The AI Precipice: A Look at Tangible Success and Realistic Failures at Data Summit 2025’s Day 2 Keynotes

May 15, 2025

By Sydney Blanchard

AI’s domination of the business world speaks to its capacity for transformation—both positively and negatively. Without proper preparation and planning, AI initiatives will fail to deliver.

During Data Summit's Day 2 keynotes, Andreas Welsch, founder and chief AI strategist, Intelligence Briefing, and Ebrahim Alareqi, principal machine learning engineer, Incorta, ruminated on the ways AI can be successfully implemented for practical, tangible success.

The annual Data Summit conference returned to Boston, May 14-15, 2025, with pre-conference workshops on May 13.

“If you follow the news, there’s always something new,” began Welsch. “So what really matters?”

While many execs believe their companies are AI-ready—and are itching to do “something with AI”—little know what to actually do. “But leaders feel the pressure…[they think,] ‘Most likely, our competitors—they’re not sleeping.’”

The push for AI adoption may be more harmful than helpful, however. Asking if AI is simply available for use is not the right question to pose, noted Welsch. An examination of current challenges, and how businesses can be adapted, is more beneficial. More awareness about AI and how it works, its potential—and more importantly, its realities—is the first step toward AI-readiness.

Welsch emphasized the need greater education around AI, what it is, and what it’s best for—instead of blindly hurtling toward a fuzzy “AI-first” idea. Readiness, according to Welsch, is what propels organizations toward being AI-first.

Though promising a wealth of opportunity, it is difficult to ignore the daunting realities of AI, where 80-85% of AI projects fail. It’s rarely because of the technology, according to Welsch, but because of the people—perhaps a lack of learning, inconsistent processes, or poor change management. Getting the people of your organization ready for AI is a large piece of AI success, according to Welsch.

AI-readiness, then, is a leadership task, where execs should own AI strategies and drive literacy and empowerment among their workers. Allocating space for experimentation with AI tools and encouraging feedback are key ways in which leaders can drive AI-readiness from a people perspective—not simply offloading a new AI tool with little follow-up.

Furthermore, AI literacy acknowledges “all the ways AI can help us and all the ways we need to be aware that AI can hurt us,” said Welsch. Understanding the power of hallucinations, of believable—but false—information persisted by AI models, is central toward realistic AI utilization. Encouraging more critical thinking about AI, instead of blind adoption and faith, is crucial.

There is a disconnect among leaders and workers, however, according to a report highlighting how 46% of workers are afraid to tell their managers that they use AI in their workflows. Whether due to perceptions of incompetence or fears of an increased workload, a lack of encouragement creates more dissonance between the possibilities of AI and the workforce.

Part of correcting this gap is achieved by imagining AI as products, rather than projects. Projects, which have a start and end date, do not suit the complexity of AI. Viewing them as products with cyclical lifecycles that center the user experience enables better outcomes.

From a more technical perspective, Alareqi began his section of the keynote by examining the major generative AI (GenAI) use cases in production today, which include:

Code generation
Chatbots and virtual agents
Auto-generation of visualizations and dashboards
Apps, automation, and workflows

While GenAI certainly posed value for these use cases, only a mere 10% of GenAI projects are in production. Alareqi explained that this is due to:

Lack of fresh, detailed, and trusted data
Difficulty in deploying AI tools, often requiring the right technical skills, data collection and processing, validation and testing, ongoing maintenance and support, and more
Lack of security across LLM models, internal data access, and other unknowns

Incorta helps alleviate these challenges with live, detailed data, fixed costs, a fully managed, open infrastructure, and private architecture. Outside of platforms, Alareqi explained that when developing an AI strategy, enterprises should consider: