Artificial intelligence transformation across industries is a buzzword, but companies on the ground often cannot trust the analysis that AI produces. The reason is a recurring hallucination problem, in which AI misinterprets data or fabricates information that does not actually exist. In a survey Anthropic released in March, the AI concern respondents cited most was "errors caused by hallucination."
Matthew Miller, vice president of product management for Salesforce's Tableau unit, met with The Seoul Economic Daily on the 19th at Salesforce's Korea office in Yeongdeungpo District, Seoul. "General-purpose AI services based on large language models give a slightly different answer each time you ask, but Tableau does not," he said, distancing himself from fears of a "SaaSpocalypse" in which AI agents would replace the software industry wholesale.
Tableau, a data analytics and visualization platform founded in the United States in 2003, was acquired by Salesforce in 2019. Once data is entered, it runs analysis according to predefined rules and a client's own policies. Miller defined the technology behind Tableau's consistent answers as "deterministic AI," stressing that "it uses tokens only when first building dashboards and reports, so it also costs less than an AI agent."
He said the trend of non-experts each writing their own code to build agents could instead become a burden for companies. "Software is not something you build once and are done with — continuous maintenance and security upkeep are essential," Miller said. "If agents acting as software are mass-produced without control, that amounts to a company piling up 'technical debt.'" Agents can replace simple services like a household budget app, he explained, but they cannot touch enterprise services that require high regulatory compliance and security.
A "semantics" system that standardizes tacit data knowledge is another of Tableau's strengths. "Many companies fail to use their data properly because departments and industries define terms differently," Miller said. "Tableau works with a client's management and business units to standardize the definitions of terms and enable data analysis that fits the context." For example, the English term "conversion rate" is read as advertising or campaign performance by a marketing department but as a foreign exchange rate by a finance department. Companies also differ in what information they count as revenue. Tableau enters this tacit knowledge into the back end of the system to draw out accurate analysis.
"Just as Steve Jobs once compared the computer to an 'e-bike for the mind,' AI and technology maximize human capability," he said. "When humans and AI learn from and use data together, an organization's capability becomes so powerful that one plus one plus one is not three but 17."
Separately, Miller attended the "Salesforce Tableau Data Fam Seoul" event held on the 20th in Gangnam District, Seoul, where he said Tableau would strengthen the application of AI across its solutions while also expanding integration with outside programs through methods such as the Model Context Protocol (MCP). The event also showcased how domestic companies including LG CNS, Toss Bank and Olive Young use Tableau.
