Salesforce TrailblazerDX 2024

It seems like vendor announcements for new virtual assistants, particularly named “copilot” are a weekly occurrence. Not all copilots are the same though. I spent a couple of days last week attending Salesforce TrailblazerDX, its annual developer conference in San Francisco. The event was, to nobody's surprise, focused on AI. There were some interesting announcements and some good discussions with company executives and developers. Salesforce has some unique features and approach to its platform, AI in general and the new Copilot offering. The first thing of note is that I used the singular Copilot, not plural like Microsoft. This is a good deal more significant than you might think.

Einstein 1 Platform

Before we dive into Copilot though, it's useful to examine the overall platform strategy that Salesforce has rolled out over the past 12 months or so. This diagram helps explain the new Einstein 1 platform:

Several important changes are represented here over the old Force.com platform that you might remember:

• The first significant change you will note is that all the applications are running on Einstein 1, including many of the acquired products. All apps that are on the platform share the benefits of any embedded features including Copilot.

• The second big change is the combined Data Cloud. The new Data Cloud is an evolution of Salesforce CDP (customer data platform) and is now the totally integrated data and metadata layer for all the Salesforce apps / clouds.

• Data Cloud supports both structured and unstructured data. This means that all enterprise data can be accessed either directly or indirectly through federation (for Snowlake, BigQuery, Databricks (pilot), Redshift (pilot)) that allows other non-Salesforce data to reside in place, but be available in the Data Cloud.

Federated Data

Federated data, or data federation, is a concept in data management that involves integrating data from multiple sources to appear as a single cohesive dataset without physically moving or duplicating the data. This approach is crucial for businesses and organizations that deal with large volumes of data stored in various locations and formats. Federated data enables more efficient data management, analysis, and decision-making by providing a unified view of disparate data sources.

Understanding Data Federation

Data federation is a software process that allows multiple databases to function as one through a virtual database. This virtual database takes data from a range of sources and converts them to a common model, providing a single source of data for front-end applications. It is part of the broader data virtualization framework, which includes additional features such as metadata repositories, data abstraction, and advanced security measures.

Key Benefits

• No Extra Storage Space Required: Since data federation integrates data virtually, it eliminates the need for additional storage systems, saving businesses time and money.

• A Single Source of Truth: Data federation ensures that the most recent and updated data is available, reducing errors and improving reliability.

• Removal of Data Silos: By integrating data from multiple sources, data federation breaks down silos within an organization, facilitating easier data sharing

Of note here is that integrated data availability and overall data quality is the foundation to a successful artificial intelligence (AI) strategy. Data Cloud also includes vector databases and semantic search that are used in AI to hold and isolate company data to prevent leakage into a large language model (LLM) yet still make the AI function with critical company context. This is a very important distinction of enterprise AI versus consumer AI like OpenAI’s ChatGPT and Anthropic’s Claude 3, that do not and should not have access to unique company data.

It’s also important to call out the Salesforce Trust layer in the platform. If you’re interested it is covered in this episode of Disambiguation from Dreamforce last year and Salesforce’s overall approach to grounding AI in this episode.

Salesforce Copilot and Einstein 1 Studio

Salesforce is approaching virtual assistants and chatbots differently from other providers by providing a single Copilot and Copilot Studio across all Salesforce applications. This means a single experience across all organizational roles and both internal assistants and external chatbots. I should probably repeat this part, since it is very unusual, Salesforce Copilot is both an internally and externally facing chatbot / assistant.

All Copilot configuration and management is accomplished through the Copilot Studio by Salesforce admins. The Studio has the ability to:

• Configure prompts (Prompt Builder)

• Manage, assign and customize Copilot “Actions”

• Manage models (Model Builder)

• Ground the AI (retrieval augmented generation or RAG and semantic search)

• Provide governance for the AI

The concept of “Actions” is the key to using the Copilot across all applications and internal and external use cases. There are prebuilt actions included, but the admin can also create custom actions based on the specific business use case. The Actions can use Flow (Salesforce workflow) and Apex code as well. Actions are assigned to a specific type of Copilot, and then the Copilot is made accessible to specific roles. In other words a sales Copilot has different actions than a marketing Copilot, etc.

Benefits of Einstein 1 Studio:

• Single experience to configure, customize and manage all Copilots.

• Admins only need to train on Studio to administer all assistants and chatbots.

• Prebuilt Actions enable many use cases with no coding.

• For advanced use cases custom Actions can be constructed using no / low and pro code methods that can incorporate custom workflows and include custom Apex code.

• Ability to build your own language models or connect to existing models.

• RAG and semantic search to provide secure and protected access to company data in a vector database in Data Cloud

• Monitoring and governance for all Einstein AI

Benefits of Copilot:

• Access to all the data in Data Cloud, with access controlled by the Salesforce admin.

• Single bot for all internal and external use cases

• Fully customizable

• Context sensitive to the user and the app in use

• Learns from user over time

• Can recommend actions and/or take approved actions for the user

• Access to company data securely to protect privacy and against data leakage while enabling effective business use of the Copilot