The Top Reasons Why Your Adoption of AI Will Fail
We had the pleasure of meeting Numa and Ben from KUNGFU.AI during an Agile Tech Town Hall. We covered a ton of ground in discussing the top reasons why organizations fail when trying to adopt AI (Artificial Intelligence). Their insights shed light on the common pitfalls organizations faced when diving headfirst into AI adoption.
Here is a recap of our conversation:
The Pitfalls of Adopting Shiny AI Solutions without a Clear Use Case
Numa highlighted the common occurrence of organizations rushing to adopt AI without a clear understanding of the problem they were trying to solve. Many clients came to KungFuAI after being swayed by media hype or competitor activities, without evaluating whether AI was the best solution for their specific use case. Numa explained, “They were kind of working with the solution as opposed to working with a problem, and they were going backwards. Oftentimes, that just didn’t work.”
By focusing on a problem-driven approach, organizations could identify the most suitable solution, which might not necessarily have been AI. Numa emphasized the need for an “unclear definition of the problem they wanted to solve” and a thorough understanding of whether AI was the most efficient and cost-effective solution. Sometimes, simple rule-based systems could provide a more efficient alternative to complex AI solutions.
The Danger of Superficial Urgency
Ben Herndon drew a parallel between the early days of the internet and the then AI landscape, highlighting the danger of superficial urgency. As AI technologies gained widespread attention, there was a sense of urgency to jump on the bandwagon without a well-defined strategy or purpose. Ben explained, “There was an almost nameless and undefined urgency to have to do something… That often led down a path where you might spend a lot of time and money… It didn’t really deliver the results you expected.”
To avoid the pitfalls of superficial urgency, organizations had to take a step back and evaluate their motivations for adopting AI. Ben suggested asking fundamental questions about the business, such as expected impact metrics, revenue generation, cost reduction, and customer retention. This helped organizations align their AI initiatives with concrete business goals and ensured they were not pursuing AI for the sake of following trends.
The Need for Organizational Alignment
Ben emphasized that embracing AI required organizational alignment across various stakeholders, not just within the leadership team. Clients approaching KungFuAI had to demonstrate alignment throughout the organization, including boards, investors, and key employees. Ben stressed, “Everybody was along for this ride… It was a commitment and initiative that, even if you started in a part of the organization that didn’t necessarily directly impact a lot of other people, everybody was along for this ride.”
This insight highlighted the holistic nature of AI adoption, where success depended on widespread collaboration and cooperation. While small-scale automation initiatives might have required less organizational alignment, they often fell short of delivering significant industry-leading differentiation. True competitive advantage through AI emerged when organizations embraced it as an enterprise-wide commitment.
Navigating Data Challenges for Successful AI Implementation
Garbage in, Garbage out: Data Quality Matters:
Data quality played a crucial role in the success of AI models. Numa Dhamani stressed the direct impact of data quality on the quality of AI models, stating, “Garbage in, garbage out. The quality of your data directly affected the quality of your model.” Organizations often struggled with insufficient or unlabeled data, hindering the training of AI models and leading to disappointing outcomes. To overcome this challenge, Numa emphasized the need for organizations to invest time in understanding data sources, ensuring compliance with privacy laws, and organizing data through attribute profiling and metadata catalogs.
Addressing Data Challenges in AI Projects:
Early identification and resolution of data challenges were vital for the smooth implementation of AI projects. According to Ben Herndon, a significant portion of the project time was spent on data preparation, as he explained, “You actually spent most of your time trying to get the data in a format to put it in a machine learning model.” Unstructured data and lack of labeling posed significant hurdles in building effective AI models. To overcome these challenges, organizations needed to involve data engineering teams, data architects, or software engineers who could restructure and organize the data for optimal use in AI models.
Aligning Data Strategy with AI Objectives:
A forward-thinking data strategy was key to maximizing the effectiveness of AI initiatives. Ben Herndon emphasized the importance of considering future AI requirements when collecting and retaining data. He stated, “Companies needed to be thinking about their data strategy in terms of what they wanted to model in two to three years.” Organizations should proactively collect relevant data, even if it only covered a shorter timeframe. This approach ensured that the data collected aligned with the future AI objectives and enabled seamless model training and deployment.
Navigating the Transition from Model to Product in AI Development
One crucial aspect often overlooked during the transition from a model to a product was the need for a robust machine learning infrastructure. Numa emphasized that AI models required ongoing maintenance and updates. She noted that companies often lacked long-term planning and failed to recognize that AI models needed to be retrained and updated regularly. This was particularly important due to the phenomenon known as data drift, where models trained on historical data became ineffective when faced with new and evolving data. Numa provided an example of the COVID-19 pandemic, where models trained on historical pricing data were unable to account for market changes caused by the crisis. To address this challenge, companies had to establish efficient and reproducible methods for updating and retraining their models to ensure their ongoing accuracy and relevance.
Ethical Implications and Considerations
Before deploying an AI model as a product, it was essential to consider the potential ethical implications it might have. Ben Herndon stressed the need to examine whether the model introduced inadvertent biases in its predictions, recommendations, or categorizations. Even if the model was unbiased, it might inadvertently cause harm through the way it provided feedback or directions to users. Ben highlighted the importance of addressing these ethical concerns, as they could have cataclysmic consequences for both individuals and businesses. This necessitated a proactive approach in identifying and mitigating potential biases and harmful effects.
Impact on Business and User Experience
Transitioning from a model to a product involved more than just technical considerations. Ben Herndon emphasized the need to assess the impact on various aspects of the business, particularly the user experience (UX). He explained that AI models often produced outputs with a certain level of confidence, which users might find challenging to interpret. The probabilistic nature of AI models could create confusion and uncertainty for users if not properly addressed in the UX design. Additionally, organizations had to ensure that customer success groups were equipped to handle AI-related issues or concerns effectively. Neglecting these aspects could lead to negative customer experiences and undermine the overall success of AI initiatives.
Challenges in Realizing Measurable Business Value from AI Investments
During the conversation, Clint raised the question of the challenges enterprises faced in realizing measurable business value from their AI investments. Numa Dhamani pointed out that one crucial hurdle lay in identifying the right problem to solve. Organizations had to define the problem they aimed to address and establish key performance indicators (KPIs) that indicated success.
Numa further highlighted the associated costs involved in AI implementation. These costs included not only infrastructure and model building but also employee training and ongoing learning. In addition, managing models in production presented a significant challenge, often requiring dedicated resources and infrastructure. Numa emphasized the need for organizations to allocate resources and support employee development to ensure successful AI adoption.
How Do We Keep Learning About Emerging AI Trends?
Credible Research Organizations as Trusted Sources
In addition to podcasts, Numa highlighted the significance of credible research organizations such as Stanford and MIT. These institutions actively contributed to the field of AI and regularly published peer-reviewed research. Numa stated that she considered these organizations her go-to sources for information on emerging technologies. By relying on research institutions, professionals could access reliable and trustworthy information, ensuring they stayed up to date with the latest AI trends.
Building a Community of Experts
Ben Herndon emphasized the importance of building a community of peers and experts to stay ahead in AI. He discussed the significance of collaborating with others, including former academic colleagues, mentors, and industry contacts. Ben stressed the need for ongoing conversations and knowledge sharing to navigate the rapidly evolving AI landscape effectively. By engaging with a community of experts, professionals could stay informed about emerging trends, exchange ideas, and collectively solve challenges.
