Setting data up for AI success with Coveo and Organon

Generative AI (GenAI)—though ambitious in its goals—is nothing without the extensive data work that must be done before, during, and beyond implementation. Data cleanliness is a critical enabler for GenAI technology, where accurate, secure AI-generated responses can become a tangible reality.

For those struggling with data quality initiatives in the quest to pursue GenAI, experts joined KMWorld’s webinar, Clean Data, Smarter AI: The Impacts of KM on GenAI Effectiveness, offering strategies, techniques, and insights for optimizing knowledge repositories to best support GenAI projects.

The surge in popularity of GenAI has led to a consistent sentiment among enterprises: Will this investment deliver tangible value? And, with more than 50% of enterprises projected to abandon their AI investments by 2028 due to costs, complexity, and technical debt, according to Gartner, the push for results is high.

“There’s a lot of questions now…around whether GenAI is going to live up to its promises,” explained Juanita Olguin, senior director of product marketing, platform solutions at Coveo. “There’s a call for customers and enterprises—rightfully so—asking vendors to show them…that GenAI works.”

Unfortunately, there is no “silver bullet” that will launch GenAI initiatives to success, noted Olguin. Instead, “it’s a series of components and elements that work hand-in-hand, together,” she added, which include:

• AI expertise in complex systems and scientific testing
• Deep search relevance to filter, factor, and identify top results
• Ability to innovate quickly, what you build today won’t be useful tomorrow
• Enterprise scalability, moving POC to production across the organization
• Agnostic approach that allows the AI model to work with any UI, platform, and use case
• Secure connectivity with integrations and connectors to popular applications
• Self-learning and closed loops, where various AI models work in tandem to improve the overall system, continuously
• Analytics and KPIs to measure the success and performance of your investment

Olguin introduced Coveo, a company working since 2005 to solve the hardest and toughest content management and knowledge discovery problems. With over 30 live generative answering deployments with leading enterprises today, Coveo’s single managed AI search platform helps enterprises implement semantic search, AI recommendations, GenAI answering, and unified personalization.

The webinar’s practitioner spotlight featured Daniel Z Shapiro, associate director, enterprise knowledge and content management at Organon, and the way Organon has delivered effective enterprise search with GenAI in the women’s healthcare sector.

To implement generative answering, Organon selected Coveo due to its strong integration with ServiceNow, its merging AI capabilities, and the ability to interface with many of Organon’s existing systems, explained Shapiro. This journey toward GenAI adoption surfaced crucial lessons learned and best practices, which boiled down to two main takeaways: GenAI relies on good content and how to handle stale content and conditionality.

Generative answering requires a foundation, noted Shapiro, which is defined by proper content stewardship. Ensuring users take accountability for their content by setting expectations for ownership is a crucial basis toward enabling effective GenAI. Organon further learned that any wrong answers will quickly erode users’ trust in your search solution, and enterprises should think critically about where there is any content on a site that should be excluded from search.

Organon’s GenAI adoption also revealed that, despite its potential, generative answering undoubtedly has weaknesses. For instance, policy and HR content has a significant amount of conditionality, and current GenAI solutions fail to understand when there are many possible correct answers. Furthermore, stale content will train the model just the same as fresh content—the only way to avoid this is not to train on any stale content at all, explained Shapiro.

Ultimately, generative answering and enterprise search is a mirror that will uncover issues in your content management—especially regarding data cleanliness. Deciding how to train the model—and with what content—is critical, as this is one of the only ways to control the model’s output. Shapiro advised that enterprises should implement effective stewardship policies, as well as consider technology to identify and clean proprietary content.

For the full, in-depth webinar, you can view an archived version here.