In our “What is Salesforce AI?” post, we introduced big picture AI concepts that laid out the overall architecture and foundation for Salesforce’s AI offerings. In this post, we are delving deeper into the Salesforce Artificial Intelligence Cloud, covering its functional product areas, intelligence capabilities and any potential limits and pitfalls. We are also covering the Salesforce Machine Learning Enabled suite of APIs and tools that will help your business transition into the adoption of AI for your day to day business needs.
Salesforce AI Cloud
The Salesforce AI Cloud is what most users now think of as Salesforce AI. The Salesforce AI Cloud consists of two key areas:
- A set of AI-enabled functions/apps/extensions built into the various Salesforce clouds
- A set of low-code tools, called builders, that allow different audiences to develop AI-based models, low-code applications, and AI-based chat assistants.
AI-Enabled Functions
Salesforce AI provides a set of AI-driven plugins/extensions (Salesforce calls these apps) that add functionality to each of their clouds. These apps are able to pull in Salesforce data for learning, analysis, and potentially data enrichment through LLM providers.
Figure 2 shows an example of the Salesforce AI-based apps added to Sales Cloud. Salesforce has leveraged underlying Einstein APIs to develop these applications just as any third-party developer on the AppExchange would, which is why Salesforce uses the term “apps” to describe core AI apps that enhance Salesforce Cloud functionality. In the end,they are extensions of core cloud functionality, which is why we use the term “functions” to describe them collectively.
Many of these plug-in apps use true AI capabilities, like Sales AI or Einstein Conversation Insights shown on the top right of the list. Others, like Sales Analytics, use more traditional predictive Machine Learning models, which are statistical models that can be used to make a prediction on the future based on previous values. Then others use AI to do something we discussed in our last post and which we said we would come back to as we delved into Salesforce AI – orchestration.
Orchestration is the act of taking individual manual or automated tasks and turning them into an organized, repetitive workflow. While orchestration is conceptually a separate capability from traditional AI, AI has made orchestration easier to accomplish and more impactful. Now, instead of having to write code to interconnect applications or apps within a larger platform like Salesforce, a non-technical user can talk to an intelligent assistant and instruct it on how to build a step-by-step repetitive process that will make them more efficient and prevent them from having to do mundane tasks manually over-and-over again.
A high level example would be to ask AI to generate a weekly report on a region’s sales summary and automate the task of sending out an update to relevant sales directors and various relevant stakeholders, AI is able to interpret the task it has been given and is able to carry out the task of making it happen.
Many of the Einstein-enabled apps are designed to make it easy to build critical orchestrations that will make various teams, in this case sales, extremely efficient. An example is the Revenue Lifecycle Management app, which allows sales managers, who are for the most part not deeply technical, to create orchestrations for product catalogs, pricing, configuration, quoting, order management, fulfillment, billing, and contracts. Salesforce states that the primary benefits of the efficiency-driven: “Reduce operational costs by automating policies, controls, and processes at scale.”
Builders
Salesforce’s AI Builders are a suite of AI-enabled, prompted tools that allow non-technical individuals to extend or customize the capabilities of Salesforce AI. There are currently three announced builders in the Salesforce AI platform:
- Model Builder. Model Builder provides a single place to manage custom ML- or AI-based models in Salesforce. It includes a no-code tool for writing predictive models, as well as tools that allow individuals to bring their company’s in-house built custom models into Salesforce AI. By bringing in pretrained custom models, organizations can leverage their tested, trusted and well understood models with little to no customization, reducing hassle, setup time and potential breaches of sensitive data.
- Prompt Builder. Prompt Builder helps individuals create structured prompts that team members can use to complete Salesforce tasks. One common example shown on Salesforce’s marketing material a structured, prompt-driven interface to help customer service reps provide fast, more accurate, and more consistent technical support, allowing teams to better serve their customers
- Copilot Builder. Copilot Builder allows customers to customize Einstein Copilot for their specific needs.Copilot functions as a conversational engine that acts as an AI assistant that is able to interpret questions and instructions. Responding with clear, trusted, and actionable information grounded on data within a customer’s Salesforce instance.
Figure 3 shows an example of the Prompt Builder template tool.
All this talk about Salesforce’s impressive line of AI powered functions, builders and virtual assistance begs the question: How do I transition my business, with all of its potentially out of date data paired with complex business processes to leverage the latest and most exciting cutting AI technology? In most cases it starts with clean, accurate data. In the end, data is core to the reliability and accuracy of AI. Data that is outdated, inaccurate or with multiple duplicates can cause AI models to hallucinate, make wild predictions or generate information that is grounded in false truths.
No matter what’s coming in the future, if your data is inaccurate, AI will generate incorrect information that is dangerous and harmful to the day to day operations of a company. It is important to start from the ground up; by building a strong foundation starting with your clean, accurate data, your business will be in a much better future position to leverage and stay ahead in the ever exciting and constantly transforming field of AI.
Salesforce ML-Enabled APIs
The key top-level element of Salesforce AI for developers are the Salesforce ML- and AI-enabled APIs. These are for their use in developing their internal apps (if a Salesforce developer) or for third-party developer partners to leverage for their apps and applications. There are 45 Salesforce APIs available to developers. Of these, only six clearly use machine learning or Salesforce’s AI capabilities:
- B2C Commerce Einstein API
- Einstein Bot API and SDK
- Einstein Prediction Service REST API
- Einstein Discovery REST API
- Einstein Vision and Language APIs
- Net Zero Cloud API
The rest are more process or data-movement intensive and do not seem to use any significant amount of machine learning.
B2C Commerce Einstein API
The B2C Commerce Einstein API is the recommendation engine that provides personalized recommendations to shoppers based on their past shopping and browsing history. Developers can leverage this API by sending payloads to the API containing customer interactions that can be
Einstein Bot API and SDK
A bot is an autonomous program that can interact with systems or users. The Einstein bot API and SDK is a RESTful API that allows developers to build their own AI-based bots and then build workflows and orchestrations around them that can be accessed from any device.
Salesforce AI has a bot builder tool that speeds and simplifies the development of bots for the Salesforce platform (Figure 4)
Einstein Prediction Service REST API
The Einstein Prediction Service REST API allows developers to programmatically run predictions, manage prediction definitions, and manage models. After deploying models with Einstein Discovery, developers use the Einstein Prediction Service REST API to embed predictions into their websites or applications. It is the foundation for the Einstein Discovery REST API
Einstein Discovery REST API
The Einstein Discovery REST API leverages the Einstein Prediction Service via its API to develop and manage prediction models. It then leverages these models via the Connect REST API to integrate these models into data sources, web sites, applications, or apps where they are then used to use predictive capabilities to provide personalized responses to user behavior on their website or in their applications.
Einstein Vision and Language APIs
The Einstein Vision and Language API allows developers to build natural language capabilities into their applications. The API can perform sentiment analysis, intent analysis, and named entity recognition. However, this API is scheduled to be retired in May 2024.
Net Zero Cloud API
The Net Zero Cloud API uses deep learning tools to help companies predict their carbon footprint. It uses global emissions factors to calculate greenhouse gas emissions. It allows organizations to collect, categorize, analyze, and report energy usage data throughout their business activities.
Feel free to reach out to us at info@datagroomr.com with any questions you may have about Salesforce AI. As experts using AI to build DataGroomr products, we stay on top of the latest developments in Salesforce’s AI platform and would be happy to share our knowledge with you.
