At Dreamforce 2023, Salesforce promised a mix of Data, AI, CRM and Trust. It delivered the promise far more, where many vendors discuss producing generative AI solutions for business use cases. Salesforce did so earlier this year. The giant SaaS vendor has announced the Einstein 1 Platform.
The new platform is fully integrated into the Salesforce Data Cloud. This means that any new generative AI solutions have real-time access to transactional data from Salesforce applications and any other data ingested into the Data Cloud via MuleSoft.
Einstein 1 Platform
Patrick Stokes, EVP of Product and Industries Marketing, Salesforce, stated, “We’re excited to announce the Einstein 1 platform, a relaunch of our Salesforce platform to create a trusted AI platform for customer’s companies.
“The Einstein 1 platform is integrated, intelligent, automated, it is and always has been low code and no code. And most importantly it is an open platform to support many, many different data providers, large language model providers, and independent software vendors.”
Underpinning Einstein 1 is a metadata framework that allows organisations to leverage Einstein across all Salesforce applications, including Sales, Service, Marketing, Commerce and Analytics. Customers can use Einstein Copilot, which delivers a conversational AI assistant that any application can leverage.
Salesforce has enhanced this with Einstein 1 Copilot studio. It enables organisations to deliver their own use cases. Building what Salesforce calls prompts to enable the AI to elicit and provide customers’ required information.
Parker Harris, Co-Founder and CTO Salesforce, commented, “A company’s AI strategy is only as good as its data strategy. We pioneered the metadata framework nearly 25 years ago to seamlessly bridge data across applications. It’s the connective tissue that fuels innovation.
“Now, with Data Cloud and Einstein AI native on the Einstein 1 Platform, companies can easily create AI-powered apps and workflows that supercharge productivity, reduce costs, and deliver amazing customer experiences.”
To kickstart Data Cloud interest, Salesforce announced it is offering free access to Data Cloud and Tableau. With Enterprise, Unlimited and Unlimited Plus editions, customers can unify up to 10,000 profiles and start exploring their data with two Tableau Creator licenses. There are restrictions, though these were not made clear.
AI a threat or an opportunity
During the Dreamforce keynote, Marc Benioff, Co-Founder and CEO of Salesforce, gave a passionate and informative keynote that flagged the issues with Large Language Models that most enterprises are concerned about. He highlighted that Large Language Models improve their knowledge and profits at the expense of public domain data and are also looking to gather corporate data.
According to Benioff, this is not and never has been the Salesforce approach. He summarised it by saying, “We want to build the trusted AI platform for customer companies. That is our mission!”
Salesforce also ensures that security and trust underpin everything within Einstein. While Benioff added, “We think trust is the highest priority”, he also noted that Salesforce was one of the first organisations to consider that the tenets of AI include trusted ethical and humane use. He expanded on this, adding that in the Salesforce belief is:
- “Your data isn’t our product
- You control access to your data
- We prioritize accurate, verifiable results
- Our product policies protect human rights
- We advance responsible AI globally, just as we always have
- Transparency builds trust.”
This is a mature approach from Salesforce meant to protect against a future where AI becomes dangerous. Benioff noted, “We all know what could happen if this AI goes really wrong.” While the threat is some time away from reality, mitigating it now is important.
Is Benioff hyping up the threat? He is probably correct to highlight it without this compliance wrap-around AI. There is a risk that it can produce poor, if not incorrect, results that will lose customers from those using AI. Also, with the long-term prospect of general intelligence, now is the time to add the guardrails, even though it seems like retrofitting this to AI.
The Einstein 1 Platform enables that with a comprehensive trust layer ensuring data flow to LLMs. Suppose a prompt initiates, as AI follows the data, that to provide that answer, data must be extracted from relevant data sources. In that case, That data must be grounded and synchronised using the metadata framework. While the data must be passed to LLMs, personal data and PII must be anonymised, ensuring the data masking can be reversed later.
The data is then passed through a secure gateway to the most relevant LLMs. This is where Salesforce has built partnerships and invested in several AI platforms, including AWS, Google, Cohere and Anthropic. Those LLMs won’t keep that data but will pass back the answers to the Einstein trust layer, where that data is validated for hallucinations and toxicity.
The Einstein Trust Layer enhancements will be generally available in October 2023 and included in Einstein products.
Getting the right results from AI
Alongside the Einstein 1 platform, Salesforce also announced Einstein Copilot. This out-of-the-box conversational AI opens up the use of generative AI across every application. However, not all are generally available immediately.
The Copilot is both reactive and proactive. It will, out of the box, provide several use cases for customers across Sales, Service and Marketing clouds and more. For example, it could answer questions posed in natural languages or provide actions following the automated transcription of a sales call.
Salesforce is taking the copilot idea beyond others by creating the Einstein Copilot studio. This will enable organisations to create prompts to develop the capabilities of their LLM to answer questions relevant to their company rather than just generic solutions. Customers can automate answers across multiple communication channels, including Slack, WhatsApp and SMS.
The Einstein metadata framework is polymorphic and ensures that data mapping and additions replicate across the lifecycle of the AI process. So that fields within the application are understood in context within the Einstein AI element.
Customers are enthused with the early developments by Salesforce on generative AI. Chip Aubry, CTO of Independence Pet Group, commented, “The way we understand and serve pet owners is quickly evolving, and so are their expectations of us. By embracing Data Cloud, we’re unifying data from across our brands to transform our customer engagement.
“Leveraging this trove of real-time data, we’re poised to harness generative AI to help improve our operations, make smarter business decisions, and deliver highly personalized offerings that truly resonate with our customers.”
Einstein Copilot is currently in pilot, with no date for general release, and Einstein Copilot Studio will be in pilot in Fall 2023.
Enterprise Times: What does this mean
This was a solid keynote from Salesforce. While there was a nod to the philanthropic initiatives that Salesforce continues to make, it was clearly aimed at Enterprise businesses and aligned with the investors who raised concerns a year ago. Benioff has redirected Salesforce skilfully over the last few months. He reiterated clearly that the company is here to deliver value to customers.
Benioff thanked the customer base for its support over the last year. He offered, as usual, the promise of more to come on AI, the hottest topic in the board room.
Over the last few years, Salesforce has continued to iterate its platform. This is not a completely new platform but an iteration that adds to the foundation, with the trust layer and the customer-facing levels with the Co-Pilot studio. What will be interesting is the take up amongst customers. The appetite for generative AI solutions is strong. Salesforce appears to be putting the power into the hands of business users in a trusted, ethical, and humane way.
The big question is whether customers adopt it and whether Salesforce will be able to monetise this to increase revenue. Or will generative AI become table stakes?