Semantic understanding, not just vectors: How the data architecture of intuit Agentic Ai supplies with measurable ROI

Intuit – The financial software giant behind products such as Turbotax and QuickBooks makes considerable progress with generative AI to improve the offer for small business customers.

In a tech landscape that was flooded with AI promise, intuit has built up an agent-based AI architecture that provides tangible business results for small companies. The company has used what it describes as “done for you” experiences that deal autonomously with entire workflows and provide quantifiable business effects.

Intuit has expanded its own AI level, which calls a generative AI operating system (Genos). The company Detailed some options for using Gen AI to improve personalization VB transformation 2024. In September 2024 added intuit Agent Ai Workflows, an effort that has improved the company for both the company and its users.

According to the new intuit data, QuickBooks online customers are paid on average five days faster, with overdue invoices being paid by 10% higher. For small companies in which the cash flow is king, these are not only incremental improvements, they are potential business-saving innovations.

The technical trinity: how the data architecture of intuit enables real agents -KI

What distinguishes the approach of intuit from competitors is the sophisticated data architecture, which was specially developed to enable agent-based AI experiences.

The company has built what CDO Ashok Srivastava calls the “Trinity” of data systems:

1. Data lake: The basic repository for all data.
2. Customer data Cloud (CDC): A special layer of serving for AI experiences.
3. “Event bus“: A streaming data system that enables real-time processes.

“CDC offers a serving layer for AI experiences, then the data lake is a kind of repository for all of this data,” Srivastava told Venturebeat. “The agent will interact with data and has a number of data that he can consider to obtain information.”

Go beyond the vector embedding beyond the KI beyond the Power Agentic -KI

The intuit architecture deviates from the typical vector database approach that many companies hastily implement. While Vector databases and embedding are important for the performance of AI models, intuit recognizes that a real semantic understanding requires a more holistic approach.

“Wherever the main problem continues, it is essentially to ensure that we have a good, logical and semantic understanding of the data,” said Srivastava.

In order to achieve this semantic understanding, intuit builds a semantic data layer on its core data infrastructure. The semantic data layer provides the context and the importance of the data about only the raw data itself or its vectors’ representations. It enables the AI agents of intuit to better understand relationships and connections between different data sources and elements.

By building this semantic data layer, intuit can expand the functions of its vector -based systems with a deeper, more contextual understanding of data. In this way, AI agents can make more informed and meaningful decisions for customers.

Beyond the basic automation: How Agentic AI autonomously concludes entire business processes

In contrast to companies that implement AI for the basic workflow automation or customer service chatbots, intuit has concentrated on creating completely acting “for you”. These are applications that do complex, multi -stage tasks and at the same time only require the final human approval.

For QuickBooks users, the agent system analyzes the client payment process and the invoice status in order to automatically write personalized memory messages, so that business owners can easily be checked and approved before sending. The system’s ability based on relationship context and payment patterns have contributed directly to measurably faster payments.

Intuit uses identical agent principles internally and develops autonomous procurement systems and HR assistants.

“We can carry out an internal agent procurement process with which employees can buy deliveries and book travel,” said Srivastava and demonstrated how the company eats its own AI dog food.

Developed for the era of the argumentation model

What may offer a competitive advantage over other companies for companies for companies is how the system was designed with foresight over the development of advanced argumentation models such as Deepseek.

“We have built up Runtime, in which we expected to be on the part of argumentation models,” said Ashok. “We are not behind the eight ball … we are ahead. We have built up the skills that have assumed that argument would exist. “

This future -oriented design means that intuit is intuitively involved in new argumentation functions in your acting experiences if you appear without requiring architectural overhaul. According to Srivastava, the technical teams from Intuit already use these functions to justify the agent in a large number of tools and data in a way that was not yet possible.

Relocation from the AI hype on business effects

The most important thing may be that the intuitation approach points a clear focus on business results and not on technological showmanship.

“There is a lot of work and a lot of fanfare on AI these days itself that it will revolutionize the world, and everything I think is good,” said Srivastava. “But I think what is much better is to show that it actually helps real people better.”

The company believes that deeper argumentation functions will enable more “experiences for you” to cover more customer needs with greater depth. Each experience combines several nuclear experiences or discrete operations that together create a complete workflow solution.

What does this mean for companies that accept AI

For companies that want to implement AI effectively, Intuit’s approach offers several valuable lessons for companies:

• Concentrate on the results via technology: Instead of presenting AI for your sake, you aim for specific business pain points with measurable improvement goals.
• Building with future models in mind: Design architecture that can include aspiring argumentation functions without requiring a complete conversion.
• Fix the data challenges first: Before you hurry to implement agents, make sure that your data foundation can support semantic understanding and cross-system thinking.
• Create complete experiences: Look beyond simple automation to create end-to-end workflows that provide complete solutions.

Since the Agentic Ai continues to mature, companies that follow the example of intuitation by focusing on complete solutions and not on isolated AI characteristics that achieve similar concrete business results instead of simply generating tech sums.

Daily insights into the economic use cases with VB daily

If you want to impress your boss, VB Daily covered her. We give you the Inside scoop of what companies do with generative AI, from regulatory shifts to practical deprivation, so that they can share knowledge for a maximum ROI.

Read our Data protection guideline

Thanks for subscribing. Check out more VB newsletter here.

An error occurred.