What makes Einstein Copilot a genius? Salesforce says it’s all about the data

Salesforce today announced the public beta release of Einstein Copilot, a conversational AI assistant that empowers Salesforce users to interact with their data and workflows.

The Einstein Copilot was first announced as an early prototype at the Salesforce Dreamforce conference in 2023. With the beta release today, more users will be able to try out the technology which provides a more open, conversational and intuitive integration of AI capabilities than what Salesforce has had to date.

“Einstein Copilot really is that new conversational AI assistant for every Salesforce customer to interact with all of their data and their workflows in a very new way,” Clara Shih, CEO of Salesforce AI told VentureBeat. “We did preview it at Dreamforce and the team has been working furiously, I don’t know if we’ve ever shipped a totally new category like this, this quickly.”

To help validate the need for Einstein Copilot, Salesforce today also released new research from its Slack business unit that claims 86% of IT executives expect gen AI to have a big impact on their enterprises. The research also found that 80% of employees who are using gen AI tools are already experiencing a boost in productivity.

How Einstein keeps getting smarter with AI

The Salesforce Einstein technology did not get its start with the generative AI era and the debut of OpenAI’s ChatGPT in 2023.

The original Einstein technology predates all the generative AI hype and is a platform that has its roots in predictive machine learning (ML) and AI algorithms. Back in 2020, Salesforce claimed that the Einstein platform was serving more than 80 billion predictions per day. In March 2023, Salesforce expanded the platform for the gen AI era with the debut of Einstein GPT.

Shih emphasized that the pre-gen AI Einstein predictive AI technology is still very much alive and well at Salesforce and remains a cornerstone of the company’s AI capabilities. With the gen AI craze, she said that the value and usage of predictive AI have only grown and become more impactful.

Predictive AI that Einstein has always had, enabled capabilities such as providing a salesperson with a recommendation for the next best action to take in a sales process or a sales forecast to help a sales manager figure out if they will meet a quota. The Einstein GPT capabilities expanded on that with specific and predefined AI use cases that were embedded in certain Salesforce workflows such as the Sales GPT and Service GPT.

“What makes Einstein Copilot different is it goes from those predefined, pre-specified use cases to an open ended assistant that you can ask any question of across your Salesforce data and your workflows, so it’s not predefined,” Shih said.

Hyperforce data architecture is the foundation for Einstein Copilot

Einstein Copilot isn’t just some kind of basic gen AI wrapper that enables organizations to ask natural language questions about data. Rather, Shih emphasized that it’s a new interface and approach to leveraging the power and data that the Salesforce data stack provides for organizations.

The foundation of that stack is the Salesforce Hyperforce architecture, which provides data residency and compliance for enterprise use cases. On top of Hyperforce is the Salesforce Data Cloud for unifying, harmonizing and cleansing data from multiple sources like Salesforce instances. The Salesforce Data Cloud can also federate with external data lakes and data warehouses such as Snowflake, Databricks, Amazon Redshift and Google BigQuery.

Shih explained that the Einstein LLM (large language model) gateway sits on top of that foundation. The Einstein LLM gateway pulls in different gen AI models depending on the use case requirement. Salesforce has developed its own LLMs and also has partnerships with leading vendors including OpenAI, Google, Anthropic and Cohere.

On top of the LLM gateway is the Einstein Trust Layer that provides data privacy, security, toxicity filtering, citations and the tools necessary to support trustworthy AI for enterprise consumption. Shih explained that the overall AI platform and strategy at Salesforce that encompasses all of these underlying technologies that enable products like Einstein Copilot is branded as Einstein 1.

The genius of Einstein is all about context

There is no shortage of organizations building AI copilots and conversational UI interfaces. 

Microsoft for example has a growing list of Copilot services across its software portfolio.

Shih stressed that there is a big difference between what Salesforce is doing with Einstein Copilot and others like Microsoft. In her view, that difference is the contextual data.

According to Shih, it’s metadata that gives context to the data, which is of critical importance. Salesforce’s metadata model has long provided a structured way to define objects, fields, relationships and business logic. For example, metadata can help to identify in a sales context what an opportunity is and how an opportunity is related to an account, with a close date and a financial amount.

“At Salesforce we’ve had our metadata layer since the very beginning and I mean, it wasn’t created for AI, but it turns out that metadata is essential for AI to function properly,” Shih said.