How Salesforce’s Data Cloud Drives the Next Generation of AI-Powered Enterprise Apps

Steve Fisher has a long history with computer science, Salesforce, and Marc Benioff. 

As a high school student in Silicon Valley in the late 1970s, Fisher teamed up with Benioff to develop software games for Apple and Atari computers. The two parted ways when Fisher went to Stanford to study computer science and Benioff went to the University of Southern California to study business. They reunited when Fisher joined Salesforce in 2004 after stints at Apple, AT&T, and NotifyMe Networks, a startup he co-founded. 

After 10 years leading the technical design and development of Salesforce software, Fisher joined eBay as its chief technology officer. But in 2021, he returned to Salesforce and picked up where he left off. Now, Fisher is leading the development of the next generation of CRM and unified data services, including Salesforce Data Cloud, a real-time hyperscale data engine that unifies and harmonizes data from any source to create a single view of the customer. 

Salesforce’s Dan Farber talked with Fisher about the evolution and future of Data Cloud.

Q. Why is Data Cloud so important to Salesforce’s future and our customers’ AI transformations?

The simple answer is, ‌AI is only as good as the data that it has. 

Data Cloud is deeply integrated with everything in Salesforce. So you don’t have to organize all your data in one place and then figure out how to use it to engage with your customers. Data Cloud makes it really easy to bring all your Salesforce data together, whether from multiple clouds or multiple Salesforce orgs. And all that data is unified and harmonized in a single space that is accessible for every application to use.

You could just use the transactional CRM data we have for AI applications, but it would only be a piece of the data puzzle. Data Cloud unifies all of your data across all different, disparate systems — not just Salesforce systems. In fact, 75% of the data in Data Cloud today comes from outside Salesforce. This means data from websites, telemetry data, and both structured or unstructured data. You can easily bring in data with connectors and federate with big data providers in a secure, compliant way from external data platforms like Snowflake, Google, AWS, and Databricks with zero-copy or zero-ETL [Extract, Transform, Load] for data sharing and AI model training.  

Data Cloud makes bringing whatever data you want into Salesforce data easy. By unifying and harmonizing that data, it creates a golden record that contains all the information about your customers, your orders, cases, vehicles, or whatever entity you choose. And much of the data processing can be done in real time, which is really important for applications like logistics tracking, fraud detection, patient information, and customer engagement. For example, we can build marketing segments by looking at how customers are engaging with our products on our website and automatically adding them in and out of the segments. 

Data Cloud makes bringing whatever data you want into Salesforce data easy. By unifying and harmonizing that data, it creates a golden record that contains all the information about your customers, your orders, cases, vehicles, or whatever entity you choose.

Steve fisher, EVP and GM of Next Gen CRM and Unified Data Services

Data Cloud is the foundation for CRM and our customers’ AI transformations. That’s why it’s become the fastest growing product in Salesforce’s history. Data Cloud is already processing 30 trillion transactions per month, and connecting and unifying 100 billion records every day.

Q. What’s the relationship between Data Cloud and the Salesforce Einstein 1 Platform?

The key to our ability to rapidly innovate is the metadata-driven Salesforce platform. When the first release of Sales Cloud came to market nearly 25 years ago, the big idea was to expose Salesforce metadata — the schema that describes data models, record types, business rules and processes, page layouts, user permissions, and more. It abstracted away the complexity of application code, and today, enables us to upgrade customers with new capabilities automatically, three times a year, without breaking any of their integrations, customizations, or security settings.

Data Cloud is deeply integrated into the Salesforce Einstein 1 Platform, and now serves as a foundation for the platform and all Salesforce apps. In fact, all the data in Data Cloud manifests through the metadata in the platform, making it seamless to use that data across Sales Cloud, Service Cloud, and our industry clouds. At Dreamforce 2023, we announced that Marketing Cloud and Commerce Cloud, consumer-scale technology stacks that joined Salesforce’s Customer 360 portfolio via acquisitions, are now native on the Einstein 1 Platform.

In addition, platform capabilities — including Einstein, Flow, Lightning, and Apex — have native access to Data Cloud, so customers can easily supercharge their business applications with powerful AI, as well as automation and analytics, using low-code or no-code tools. Tableau also natively connects to Data Cloud, instantly analyzing data with the click of a button and taking action with AI-powered insights in the flow of work.

Data Cloud provides the data and grounding for Einstein Copilot, our new generative AI conversational assistant, as well as Einstein Copilot Builder, a new way to build and tailor AI assistants, to deliver a new generation of AI-powered apps. With Data Cloud, we can create a data graph that provides a real-time, consolidated view of a customer or any entity. And with just one click, customers can send all the relevant data to the prompt that then feeds the LLM. You don’t need to send SQL queries or create data joins manually. 

We’re going to move from building apps to building prompts and copilots that reason over the data and take actions on your behalf. And, you won’t have to learn a new tool set — you’ll be able to use all the Salesforce tools that you’ve already learned.

We’re going to move from building apps to building prompts and copilots that reason over the data and take actions on your behalf.

Steve fisher, EVP and GM of Next Gen CRM and Unified Data Services

Q. What’s an example of how Data Cloud powers generative AI applications?   

Let’s say you’re in Marketing Cloud. With Data Cloud and Einstein AI, you can generate marketing campaign segments, create website landing pages based on consumer browsing and buying preferences, and personalize emails for specific marketing campaigns. When you create a segment, for example, the Einstein Copilot conversational assistant is there and you just tell it what you want.  

What you’re doing is creating a prompt. And the output of the AI model is only as good as the quality of the prompt. You could prompt the AI to generate an email inviting a prospect to an event. It will do that, but it’s going to be a pretty generic email without grounding in contextual data, such as their name, company, and title; all the products the prospect has purchased or is interested in; how long they’ve been a customer; what emails they’ve opened; and whether they have any open service cases. A salesperson could look up all that information, and type in paragraph after paragraph and feed it to an LLM. But what a nightmare — it could take hours. 

This is where Data Cloud comes in. From the prompt to the output, it’s all fully integrated. Data Cloud automatically sends all that relevant data to the AI model, giving it the context to reason over the prompt and generate quality output. So, without any heavy lifting, you’re going to get a 100 times better email than any sales rep is ever going to generate by typing something into ChatGPT.

Q. How does Data Cloud deal with security and privacy regarding moving business data into Einstein Copilot apps and workflows?

Data Cloud is very good about providing data access only to entities that should have access to that data. All of the privacy, security, and other data controls are part of how Data Cloud manages the data. This is particularly important with large language models that power generative AI.

LLMs are very good at language, but they’re not really that good at data. And what I mean by that is they’re fine for public data, such as what’s in Wikipedia, which doesn’t change very often. But if you have proprietary data or data that’s changing often, that’s not ideal for an LLM. For LLMs to learn about new data, you have to train the model, which is expensive and isn’t typically done in real time. 

You also have to consider security or privacy requirements — meaning not everybody within the company or within the world should see this data, so you should not put it into the LLM because chances are it’s going to be exposed. Once you put it in an LLM, you really can’t delete it. There’s no row that you can delete from a relational database table, so you have to be super careful. I think this is what worries customers most — they don’t want their proprietary data training those public models.

We have a very robust, rich sharing and security model, and we preserve that when you’re engaging with AI models. The Einstein Trust Layer is natively built into the Einstein 1 Platform. When you enter a prompt in Einstein Copilot or use a pre-built prompt template, the relevant data is securely retrieved from Data Cloud. Before we send it to the LLM, we mask any information that’s proprietary, sensitive, or confidential and then send it to the AI model through our Secure Gateway. And, thanks to a zero-retention policy enforced across all of those models, the data is never stored outside of Salesforce and will never be used to train the models. 

Once the output is generated, the Einstein Trust Layer also goes through a series of checks against bias and toxicity. Then we track and maintain an audit trail of this entire interaction, so you know what’s been used to generate that content.

Q. What’s next for Data Cloud?  

At Dreamforce, we announced expanded support for unstructured data, like large text data in PDFs, call transcripts, or Slack conversations. Data Cloud will also make it easier to index these data types from across internal and external knowledge stores to simplify semantic search and retrieval without added custom work. 

We’re really excited to get Data Cloud into the hands of every customer — it’s the foundation for enterprise applications going forward. As part of that goal, every Sales Cloud and Service Cloud customer with Unlimited Edition or Enterprise Edition can get a free Data Cloud Starter, which gives them the ability to unify 10,000 customer profiles and gain insights with two Tableau Creator licenses.

More information:

• Learn about Data Cloud
• Read how Salesforce uses Data Cloud in this interview with Janani Narayanan
• Get all the news from Dreamforce 2023, including Salesforce’s data sharing partnerships