Keep up with the innovative tech transforming business
Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.
Salesforce is beefing up its AI offerings as it gears up to roll out Einstein Copilot, the company’s new all-purpose conversational AI assistant for the office, early next year.
The enterprise giant announced at an event in New York this week that it will begin offering support for unstructured data like emails, PDFs, and transcripts in combination with structured business information, to more easily customize generative AI applications without intensive fine-tuning. The company also said it’s adding new search capabilities to its forthcoming Copilot product to better navigate that data.
“Einstein Copilot is an entirely new paradigm for Salesforce, an entirely new user interface,” Patrick Stokes, EVP of product and industries marketing, said in the keynote. “It is a fully working generative AI embedded right into the flow of work, embedded right into all of our best-in-class applications.”
Salesforce is not the only company eyeing the enterprise market for so-called Copilots, or generative AI-powered workplace assistants that can retrieve information and help draft various communications. Microsoft’s similar Copilot 365 is now widely available as of last month. Google also released a version of its new Gemini model for enterprise this week.
Salesforce is hoping this new announcement will help better integrate one of its key edges in this race: the sprawling databases of existing enterprise data that its clients keep on the platform.
As an example of how this might look, a customer service rep could use an Einstein Copilot search to call up both the customer’s support ticket history (structured data) and history of call transcripts and emails (unstructured data) to better understand a dilemma.
The company has found that combining these two forms of data and asking large language models (LLMs) to draw from that pool cuts down on the amount of fabrications, or “hallucinations,” the AI produces, according to Rahul Auradkar, EVP and GM of Data Cloud and Einstein Platform at Salesforce.
“We have looked at generative AI outcomes when we don’t ground it; the LLM comes back and hallucinates a lot, as in, ‘You might want to talk to the customer about XYZ,’ and every time there’s a different answer,” Auradkar told Tech Brew. “But when we ground it with [unstructured data layered with structured data], we are all of a sudden getting very relevant responses.”
Salesforce cites an IDC stat that “90% of enterprise data exists in unstructured formats,” and a Forrester Research prediction that the volume of that data will double by next year. Salesforce joins enterprise data platforms like Databricks in offering this kind of combination of structured and unstructured data. This kind of data can also be a key ingredient in training specialized LLMs.
While structured information like its customer relationship management (CRM) system has been Salesforce’s bread and butter, Auradkar said “unstructured data is the place where we’re bringing it to life.”