AIwire recently spoke with Cindi Howson, Chief Data and AI Strategy Officer (CAIO) at ThoughtSpot, following the company’s inclusion in the Gartner Magic Quadrant. As organizations move beyond AI experimentation and look to generate measurable business value, Howson argues that success will depend less on LLMs and more on giving AI the business context it needs to deliver accurate and trusted results.
Founded in 2012, ThoughtSpot is an enterprise analytics company that helped popularize search-driven analytics, making it easier for business users to ask questions of their data using natural language. In recent years, the company has expanded its platform to include generative and agentic AI capabilities.
Gartner recently predicted that by 2027, organizations that prioritize semantics in AI-ready data will improve agentic AI accuracy by up to 80% while reducing costs by as much as 60%. According to Howson, those gains won’t come from the language models themselves but from the semantic foundation that sits beneath them.
“The number one mistake that customers are making is confusing a BI semantic layer with an agentic semantic layer,” Howson told BigDATAwire. “A BI semantic layer in its most basic form really just took business terms and pointed them to the physical tables in a data warehouse. With an agentic semantic layer, the data sources are no longer just your data warehouse. It can be transactional databases, operational systems, and repositories of unstructured data.”
Cindi Howson is the Chief Data and AI Strategy Officer at ThoughtSpot
The role of the semantic layer has also expanded beyond traditional business intelligence tools. “It’s no longer just a BI tool that needs this. It can be a range of office tools. It can be agents talking to other agents,” she explained. “I can ask an insight question and tell it to take action.”
ThoughtSpot refers to its approach as an AI semantic layer, combining metrics, ontology, context and memory, and knowledge graphs into a unified foundation for AI agents.
“It still has your metrics and data model… but then it also has this ontology layer that tells me how things relate. Your context and memory… is also part of it and then the knowledge graph… is part of the knowledge graph. That is what I would say is an AI semantic layer.”
The distinction has become increasingly important as enterprises attempt to connect generative AI directly to corporate data. “So think about the training data for the large language models,” said Howson. “They’ve largely been trained on public data sources and on code. They were not trained on structured corporate data. They cannot just infer that context on their own. The semantic layer gives them that trusted context.”
Without that business context, organizations risk inaccurate answers, hallucinations, and potentially significant business consequences.
“If the AI agent says, ‘Yes, here’s how many widgets you have,’ and that is wrong, then first off you have dissatisfied customers, missed sales opportunity. If you are reporting those numbers externally, you have loss of brand trust and you have the risk of regulatory fines.”
Howson also believes semantic layers play a significant role in lowering AI costs, not only by reducing manual work but also by improving how models access enterprise data.
“If you’re not using AI, you’re doing an old-school process of asking the data expert to give you that insight. By letting the business person ask the questions themselves, that reduces the costs to get to the insights.”
She added that token consumption represents another major opportunity for savings.
“In the absence of a semantic layer, if I just go LLM straight to the database, the LLM is going to read as much as it can from the database and stuff that into the context window. I am consuming both a lot of token costs and, if my database is consumption-based, I’m consuming data warehouse consumption credits. With the ThoughtSpot semantic layer, I already have the context and I’m just giving a very precise question and context to the LLM.”
(Shutterstock/tadamichi)
Customer adoption also reflects growing confidence in governed AI, according to Howson. “Sixty-four percent of our customers are using Spotter. Compare that to the industry average that Gartner has quoted… only 8% are using agentic AI.”
She expects semantic layers to become a permanent component of enterprise AI architectures – but not as a single centralized platform. “The number one mistake that customers make is thinking that they will have one semantic layer to rule them all. They will not.” Instead, organizations should focus on interoperability.
“Gartner uses the word mosaic. Customers should not be pursuing one semantic layer to rule them all. Instead, they should be pursuing interoperability between these different elements that make up a robust semantic layer.”
Howson believes enterprises should judge AI initiatives by user trust rather than infrastructure metrics. “ROI is a lagging indicator. People will use the system that they trust. If you don’t trust it, if it gives you wrong answers, you’re not going to adopt it… Adoption comes from trust. The semantic layer enables the trust.”
Editor’s note: This article originally appeared in BigDATAwire.
The post Why the AI Semantic Layer Is Becoming the Foundation of Enterprise AI appeared first on AIwire.
