Why Your ‘Data Exhaust’ Is Your Most Valuable Asset

Key Concepts

• Enterprise AI
• Data unification/harmonization
• Agentic AI
• Bad data (incompleteness, inaccuracy, lack of context)
• Data silos (databases, warehouses, data lakes, lakehouses)
• Data gravity vs. data fluidity
• Zero-copy network
• Deeply unified platform
• Trusted AI
• Data governance (tagging, policies, attribute-based access control)
• 360 experiences
• DataOps

Main Topics and Key Points

The Problem of Bad Data in Enterprise AI

• The Persistence of Data Cleanup: Despite technological advancements, data projects still require significant time for data cleanup.
• Bad Data Defined: Bad data manifests in various forms, including incompleteness, low quality/fidelity, and inappropriate summarization.
• Impact on AI Experience: Inaccurate data negatively impacts user experience and trust, leading to skepticism and rejection of AI-driven recommendations.
• AI Amplifies the Problem: AI's intelligent automation at scale magnifies the consequences of bad data, potentially violating trust and leading to inaccurate or irrelevant outcomes.
• Example: Inaccurate sports statistics leading to loss of trust in the data source.

Evolution of Data Architecture and the Creation of Data Silos

• Historical Context: The evolution from structured data in relational databases to data warehouses for analytics, then to machine learning models trained on data snapshots.
• Generative AI Shift: A paradigm shift where applications no longer treat data as an exhaust but as a means to define the applications themselves, enabling personalization.
• Proliferation of Data Systems: The emergence of diverse data systems (databases, warehouses, data lakes, lakehouses) with their own storage protocols, table constructs, file constructs, and query protocols (including SQL variants).
• Creation of Data Debt: These disparate systems create data silos, trapping valuable data and hindering seamless access for meaningful business interactions.
• Salesforce's Solution: Salesforce aims to solve this problem with Data Cloud and Agent Force, focusing on untrapping data from silos.

Unified Data and Agentic AI

• Harmonization vs. Unification: Harmonization refers to creating consistent schemas across different data sources, while unification involves connecting data from different modalities (e.g., email address and phone number) to represent the same entity.
• Data Fluidity: The goal is to bring data to life in a fluid manner, enabling real-time access and integration without replacing existing data silos.
• Agent First Experience: Example of Salesforce's help.salesforce.com, where AI agents provide richer answers by combining unstructured textual data with structured data.
• Customer Service Example: A customer service representative using an AI agent to determine whether to raise a customer's credit limit by combining transaction history (from Snowflake), credit score (from a warehouse), and credit limit criteria (from S3 or Google Files).
• Benefits of Unified Data: Increased productivity for human agents, more relevant and precise AI agents, and improved decision-making.

Data Governance and Trust in AI

• Elevating Identity Management: Moving beyond identity management to encompass access, trust, governance, and security.
• Importance of Data Governance: Enterprises need to ensure data governance to protect their brand, security, customer experience, and data privacy.
• Salesforce's Data Governance Offering: Automated tagging of data (structured and unstructured) with classifications like PII, PCI, social security number, etc.
• Policy Enforcement: Defining and enforcing policies (default or human-driven) using attribute-based access control to restrict data access for agents and humans.
• Trusted AI as a Differentiator: Delivering trusted AI at the right performance and scale is a key differentiator for businesses.
• "Went slow to go fast": Salesforce's approach to delivering trusted generative AI, prioritizing data governance and security.

Technical Direction and Team Structure

• Outcome-Driven Approach: Start with desired outcomes and work backward to determine the necessary architecture and systems.
• Data as a Central Role: Data should play a central role in the agentic enterprise by providing unified, trusted, and real-time data to fuel agents.
• Breaking Down Silos: Focus on breaking down data silos and enabling data fluidity through normalized protocols, schemas, and unified data.
• Emergence of DataOps: The rise of DataOps reflects the growing importance of data management and its integration with DevOps practices.

The Role of Relational Databases

• Transactional Systems: Relational databases continue to serve as the backends for transactional systems.
• Evolution of Databases: Databases have improved to handle structured and unstructured data, create graphs, and scale out in the cloud.
• Data Platforms: Data platforms like Salesforce Data Cloud provide lakehouse foundations and tooling to make data relevant for AI agents and transactional systems.
• Zero-Copy Network: Salesforce created a zero-copy network to allow customers to use data where it resides without moving it, addressing sensitivity and economic concerns.
• Normalization and Standard Interfaces: A trend towards normalization and standard interfaces (e.g., Apache Iceberg) to enable seamless data exchange between systems.
• Connectivity Through Graph: The concept of connectivity through graph, where all touchpoints are driven through profiles (human, lead, account, vehicle, etc.).

Salesforce's Deeply Unified Platform

• Core Components: The deeply unified platform includes the agentic foundation, Data Cloud, deterministic flow-based automation (Flow), Tableau (with Tableau Next running on a semantic model built inside Data Cloud), Slack, and industry applications.
• Consistent Semantics: The platform provides consistent semantics across the enterprise, enabling better understanding and utilization of data.
• Fueling the Agentic Enterprise: The deeply unified platform serves as the fuel that drives the agentic enterprise.

Notable Quotes

• "Bad data really comes in many different forms, right? Is it complete? Is it high of high quality, high fidelity? Is it been uh is it been summarized in a manner in which it's meaningful for the topic at hand?" - Rahul Arakar
• "What AI is doing is it's it's driving intelligent automation and doing that at scale." - Rahul Arakar
• "We went slow to go fast because we wanted to make sure that we deliver trusted data for trusted agents." - Rahul Arakar
• "Data should play a central role in the agentic enterprise is to provide unified trusted and real-time data as a foundation that fuels agents" - Rahul Arakar

Technical Terms and Concepts

• Agentic AI: AI systems that act autonomously to achieve specific goals.
• Data Gravity: The tendency of data to attract applications and services to it.
• Data Fluidity: The ability to move and access data seamlessly across different systems.
• Zero-Copy Network: A network that allows data to be accessed and used without being physically moved.
• Deeply Unified Platform: A platform that integrates various data and application services to provide a consistent and unified experience.
• Attribute-Based Access Control (ABAC): An access control method that grants or denies access to resources based on attributes of the user, the resource, and the environment.
• PII (Personally Identifiable Information): Information that can be used to identify an individual.
• PCI (Payment Card Industry): Standards for protecting credit card information.
• Lakehouse: A data management system that combines the features of data lakes and data warehouses.
• Apache Iceberg: An open-source table format for large analytic datasets.

Logical Connections

The discussion flows logically from the initial problem of bad data to the evolution of data architectures, the creation of data silos, and the need for unified data. It then explores how unified data enables agentic AI, the importance of data governance and trust, and Salesforce's technical direction in addressing these challenges. The conversation concludes with insights on team structure and the role of relational databases in the modern data landscape.

Data, Research Findings, or Statistics

• Salesforce's Flow service drives trillions of automations within their business applications.

Synthesis/Conclusion

The key takeaway is that enterprise AI's success hinges on addressing the persistent problem of bad data and the challenges posed by data silos. Data unification and harmonization are crucial for enabling agentic AI and delivering trusted, relevant, and accurate insights. Salesforce is tackling this challenge with its Data Cloud and Agent Force, focusing on data fluidity, data governance, and a deeply unified platform. The future of data architecture involves a shift towards normalization, standard interfaces, and zero-copy networks to facilitate seamless data exchange and connectivity across diverse systems.