Learning about Taxonomy, Enterprise Search, Enterprise AI, and Text Analytics at KMWorld 2026
The KMWorld 2026 conference has four additional co-located events: Taxonomy Boot Camp, Enterprise Search & Discovery, Enterprise AI World, and Text Analytics Forum. The following is a look at some of their sessions.
Taxonomy Boot Camp
The theme of Taxonomy Boot Camp was Practical Taxonomy for Search, AI, and Findability. Sam Stringer, information architect at Factor, presented Domain Modeling as a Foundation for Taxonomy. She discussed a project she was involved in to create new information architecture for a cloud computing company’s website. The site’s landscape was a mess, Stringer noted. Her team needed to create a unified taxonomy. That’s where domain modeling came in, which Stringer defines as a representation of information objects and the relationships between them. A domain is a conceptual subject area, and relationships should be universal within a domain. For the website project, she and her team created and revised domain models until everyone had a shared understanding.
In Creating a Human-Driven, AI-Assisted Source of Truth for Taxonomy Definitions, Jennifer Kaari, catalog manager at OpenSesame, described how she and her team needed to quickly connect a taxonomy of 4,000 skills. But their custom GPTs inconsistently used definitions, and this would undermine confidence in the company’s catalog, Kaari said. Her team came up with a four-step workflow: 1) Prioritize terms and build a template, 2) Ground definitions in real usage with a Custom GPT, 3) [Implement h]uman review and approval, and 4) Differentiate similar terms. She stressed the importance of having documentation, which is something she had neglected previously.
Taxonomies for Humans and Machines: Language, Politics, and AI featured Joseph Busch, principal analyst at Taxonomy Strategies, who noted that there’s a role for taxonomists in constructing prompts using natural language. He provided a brief history of information retrieval and discussed what prompts should be: precise, free of jargon, etc. Busch reviewed the different types of mapping and noted that LLMs help identify gaps. During the same session, Ahren Lehnert, senior taxonomist at Genentech, presented Taxonomies, Memeches, and the Politics of Language. He said taxonomies are not inherently political—but contextually, they can become so when the term is something like “African Americans.” Taxonomies are rigid hierarchies, he noted, and in society, hierarchies are power structures. Lehnert compared a societal regression to the mean to LLMs’ resistance to innovative ideas. Taxonomists are uniquely positioned to help LLMs improve, he said, by, for example, using diverse data sources.
Enterprise Search & Discovery
The Enterprise Search & Discovery theme was Making AI-Powered Search Real for the Enterprise. Agentic AI, AI-Ready Documents, and Search: Real-World Applications featured Randall Smith, chief revenue officer of Forefront Technologies, and Jeff Fried, director of platform strategy and innovation at InterSystems. Smith spoke first, with his talk Corralling the Silos. He said we’re in a Wild West of enterprise information, with it spread across organizations, and “There’s gold in those silos!” InterSystems tells its customers that getting AI-ready is a journey; Smith provided some examples from companies he works with and shared the milestones on the trail (to continue his Western metaphor) of AI readiness of information: chaos, discover, clean, understand, protect, govern, and modernize. He introduced the questions necessary to ask at the beginning, such as, What is redundant or obsolete? Is sensitive data where it shouldn’t be? Organizations need to survey their landscape, connecting repositories and searching across them, so they can better use their data.
Fried, who participated via live stream, presented Agentic AI and Search: Real-World Applications That Deliver Results. His metaphor was AI as “an amazing roller coaster ride.” He shared experiences from InterSystems customers and went over common optimization areas, such as enriching data for ingestion. Text-based embeddings have had breakthroughs with AI, in addition to non-text embeddings, he said. The latter is a crucial area for AI. KM is coming into its own with context engineering, Fried believes, so understanding the structure of the knowledge is key.
Enterprise AI World
The theme for Enterprise AI World was Driving Real AI Adoption Across the Enterprise. AI-Driven Promotion Platforms: Increasing Customer Engagement was presented by Shubhatha, a software engineer at Northwest Bank. He said that where traditional marketing falls short, AI can close the gap, and shared statistics on how AI use in marketing is growing. AI can offer benefits: real-time personalization, scalable segmentation for more granularity, automated knowledge creation, and closed-loop improvement from customer responses. The building blocks of integrating AI into marketing include decision files, a customer-facing online company portal, and AI software that monitors customer actions and feedback. There are considerations for responsible AI adoption, such as data privacy, accountability, and human oversight. His key takeaways are: Traditional promotions are no longer enough, AI enables true personalization at scale, a closed-loop platform is required, and ethics determine long-term success.
Whitney McDonald, senior manager of KM, and Karen Romano, senior director of customer content, both from Okta, lamented fewer content views because of zero-click search in From Clicks to Citations: The Human-AI Partnership Fueling the “Atomic Brain.” Their solution to this problem is the operating model they call the Atomic Brain, which is getting results; Okta now has robust, AI-ready content that agents are using. The pillars of this model include a foundation (Okta had a strong KM program already, with content standards, but they had to adjust their writing to be more readable by AI crawlers), optimization (McDonald and Romano explained other processes they had to implement behind the scenes; there’s no one-size-fits-all approach), and AI scaling (they built an AI agent ecosystem so that the humans could focus on QA and the AI could find content gaps that humans couldn’t).
McDonald and Romano had five guiding principles: Convert editorial standards into system instructions; target repeatable tasks by automating first drafts and routine style edits; identify content gaps, including hidden user friction points; start small by deploying simple agents, and then iterate rapidly; and keep the human in the loop who validate accuracy and safeguard confidential data. They provided examples of the evolution of their AI ecosystem and reiterated the impact and benefits of their Atomic Brain endeavor.
Text Analytics Forum
The theme for Text Analytics Forum was Turning Text Into Actionable Insights With AI and Analytics. The AI, Text Analytics, and Taxonomies session featured two talks: Governing Known and Novel Entities to Reduce Hallucinations in Gen AI by Lauren Clark Hill, expert solutions engineer at Squirro, and How to Augment Taxonomy Change Management With Statistics by Kyle Garcia, technical solutions lead, and Urmi Majumder, chief architect, both of Enterprise Knowledge, LLC. Hill shared how to use your skills to prevent GraphRAG hallucinations (i.e., resolution errors, in which the model faithfully reports a graph that merged, split, or mistyped your concepts). When you review a hallucination, you don’t see all of the steps that went wrong, she said. She introduced label-based resolution, which consists of the collapse concept and the fragmentation concept. Small inconsistencies compound at graph scale, so there has to be a level of review. She explored the example of a chatbot with two different nodes that can cause confusion for an end user. We need to draw a boundary between known and novel entities, she said. The risk isn’t with novel entities; it’s resolving them with the same rules and trust as known ones. Novel entities won’t merge to known entities without human review, so you can build the boundary into a workflow. Hill shared some examples of that, noting that GraphRAG needs to be governed to ensure more accuracy.
Garcia and Majumder said the challenge they face is that enterprise taxonomy doesn’t account for a new domain terminology. How can organizations track changes in business terminology and identify what needs to be changed within their enterprise taxonomy? they asked. The problem is compounded at scale because of small taxonomy teams and the way generative AI has increased the speed of research. Garcia and Majumder found the solution to be bottom-up (statistical analysis to detect less-perceptible changes at a data level), top-down (human-driven validation of taxonomy statistics to verify their veracity and address findings), and hybrid analysis (tailored, validated insights that effectively drive taxonomy change and change management). By pairing statistical analysis with end-user validation, they shared, a hybrid approach ensures every taxonomy change is evidence-based and user-approved. Statistical analysis treats a tagged content corpus as a measurable dataset, they noted, before exploring some common methods of statistical analysis and sharing examples of the Enterprise Knowledge dashboards they use. They suggested asking, When content doesn’t fit the taxonomy, is it noise, or is it evidence of a new concept?
KMWorld returned to the JW Marriott in Washington D.C. on November 17–19, 2026, with pre-conference workshops held on November 16th.
KMWorld 2026 is a part of a unique program of five co-located conferences, which also includes Enterprise Search & Discovery, Enterprise AI World, Taxonomy Boot Camp, and Text Analytics Forum.