Impact of AI on KM Strategy: A Two-Way Street

This desire leads naturally to the need for well-organized, accurate, and verified content. “The knowledge agents get trained on controlled datasets, including content from Happeo Pages, our content management system, and integrated sources like Google Drive and OneDrive. Gap detection and verification processes ensure answers are up-to-date and accurate,” Pinkham added. “We automate as much as possible to help overcome the laborious parts of the process, such as development of a taxonomy, but we keep humans in the loop. There is no substitute for reliable and timely data. People do not want to deal with misinformation or conflicting answers.”

However, wishful thinking prevails in some cases. “People are looking for a magic solution,” said Victor Manrique Yus, principal product manager at Happeo, “but putting a chatbot on top of the existing layer of information accumulated over the past 20 years will not produce the results they want.” Sometimes, the outcome is that AI gets blamed, and the company abandons the project. Other times, the company regroups and works on cleaning up the foundational content.

When properly supported by verified data, AI is empowering. “AI looks at all that is happening in an organization and will bring up things you don’t know about,” commented Yus. “Knowing your role in the company, for example, an AI agent could alert you to a relevant policy update that you had not heard about.” The proactive abilities of AI agents bring a new dimension to KM and enable organizations to elevate their KM strategies.

Orchestrating AI Agents

Nearly any organization can benefit from extracting accurate, actionable insights from customer feedback. This involves identifying and validating topics mentioned in each piece of feedback, checking whether similar topics have already been detected, tagging documents accordingly, and suggesting when it may be useful to split broader topics into more granular subtopics. A common business goal is to understand which topics are key drivers of important KPIs, such as customer satisfaction or sales volume.

Since organizations cannot always predict what issues customers will raise, an insight extraction solution based on analyzing customer information should be able to discover topics and subtopics through semantic analysis of customer feedback with minimal guidance from a human analyst. When initial seed categories (small sets of words used to train a learning model) do exist, the system should be able to monitor and expand these categories while also surfacing previously unknown patterns.

Megaputer’s AI Insight Extractor is a solution built on the PolyAnalyst no-code platform. “PolyAnalyst provides built-in access to multiple LLMs [large language models] and supports visual design of multistep data processing workflows,” said Sergei Ananyan, CEO of Megaputer, “which makes it easy to comprehend the analysis logic and fine-tune the solution.” The platform accelerates prototyping and simplifies ongoing maintenance of business solutions.

AI Insight Extractor includes a human in the loop to validate the results and align them with business objectives. “Analysts typically need an interactive visual presentation of the resulting topic taxonomy,” observed Ananyan, “and it is helpful to see the strength of each topic’s impact on the KPI of interest.” Humans in the loop can also inspect the underlying feedback that contributed to each topic, with relevant patterns highlighted in the text. Once a human analyst has validated the results, the key topics can be passed to another AI agent for deeper analysis. This downstream agent focuses on uncovering root causes of issues and providing actionable recommendations on how these issues can be fixed.