Einstein 1 is a relaunch of the Salesforce platform to create a trusted AI platform that Salesforce customers can use with confidence. The key to Einstein 1 is Einstein Copilot, the built-in side panel of the user interface we’ve all become used to. Einstein Copilot Studio is made up of three components: Prompt Builder, Skills Builder, and Model Builder.
We’ve found ourselves in the middle of a data and AI revolution. Salesforce believes that they are pioneering the way for businesses to bring data together, and apply that to AI.
However, there are challenges to overcome to reach these AI strategies. The majority of organizations may not have the platform that they need, but instead have different applications, custom built data lakes, separate APIs – all of which are disconnecting their data.
What organizations need is one platform that’s capable of connecting their data. This doesn’t necessarily mean that all data needs to be in one place, but there needs to be a connection. In Salesforce’s eyes, it has to be low-code (democratizing the technology), and also open, to enable different providers (data, LLMs, and ISVs) to support Salesforce customer organizations.
What is Einstein 1?
The Einstein 1 platform brings together a suite of tools that enables you, as a Salesforce professional, to bring AI into your users’ workflows – while guarded by the Einstein Trust Layer that gives your organization peace of mind when it comes to data privacy and security.
It is integrated, intelligent, and automated – but what does this actually mean? Hopefully, by taking you on a tour of the layers that make up Einstein 1, you will see how Salesforce are meeting organizations where they need/want their AI to be in this AI flux.
Einstein Copilot (For Users)
Einstein Copilot is built-in in the side panel of the user interface of any Salesforce application. Internal users, or customers (via Experience Cloud portals, or similar), can interact with this conversational AI assistant in natural language – that is, how a human user types, as they would in conversation with another human.
Copilot goes further, offering what’s similar to the ‘next best action’ we’ve become used to, served up by Einstein since circa 2017 – except this time, the recommendations are in the form of multi-step action plans where users can select/de-select the follow-up actions to take forward.
Admins, this is what Salesforce are delivering so your users can craft their own prompt all safely and securely, as you would have configured Einstein Copilot Studio accordingly.
Einstein Copilot (For Admins)
Now that we’ve seen it from the perspective of the users (the polished product, you could say), it’s time to explain the levers that are operating behind the scenes that Salesforce professionals, such as Admins, can ‘pull’.
Copilot Studio gives you both power and control – the power to implement impactful generative AI to support your users’ productivity, while having the control over which processes generative AI is used, and the extent that the prompts should reach.
For example, a user may ask Einstein Copilot a question, and not receive an answer. Here, we should question: was the user entitled to access that information? If yes, you will benefit from the Skills Builder we’ll cover soon.
Overall – this technology also puts the information right in your hands, as the admin, to avoid the feared negative consequences of generative AI, such as toxicity, bias, and hallucinations.
Prompts are simply instructions for large language models (LLMs) – the more context you provide to a prompt, the better the generated output you’ll receive.
Prompt Builder will enable admins to create prompt templates, choose how they want to ‘ground’ the prompts in Salesforce data, and activate for users quickly and easily. In simpler terms: build, test, and deploy generative AI prompts for your users.
What’s more, is a new-found importance placed onto the unloved ‘description’ field. This seemingly superfluous field that you’ll find on fields, validation rules, custom report types (and much more), may seem like a chore to fill in even when it is for the benefit of your team/successors. Now, this field actually serves as context for Einstein Copilot Studio to generate a prompt (in Prompt Studio) that’s more aligned with what you need – you could say that the description field is giving you a solid head start!
When you’re ready to test, using clicks in Prompt Builder, you can pull in instructions. While it looks like code (because it is JSON), this is only just the context that’s been embedded in the prompt you’ve instructed.
Having done some considered tweaking with Prompt Builder, the output generated is better, i.e. pulling in more context from our organization’s data.
As we have seen, a user may ask Einstein Copilot a question, and not receive a quality answer (if any answer, at all). If the user is entitled to access that information, then they should be granted access.
Should is the keyword here. You would only grant the power of generative AI to the users who are using the Salesforce platform (and any trusted, integrated platforms) for defined use cases.
When the should is a definite yes, then you can grant access to generative AI (on the Salesforce platform) via ‘skills’. For example, this enables you to say: for process A, user X should have N generative AI capabilities, whereas user Y shouldn’t have the same outcome.
You can enable checklists for users – like recommended actions that could take the form of multi-step action plans.
However, if a user asks Einstein an additional question, in some instances, it could appear that Einstein doesn’t have the skill to answer that question. For organizations, being able to extend the ‘Copilot’ functionality can be important. This extension is bolting on custom skills via Copilot.
‘Skills’ is a concept that, personally, I found tricky to grasp. In simple terms, it involves an admin/other similar professional, embedding a ‘skill’ into a Salesforce Flow.
Finally, an additional part of the Einstein Copilot Studio is the Model Builder. This provides customers with the flexibility to select their own AI models.
This includes the ability to select one of Salesforce’s proprietary large language models (which Salesforce have been training for years), or one of their preferred partners, including Anthropic, Cohere, Databricks, Google Cloud’s Vertex IX, or OpenAI.
Additionally, Salesforce announced that they have partnered with AWS to allow you to bring your own data models to the Salesforce platform. This includes “Bring Your Own Lake” model or “Bring your own large language model” using Amazon Bedrock.
This is all entirely supported using the Einstein Trust Layer, ensuring that sensitive and proprietary company data is masked, and therefore, unable to actively train LLMs.
We’re living in a historical moment right now for AI – we’ll remember this day years from now.
No surprises that the key piece of news from Salesforce’s flagship conference, Dreamforce, is centered on AI and Data Cloud coming together. The tagline for the keynote, hosted by Founder, Chairman, and CEO, Marc Benioff, titled: “Now everyone is an Einstein – Data + AI + CRM + Trust”.
Salesforce’s flagship conference, Dreamforce, is taking place this week (September 12th-14th, 2023). This year, it comes with impressive statistics, being the largest AI event in the world, with 1500+ sessions, 40,000 attending in-person, and millions of viewers online.