What’s next for Salesforce Einstein AI and Data Cloud

Salesforce users await the rollout of a bundle of products previewed last year at Dreamforce, a broad outlay of Einstein generative AI tools as well as the Einstein 1 and Data Cloud platforms. Some have been released, while others are in beta or still in development.

In this interview, Rahul Auradkar, executive vice president and general manager of both Data Cloud and Einstein, discusses the interwoven nature of AI, data and security in the enterprise, and how Salesforce is building its platform with those features in mind.

Editor’s note: This Q&A was edited for clarity and brevity.

You’re head of both Data Cloud and Einstein. Explain how data and AI are so inextricably linked that Salesforce built its org chart around this idea?

Rahul Auradkar: We looked at data-driven and AI-driven customer experiences as two sides of the same coin to build Data Cloud. Einstein — what we’re doing in predictive AI — has existed since 2016. We do more than a trillion predictions per week. That number is only going up significantly, year over year. On the generative AI side, we have been shipping Einstein GPT products. We talked about copilots at Dreamforce [2023], which we’re going to have ready very soon.

Einstein 1 combines the data platform that we have, which made us successful, along with Data Cloud and all the things that come together. That’s why Data Cloud is an integral part of the Einstein 1 platform. We deliver AI use cases and data-driven use cases through our CRM service. Our partners do it, and our customers do it as well.

‘Customer 360’ has meant numerous things over the years — a philosophy, a bundle of products and whatnot. What does Customer 360 mean right now?

Auradkar: What it means is the ability for our customers to become customer companies. They have fully integrated and consistent experiences they provide to their customers across sales, marketing, service and commerce. Whether they want to provide analytics through Tableau or whether they want to find data integrations through MuleSoft, etc., the idea is that you provide a consistent and very rich integrated experience across all touchpoints, modalities and channels.

A good example would be our users providing service to their customers, who would definitely appreciate the fact that that business knows a lot about that customer. As in, I opened up a few tickets, you’ve already sent a bunch of marketing emails to me, I bought these seven products from you, I have had issues with these three products. My profile is a highly personalized profile, and this is my highly personalized view of what issues I’ve been having. That comes together using a combination of data and Data Cloud, which brings structured and unstructured engagement data to life.

Explain what a vector database is for Salesforce line-of-business users who might not understand — and why it is important to AI.

Auradkar: We continue to have unstructured data in Word documents or PowerPoint documents, etc. — all of that sitting on our hard drives or in the cloud. More than 80% of our enterprise customers’ data is unstructured data, whether it is voice transcripts, call transcripts, documents associated with contracts. Up until now, the only way that people have really brought unstructured data to life is by text search for similarities — you find a document.

With the advent of the LLM [we have] embeddings, a form of models within the LLM landscape. [This enables AI to] reason over unstructured data in a more semantic manner, to break up the unstructured data into chunks. Those chunks are the vectors we’re looking at.

[Think] about a cat, a dog and a bear. All three are fuzzy animals. So, there is one way of looking at them — they’re all fuzzy. But if you say ‘fuzzy and domesticated animal,’ you move away from a bear and you come closer to a cat or dog. When you move to ‘domesticated animal that has claws,’ you move away from the dog and closer to the cat. Those are all dimensions that allow you to come closer to that semantic meaning. Those are represented through vectors.

At the World Economic Forum meeting in Davos earlier this year, Salesforce CEO Marc Benioff said AI lies and hallucinates, but at the same time can be a helpful technology. Is it hard to get people to trust AI even though Salesforce does its best to build in transparency and security features?

Auradkar: You’re asking about the hallucination piece. In predictive AI, we have probabilistic methods — is there a propensity for something to happen? You have a threshold that if there’s something [negative or positive] expected to happen with a customer, I need to do something about it. I need to reach out or I need to have an alert going out to whoever’s managing that customer.

On the generative [AI] side, there is no deterministic response that we get. For example, I might ask a generative model to create a response to send to that customer. That may be quite different if I ask it to do it a second time — because it’s not deterministic. So, that makes it even harder, but all of that is a function of what you asked and how you asked it. That matters. The context matters and how you ground the context in the way you ask the question — that’s where the data, plus AI, plus CRM, plus trust comes in. Are you using the right data, in a trusted manner, to ground the question that you’re asking?

What can you tell us about what to expect later this year as far as the roadmap for Data Cloud and Einstein?

Auradkar: Data Cloud is our hyperscale data platform. Using that, we are providing a consistent integrated experience across sales, service, commerce and marketing. We’re bringing analytics to life, and we’re bringing AI to life as well. Now, in addition to that, the biggest thing we have done is bringing what we refer to as ‘trapped data inside of enterprises’ to life.

We are an open and extensible platform — we are not creating a data silo. We are not creating this concept of a data gravity and data silo in that if you have data sitting in Snowflake, Databricks, Azure or Redshift, we’re saying we’ll coexist with them. We’ll have zero ETL and zero copy from them, and we’ll bring that to life in the flow of work. We shipped integration with Snowflake, we’re shipping integration with Databricks, we’re doing the same thing with Redshift — we’re doing the same thing with a few other vendors as well. We have adopted open source standards right down at the storage layer all the way up to the semantic protocol layer. We want to extend that roadmap to more vendors, more integrations.

That’s one big area of our roadmap. The second big area is we’re going to have this pilot on unstructured data that I talked about earlier, where we can bring it to life.

The third area of big investment in Data Cloud is an integral part of the Einstein 1 platform. To that end, Data Cloud will be deeply integrated into our core platform, which then starts to light up some more use cases for service, sales [and] marketing. That is a substantial amount of work.

AI regulation is probably coming from the U.S., the U.K. and the EU. How are you preparing for that in your product planning?

Auradkar: The good news is that Salesforce has the Office of Ethical and Humane Use. We even have an executive vice president for that — Paula Goldman is the EVP. She [tracks] all of the regulations and [represents Salesforce in conversations with regulators] and monitors us for the ethical and humane use of AI.

We do expect that there will be ongoing debate at the public policy level. And we will have a strong voice in that public policy debate as to what AI really brings and what it means. Because we start with trust as the underlying principle, we probably are far ahead of any public policy in terms of providing trust to our customers. Any policy that comes, we will be influencing it — and we will be the first ones to actually step forward and say, ‘That’s right.’

Don Fluckinger is a senior news writer for TechTarget Editorial. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.