Deasie Secures $2.9 Million Seed Funding to Enhance Data Governance for Generative AI
Deasie, an innovative startup creating tools to empower businesses with enhanced control over text-generating AI models, has successfully raised $2.9 million in a seed funding round. This round saw valuable participation from investors including Y Combinator, General Catalyst, RTP Global, Rebel Fund, and J12 Ventures.
The founding team of Deasie—Reece Griffiths, Mikko Peiponen, and Leo Platzer—previously collaborated on data governance solutions at McKinsey. During their time there, they identified significant challenges and opportunities within enterprise data governance, particularly issues that could impact a company’s ability to successfully adopt generative AI.
Their observations align with findings from a recent IDC survey, which involved over 900 executives from large enterprises. The survey revealed that 86% of respondents believe more governance is essential to ensure the “quality and integrity” of insights generated by AI. However, only 30% of respondents felt “extremely prepared or ready” to utilize generative AI effectively at present.
To address these challenges, the Deasie team has developed a product designed to enhance the reliability of generative AI models, particularly large language models (LLMs) like OpenAI’s GPT-4. Their platform connects to diverse unstructured data sources—such as documents, reports, and emails—automatically categorizing them by content and sensitivity.
For instance, Deasie can auto-tag a report with labels like “personally identifiable information” or “proprietary information,” while also noting that it is the third version of that document. In another case, it might tag a specification sheet as “proprietary information” and indicate restricted access rights. Customers define the tags and labels according to their classification strategies, which in turn trains Deasie’s algorithms for better data categorization in the future.
Once Deasie auto-tags documents, the platform assesses the generated library of tags to evaluate data relevance and importance. Based on this analysis, it decides which data to input into a text-generating model.
“Enterprises generate massive amounts of unstructured data that often lacks proper governance,” Griffiths explained. “The risk of language models producing nonsensical answers or mishandling sensitive information increases with the volume of data. Deasie functions as an intelligent solution that sifts through thousands of documents within an enterprise, ensuring that only relevant, high-quality, and secure data is channeled into generative AI applications.”
Deasie presents a compelling solution. Implementing a strategy that restricts an LLM to vetted data is a wise approach, particularly in light of the potential ramifications of exposing LLMs to outdated or contradictory information. However, questions remain around the consistency of Deasie’s algorithms in data classification and the frequency of errors in evaluating document significance.
As with any demo, organizations need to address these concerns adequately. Griffiths shared that Deasie, which is currently a team of three, has secured its first pilot agreement with a "multi-billion-dollar" enterprise in the U.S. Additionally, they have an impressive pipeline of over 30 enterprise customers, which includes five Fortune 500 companies.
“Previous solutions have either centered on the ‘data safety’ aspect or the ‘structured data governance’ approach for LLM governance,” noted Deasie. “What has been absent is an effective method to assess the quality and relevance of unstructured data. No one has directly tackled the challenge of matching every generative AI use case with the most suitable dataset. Deasie has pioneered innovative strategies in this space.”
In the coming months, Deasie plans to expand its engineering team and make key hires, focusing on developing features that will set it apart from competitors like Unstructured.io, Scale AI, Collibra, and Alation.
Ranking
