How Salesforce’s AI Research Is Shaping CRM, Society, and the Future

Thinking big – like industry-or-paradigm-shifting innovation big – requires a lot of resources. Smarts, time, planning, patience, investment. But the most crucial element? An innovation mindset made up of equal parts culture and left-brain, right-brain thinking.

Thinking big also requires an ability to see into the future.

“To stay ahead, you have to be looking ahead,” said Shelby Heinecke, an AI Research Manager at Salesforce. She would know: Her team is laser-focused on driving AI innovation through a variety of groundbreaking projects, from academic research to society-shifting applications of the technology.

The team is also using their innovation mindset to identify new ways to apply AI to CRM.

“We’re all scientists at heart, but in the back of our heads, we’re always thinking about how AI can make our customers more successful,” Heinecke said. “How would this idea fit into one of our products? Are there ways that a customer would use this that we can include in how we frame our fundamental research?”

The road to the #1 AI CRM

Salesforce has been investing and innovating in AI since 2014. It launched Einstein, the company’s first AI product in 2016, and began work on large language models (LLMs) in 2018.

This year, the company released a host of new Einstein products to help customers get the most out of new generative AI technology. These products are especially effective because of Salesforce’s bread and butter, CRM, data, and trust.

At its core, CRM is all about organizing huge amounts of data to help organizations connect with their customers in more personalized ways. Salesforce has been a pioneer in this market for nearly 25 years, innovating and building products that leverage connected data like no one else can — and with trust at the center.

Salesforce Data Cloud, which connects and harmonizes customer data to deliver more real-time insights, gives customers a complete, trusted data foundation, so they can reap the benefits of more accurate AI. And, Data Cloud is deeply integrated into the core of the metadata-driven Salesforce platform, which allows customers to quickly action all their data from any source with AI, automation, and analytics across sales, service, marketing and commerce applications.

Caiming Xiong, VP of AI Research and Applied AI at Salesforce, explained, “The scale of customers, use cases, and data stored in Salesforce makes the company an amazing testbed for enterprise AI. Thousands of companies are running their businesses on Salesforce, generating and storing tons of data. AI is excellent at analyzing huge amounts of data, while humans aren’t.”

Applying research expertise to CRM – and society

Critical to every iteration of Salesforce’s AI roadmap has been its AI research group, now led by Silvio Savarese. Having spent a decade in academia before joining Salesforce in 2021, Savarese knows a thing or two about learning – be it by humans or machines.

“I decided to join Salesforce to lead the AI research org because Salesforce provides an amazing platform for innovation,” Savarese said, noting he couldn’t pass up the chance to work with “top-tier researchers, engineers, scientists, and be able to make an impact on the company and society at large.”

To date, Salesforce’s research team has published hundreds of AI research papers and AI patents. This work has informed Salesforce’s core AI offerings; supported other projects aimed at helping Salesforce employees and customers be more productive; and applied Salesforce’s AI to interesting and even groundbreaking use cases, like breast cancer treatment and protein development.

What makes the research team so valuable is that they take a customer-centric approach. The tools and products they develop aren’t arbitrary: They’re built specifically to solve or improve a customer need or problem.

Take their work developing applications like CodeGen and Apex Guru to apply generative AI to writing and optimizing code. With the knowledge that Salesforce customers “write a lot of code in Apex and related languages,” Savarese and his team set out to build something that made that process much faster and easier.

Using the same logic but a different “language,” developers can simply use commands and text prompts to create large amounts of code trained on high-quality examples, saving huge amounts of time rather than writing code from scratch. A few directional prompts in the English language, and what used to take hours is now reduced to minutes.

Another tool that the research team has led development on — co-created with and for a customer — uses generative AI to make chatbots more knowledgeable about products and inventory, providing the context, to handle more complex service inquires and even create upsell opportunities.

After hearing that the brand’s voice was getting lost in automated responses via chatbot, Heinecke and her team set out to use their AI technology to generate replies to a customer in their voice and style and include lots of detail and rich information about their products.

For brands, it keeps the authenticity and voice they’re known for within the entire process even when it’s automated, and customers get a better experience and recommendations that line up with what they’re looking for. The more the models are trained with further data, interactions, and experiences, the more capable and reliable the AI becomes, and the refined and successful the customer interactions become.

Salesforce AI… did what?

The research is transforming Salesforce products and benefiting Salesforce customers, but the research team is always thinking bigger.  As a company that believes business is the greatest platform for change, Salesforce has also taken the AI research that’s been the basis for advancing Einstein and applied it to areas outside the enterprise software space for the greater social good.

For example, Savarese and his team have used Salesforce’s AI expertise to:

• Track great white sharks with the aim of keeping beaches safer while also protecting the rebounding populations of sharks through Project SharkEye.
• Study financial models with the goal of finding solutions to complex economic scenarios through AI Economist.
• Help improve breast cancer treatment decisions while reducing costs and outcomes for breast cancer patients through ReceptorNet.
• Build novel proteins to help uncover new medicines, vaccines, and treatments for diseases through ProGen.

Using technology to create societal good is in Salesforce’s DNA, and according to Naik, that can only happen when a team is working together, arms locked, with a single motivation.

“You have to take a big risk by betting on a particular research technology and working on something that can have a high chance of failure. It requires innovation, courage, and the ability to constantly improvise and improve your technology. But most importantly, it requires teamwork and belief in each other,” he said.

The next AI Revolution

Savarese believes the next phase of AI is now coming to life in the form of AI Agents.

Think of AI Agents as an advanced, more capable and proactive version of AI. “AI Agents can perform actions on behalf of users. A digital personal assistant that can actually perform tasks on their behalf to make them more productive and let them focus on the most important critical creative aspects of their work,” Saverese said.

This happens thanks to what Savarese initially dubbed Large Action Models (LAMs), or AI Agents.

While initial generative pre-trained transformers – commonly known as GPTs – had individual LLMs working independently to generate something – like code, text, or images – based on a prompt, AI Agents are the next step in their evolution: they can perform tasks autonomously. And eventually, AI agents will connect and work together to create an exponentially more powerful and efficient AI network.

Here’s a real-world example from Heinecke that helps paint a picture: Imagine you want to plan a trip to Mexico, and you’re working with a chatbot. With something like AI Agents, not only will the chatbot give you options for what to do, it will actually take the actions for you – go to the websites, make the plans, and book the tickets. It’s a set of AI Agents with specific skills that know how to communicate with each other to manage tasks on your behalf. “This is how we take the LLMs to the next level, and this is what AI Agents are all about,” said Heinecke.

This isn’t just a theory or a hypothetical: It’s what Salesforce is building for customers with Einstein 1 Copilot. Announced at Dreamforce, Einstein 1 Copilot is a conversational AI assistant integrated directly into the user experience of every Salesforce application. Einstein 1 Copilot proactively offers options for additional actions beyond a user’s query – such as a recommended action plan after a sales call, or creating a new service knowledge article.

The future of AI at Salesforce 

As many are left playing catch-up in this new AI-first world, Salesforce is uniquely positioned thanks to the symbiotic relationship between CRM, data, and AI, and the research and innovation that powers it.

“It’s great to work at the frontier of AI for Salesforce,” Heinecke reflected. “There’s just an abundance of opportunity for what we can do.”

Savarese agrees – and recognizes that the work Salesforce is doing can shape the future of AI for the industry as a whole.

“AI is ushering in a new era of productivity, and there is no better place to pursue this vision than at Salesforce,” he said.

More information

• Learn more about Salesforce AI here
• Read more news and stories from Dreamforce 2023

Follow Revenova on LinkedIn, YouTube, and X (formerly Twitter) for the latest updates and news about RevenovaTMS, the original CRM-powered Transportation Management System.