Einstein Studio 1: What it is and what to expect

To make the most of AI’s potential, companies need access to data, and for many organizations looking to capitalize on AI for their customer analysis, Salesforce is where that data lives. The company has been a leader in organizing the details of the relationship between businesses and their customers. And now that the AI revolution is here, Salesforce wants to be a prime player in how companies use AI to unlock the patterns and unfilled promises in their databases.

This isn’t a new push for Salesforce. The company has been bundling various forms of automation into its Einstein brand since 2016. This year, however, Salesforce has accelerated its agenda, integrating much of its recent work with large language models (LLMs) and machine learning into a low-code tool called Einstein 1 Studio. With this platform, Salesforce seeks to help organizations apply the cleverness of LLMs to the customer data they have squirreled away in Salesforce data lakes in the hopes of selling more.

Following is a look at Einstein 1 Studio and what companies can expect in putting it to work in leveraging the power of AI.

What is Einstein 1 Studio?

Einstein 1 Studio is a set of low-code tools to create, customize, and embed AI models in Salesforce workflows. The platform is composed of three major parts: Copilot Builder, Prompt Builder, and Model Builder.

Copilot Builder enables you to customize Einstein AI for tasks specific to your company. It has the potential to create new actions to integrate the AI into your workflow.

Prompt Builder lets its users iterate the various text prompts fed into Einstein’s LLM. This is where marketing teams will probably spend much of their time, as finding the right prompt to generate the optimal messaging to customers is very much a combination of art and science. 

For teams that want to boil down their own data into predictive tools, Model Builder will turn all those records of past purchases sitting in the data lake into a big statistical hair ball of tendencies that passes for an AI these days. Model Builder lets you both train new models and work with existing models from other sources.

All are designed to be worthy of the label “low code,” a category of tools that aim to accelerate application development by enabling “citizen developers” to leverage the tools’ intuitive interfaces for building new tools and apps without having to be professional programmers.

Of course, the meaning of “code” is shifting rapidly as AIs enter the workflow. There’s little need to fuss over classical coding issues like misplaced semicolons or uninitialized variables. But there are still plenty of challenges, like trying to understand just what the models are capable of doing — and where they fall short of the hype. Also, keeping the data clean and consistent is an endless chore.

What can you achieve with Einstein Studio 1?

Einstein Studio 1’s three new tools add another layer of automation to what teams have already been doing with Salesforce. The company has been aggregating data about sales and customers for years so that humans can connect with customers with better precision and accuracy. Now these tools make it easier to unleash the power of LLMs to handle more of what humans used to do, such as write personalized messages or respond to simple questions.

The biggest advance is probably the ability to automate much of a company’s contact with its customers. Marketing teams can use the Einstein 1 Copilot to create personalized marketing campaigns based on all past interactions detailed in their data lake. One predictive AI will try to guess what a customer wants to buy next, and then the other will write an email filled with references to past purchases and, if marketing wants, include a customized coupon code. In the past, much of this drafting might have been done by humans. Now with Einstein Studio 1, the AIs use your prompts to generate emails to any customer who might fit a sales profile.

Of course, it may be better to use the word “semi-automate” because Salesforce is trying to keep the safety brakes engaged just a bit. The company’s announcements around the technology push verbs like “integrate” and “iterate” more than the idea of taking a nap while the AI does everything.

In the past, the part of Einstein labeled “AI” was more for data analysis and prediction. Now, it connects foundational LLMs with your data to produce new text. In addition to speeding up work for the marketing team, it also means no more boilerplate emails, as the AIs will create semi-customized solutions based on past behavior.

Where will Einstein Studio 1 work best?

Einstein Studio 1’s approach works best in scenarios where the enterprise has already found ways to move lots of behavioral data into the system. It does better when details such as past purchases and past browsing are carefully collated and squirreled away in the data lake so they can be fed into the AI. Enterprises that have done all the hard work of cleaning up data and building a harmonious data lake with consistent fields will be all set to monetize.

One of the real challenges for marketing teams will be finding the sweet spot between boilerplate sales literature and creepy messages filled with too many personalized facts. Salesforce is pushing the idea that Einstein 1 is a vehicle for experimentation and iteration. The messages go out and, ideally, IT has already arranged for the details of any completed sales to be piped into the data lake so the results can be analyzed.

I think many of us dream of a world where we can push a button and go to lunch while the AI just generates oodles of sales automatically. This probably won’t be the case most of the time. Marketing teams will end up doing plenty of experimenting and tweaking of prompts. Sure, the basic messages will be written by the LLM, but the humans will still be fussing over the mechanism, adjusting the data flows, and worrying about the connection between the data and the LLM.

How will Einstein Studio 1 impact the skills equation?

One interesting question will be how much value can be found in fine tuning the LLMs versus just working with the foundation models. Salesforce suggests it is better to put the time into creating a very detailed and elaborate prompt. Instead of staying late to write individualized emails to potential customers, the marketing team will be working a full day fiddling with the prompts and trying to get the right messages for the right customers.

Presumably, this will give marketing teams the opportunity to become experts in prompt engineering because much of their work will be to fiddle with just the right collection of words and past purchases to be bundled and fed into the LLM. The workload of marketers will become much more editorial, as the sales prompt engineers will in effect be crafting meta instructions for the machines.

What about privacy?

Salesforce is working to address many of the privacy concerns users, customers, and CIOs have over the use of AI. For example, It has created what it calls the “Einstein Trust Layer” to protect customer data. CIOs who worry about letting customer data out of their control can feel better because the Einstein Trust Layer is supposed to contain all a company’s data in one private corner.

In this way, Salesforce intends to keep customers’ personal information from leaking out and comingling with another company’s data. Moreover, they want to avoid a company’s sales details from becoming training fodder for the foundational LLM. To that end, Salesforce insists AI companies make a commitment not to use the data for any general training.

The bottom line

In essence, with Einstein Studio 1 Salesforce wants to handle most of the rough chores of dealing with AIs, LLMs, and the vendors that sell them. Einstein 1 Studio handles the piping so the data from your Einstein 1 platform instance will flow smoothly into the AI. And for enterprises, all the time and effort they put into building an elaborate sales data lake will start to pay off.

Where the energy will need to go is in crafting the brand and messaging so it’s consistent and focused. AI is still a new and quickly evolving field. Einstein 1’s latest additions are meant to create a place for experimentation and iteration. The data is there. The low-code tools are ready to deliver it. Now it’s up to the sales team to start turning into prompt engineers and see who buys this new messaging. And, of course, the real challenge is building a product that people want to purchase. That’s a completely different task that Einstein 1 isn’t designed to solve.