Designing the Ideal Gen AI Data Layer: Key Insights from Intuit

In media coverage of generative AI, one enterprise company stands out for its rapid and effective deployment of the technology: Intuit.

In September, Intuit launched its LLM-driven assistant, Intuit Assist, across all its products, including TurboTax, QuickBooks, Credit Karma, and MailChimp. Earlier, in June, the company unveiled its Gen AI operating system, designed to orchestrate large language model (LLM) activity throughout the organization—an ambitious vision that precedes those of other major players in the industry.

In a recent interview, I spoke with Alon Amit, Intuit’s VP of Product Management, about a crucial aspect of achieving success with generative AI: developing a robust data management layer.

Amit noted that Intuit invested several years refining this data layer to ensure accurate, well-integrated, governed, and deduplicated data. Only after establishing this strong foundation were LLMs able to utilize the data to foster personalized interactions with Intuit’s 100 million small business and consumer customers.

During our discussion, Amit shared a pivotal slide illustrating Intuit’s data layer model, which outlines best practices for an effective data architecture.

For enterprise data leaders, I recommend watching the full interview linked above, as Amit details the key initiatives Intuit is pursuing and outlines goals for 2024. The interview is part of our AI Unleashed event, with the complete video included above.

Here are a few key takeaways from our conversation:

1. Data Map Registry: Intuit developed a centralized repository for all data assets—both real-time and batch—generated within the company, including all data schemas. This ensures data governance, with clear ownership and purpose established for each asset. While Amit acknowledged imperfections in the process, he is optimistic that they will achieve near-perfect governance by the end of next year.

2. Culture of “Data as a Product”: With the help of this data map, Intuit has cultivated a culture among developers, product managers, and engineers that regards all generated data as a valuable product, not just the data within customer-facing products.

3. Uniform Data Schema Governance: Intuit ensures that all data schemas—whether from click-streams or third-party sources—are managed consistently to prevent disruptions in downstream systems, especially those supporting generative AI. Data inflow, depicted on the left side of Amit’s chart, includes “domain events” like real-time data generated from application event buses, automatically integrating into Intuit’s data lake.

4. Governed Data Derivation: This term encompasses all transformations applied to source data, such as analytics computations, AI feature extraction, and marketing attributes. Developers are notified if they attempt to derive a feature that already exists in the data registry, thus avoiding redundancy.

5. Real-time Data Derivation: Planned for 2024, this initiative aims to enhance Intuit’s capabilities in providing near-real-time data responses. Amit emphasized that the company is committed to refining its processes to enable timely insights, ensuring that Intuit can swiftly understand user actions when responding to inquiries or providing expert support.

These strategies illustrate how Intuit is leading the charge in effectively leveraging generative AI through a robust data management framework.