, it failed due to market misreading—launching during the COVID-19 pandemic when viewers preferred long-form content on larger screens. Quibi's unique format and lack of essential features like screenshot sharing didn't resonate with users. Poor marketing and a complex subscription model further hindered its success. Unable to attract a substantial subscriber base, Quibi shut down in December 2020, just six months after launch, resulting in significant financial losses for investors.
Over-automation
Over-automation as a driver for adopting AI/ML involves excessively applying these technologies to automate every possible process. This approach prioritizes automation over practical needs and human judgment, often leading to unnecessary complexity and overlooking areas where human intervention is more effective.
Why It’s Bad: This undermines human expertise, overlooking the intricate nuances of business processes. Relying solely on machine learning to replace human judgment risks oversimplification and ignores the value of human insight. It can lead to errors, decreased adaptability, and detachment from customer needs, hindering long-term business success. Without defining the specific instances where learned automation can support the business and reduce cost/increase revenue, the effort will not be worth the investment.
What to Do Instead: Augment Human Capabilities. Use AI to enhance employee productivity rather than replace them. Identify repetitive and mundane tasks that machine learning models can handle, allowing employees to focus on strategic, creative, and customer-facing activities. This approach boosts morale and maximizes the combined strengths of human and machine intelligence.
Example: Amazon developed an AI-driven hiring tool in 2014 to automate recruitment. The goal was to reduce/eliminate reliance on people to review and assess resumes. The tool was trained on available resumes, which were primarily made up of male candidates. The results were models that penalized women, leading to inaccurate rankings. The AI's lack of transparency and consistent performance made it reliable. By 2018, Amazon abandoned the project due to its inability to fairly and accurately evaluate candidates. The failure highlighted over-automation limitations, emphasizing the need for diverse training data, human oversight, and transparent AI decision-making in complex tasks like hiring.
Misunderstand Limitations of AI/ML
Misunderstanding the limits of AI/ML as a driver involves overestimating these technologies' capabilities and expecting them to solve all problems. This includes assuming AI/ML can function without sufficient data, handle any task without human oversight, or deliver flawless results, leading to unrealistic expectations and misguided projects. This is sometimes described as “sprinkling” the magic AI dust onto a problem to solve it.
Why It's Bad: AI has its limitations and is not a silver bullet. Overestimating what AI/ML can achieve can lead to unrealistic expectations and underperforming project delivery. Systems powered by machine learning models are typically designed to perform specific tasks and may not adapt well to unforeseen challenges. Deploying without understanding its constraints can result in poor decision-making and operational inefficiencies. Moreover, reliance on ML without appropriate human intervention can lead to errors and missed opportunities for nuanced problem-solving. Companies must recognize the specific strengths and weaknesses of AI, using it as a tool to complement, not replace, human judgment and expertise.
What to Do Instead: Understand and Leverage AI's Strengths. Establish a formal framework for investigating solutions where an AI/ML solution is appropriate. This framework should focus on understanding the desired business value first, and then the data needs before iterating on the development of machine learning models.
It is common to establish these frameworks via pilot projects to understand what can be accomplished using ML. In most cases using AI/ML for tasks it excels at, such as data analysis and pattern recognition, while relying on human judgment for complex and nuanced decision-making is a winning combination.
Example: In 2016, Microsoft launched Tay, an AI chatbot on Twitter intended to showcase advanced conversational abilities by learning from interactions. However, Tay's vulnerability to manipulation and lack of content moderation led users to teach it offensive content quickly. Within 24 hours, Tay began posting racist and inflammatory tweets, forcing Microsoft to shut it down. This failure highlighted Microsoft's misunderstanding of AI's limitations, emphasizing the need for robust safeguards, ethical considerations, and continuous monitoring of AI systems to prevent misuse and ensure responsible deployment.
Keeping Up with Competitors
Keeping up with competitors as a driver involves adopting AI/ML solely because other companies are doing so. This approach is driven by fear of falling behind rather than a strategic understanding of how these technologies can benefit the business, leading to rushed and potentially misaligned implementations.
Why It's Bad: Many enterprises feel that they need to stay competitive by implementing solutions based on their competitors' sales materials. These days, it seems like every solution has a tagline of “powered by AI.” Each company has unique processes, goals, and challenges, and what works for one may not work for another. Blindly implementing machine learning is expensive and will lead to ineffective solutions and wasted resources.
What to Do Instead: Evaluate Internal Needs. Conduct a thorough internal assessment to identify areas where AI/ML could provide genuine benefits. Focus on addressing your unique challenges and opportunities rather than simply mimicking competitors. Tailor your AI/ML strategy to fit your specific business context and goals.
Example: Uber's self-driving car project, launched in 2016, aimed to compete with rivals like Waymo and Tesla but faced significant challenges. Safety issues, including a fatal crash in 2018, exposed flaws in the technology and testing procedures. Technical limitations and regulatory hurdles further impeded progress. Public trust and regulatory relationships were damaged, leading Uber to suspend the project and sell its autonomous vehicle division to Aurora Innovation in 2020. The failure highlighted the risks of rushing to compete in complex, safety-critical fields without thoroughly addressing technological, safety, and regulatory requirements.
Lack of a Data Platform
The lack of a data platform as a driver involves adopting AI/ML technologies without having a robust infrastructure for data management. This includes insufficient data storage, processing capabilities, or integration systems, leading to challenges in effectively utilizing AI/ML for meaningful insights and outcomes.
Why It’s Bad: The adage “garbage in, garbage out” applies doubly when implementing machine learning models. Data is at the core of AI/ML. When models are trained, they create relationships between and within the data it is provided. If the data is biased, incomplete, or noisy, it will create models that provide bad, if not harmful, inferences. Bad data quality is usually a byproduct of data silos in the business, poor data hygiene within applications, and a lack of governance around how data should be managed.
Attempting to implement AI/ML with a common process to consume the data is, at best, difficult. At worst, it will produce a harmful machine-learning model that needs more transparency and can be repeatable, developed, or built.
What to Do Instead: Implement a Data Mesh. Data Mesh is the new hotness in the data realm. It has many benefits over more traditional data warehouses. Key benefits are that data is treated as a product, the products are decentralized and searchable, and quality and governance are embedded into the data product.
Data meshes enable data sciences teams to consume quality data more easily, which leads to quality AI/ML products.
Example: Theranos, founded in 2003, aimed to revolutionize blood testing with AI-driven technology but failed due to inadequate data infrastructure. The company struggled with data accuracy and reliability, exacerbated by a poor data platform and secretive practices. This lack of robust data management led to unreliable test results, regulatory scrutiny, and the company's eventual shutdown in 2018. Founder Elizabeth Holmes faced fraud charges. The failure highlighted the crucial need for a solid data platform and transparency in AI/ML projects to ensure accurate results and compliance with regulatory standards.
Conclusion
We don’t want to discourage you from embracing Machine Learning. Machine Learning has many novel and beneficial uses, from making better beer to helping track marine life. However, the reasons MUST be in support of your business goals and values.
