Understanding What Generative AI Means for Salesforce Admins

Understanding the fundamentals of artificial intelligence (AI) is critical to being an #AwesomeAdmin. This technology has opened up broadly and both people and businesses are finding new and inventive ways to weave it into products and our lives. There’s no better time to get up to speed on AI and how you will be able to use it.

AI is already part of your life

While AI has been thrust into the spotlight thanks to the release of popular products like ChatGPT, it’s good to note that many of the roots and interactions of generative and conversational AI have been hallmarks of our digital lives for quite some time. For instance, consider how a search engine knows how to complete searches for you.

To do this, the engine relies on a vast well of data to do two things: 1) Determine the phrase you’re forming and 2) Apply a predictive model based on how many people finished that phrase. It’s like when someone finishes your sentence for you, but generally less annoying and more socially acceptable. This same concept of trying to predict your next outcome, or more specifically, your desired outcome, is now embedded in how we navigate the internet, shop online, and communicate with each other.

Salesforce is no different. When you’re having a conversation with an Einstein chatbot, it uses tech called natural language processing (NLP) to read and write in a human-friendly way. Likewise, Prediction Builder uses predictive AI to perform its magic on custom fields.

So why is AI having its inflection point now?

The simple answer? Data.

Lots and lots of data.

For the longest time, we relied on relational databases as the best way to connect data points together. And of course, they’re still the bedrock of much of the technology we use. However, there has been a decades-long slow adoption of neural networks to solve problems. Unlike a relational database, which is organized in rows and columns, neural networks layer on interconnected nodes of data. This is particularly useful for certain kinds of solutions and one in particular is NLP. When you talk to a digital speech assistant or certain chatbots, they aren’t analyzing your text as a singular point of data—they break down the nouns and verbs into tokens, which a neural network can read like language. As a result, they can respond like language. This opened the door to having artificial but somewhat lifelike conversations with computers.

Enter large language models (LLMs), or more generically called foundation models, to help build conversational and generative AI models. Now when we say large, we mean enormously, vastly huge. The original GPT-1 data size held 117 million parameters. So, let’s represent it with the size of the moon:

GPT-3.5, the currently public model for ChatGPT, has 175 billion parameters—which makes it the size of the entire Earth:

This allows for a much more capable, more advanced, and more accurate AI to converse with.

There are pitfalls…

While this technology is very exciting with lots of possibilities, there are limitations and even concerns. When you chat with a product like ChatGPT, it will often warn of them, but these are the three big areas:

1. Inaccuracy: These models are designed to give you an answer even if it has to fill in the gaps. Those gaps might just end up randomizing the answers, but occasionally they respond with outright misinformation. They’re not designed for use without human oversight at this point. Take what you get as a response with a grain of salt.
2. Timeliness: You know how we compared the ChatGPT model to Earth? It’s more like Earth…in 2021. It’s a snapshot of data from a moment in time. So new events and information won’t be baked into the model.
3. Ethics: Again, factors like bias are still better left to humans. AI might offer a great marketing campaign without realizing it has ties to the Holocaust. And yes, that actually happened. There’s a longer and more complicated conversation to have on ethics and AI, which we will cover at a later date.

How can I become an AI-minded admin?

Being an AI-minded admin isn’t as hard as it may seem. It starts with understanding the key principles of AI and applying them to deliver the future of business at your company.

Understand AI and machine learning fundamentals

There’s so much to learn about AI—and you don’t need a PhD in AI to start learning. Learn the basics of AI, starting with a few key concepts:

AI: The concept of having machines “think like humans”—in other words, perform tasks such as reasoning, planning, learning, and decision-making. AI can amplify our capabilities and provide actionable insights.

NLP: The development of algorithms and techniques that enable computers to understand, interpret, and generate natural language in a way that’s meaningful to humans.

Machine learning: The type of AI that leverages data to generate predictions. The more data we feed into a machine learning system, the better it can predict.

Deep learning: The type of machine learning that uses complex algorithms (also known as neural networks) to learn and perform tasks without explicitly programmed step-by-step instructions.

Language models: Deep learning algorithms that can summarize, predict, and generate text based on massive amounts of data. Using these complex algorithms, deep learning processes data to make suggestions and decisions based on context.

Dedicate time every week to growing your AI knowledge. Whether you’re reading blog posts, completing Trailhead modules, or getting hands-on with new AI tools, the knowledge you build today will give you a boost as more tools and technology are developed.

Salesforce AI features

Learn the basics of Salesforce Einstein and head to Trailhead to learn more about AI-powered solutions:

• Einstein Bot Basics
• Filter-Based Predictions with Einstein Prediction Builder
• Einstein Next Best Action

Deliver the future of business with AI + data + CRM

Salesforce is looking for ways to weave AI into the CRM experience for users, developers, and admins. In the early ’90s, Bill Gates framed a pivotal moment for the internet. He said that one day we wouldn’t want to reach for a phone book, because we would just ask the internet. We are currently moving from searching for information to asking for information. That might seem like a slight difference, but there are three big differences. One is that instead of combing through various websites for the best answer, you will just receive the answer. Second, since it’s a conversational model, you can ask follow-up questions and fine-tune that answer. Third, that answer can generate an actual solution.

So if you wanted to build a new flow in Salesforce right now, you would search for the information and solutions on how to build the flow, and then you would use that knowledge to craft a new flow.

Soon you will be able to ask for a flow based on business requirements and then go to that flow, test it, maybe tweak it, and then send it over to your users.

However, you don’t have to wait for Dreamforce to get into the AI game. ChatGPT has a free version and Microsoft is opening it up via their Bing model. Google just recently opened up its conversational AI, Bard, to the public. Now is the best time to start seeing where these AI models can help you in your Salesforce life. Here are some example prompts you could try to see the type of solutions ChatGPT can help you craft:

• Write a friendly email text when a contact gets converted to an opportunity.
• Write a SOQL statement that shows me users with API access.
• Give me five slogans for an onboarding campaign for new Salesforce users with my financial company.

We’ve got a lot more coming on AI, prompt writing, and product announcements. Everyone has the opportunity to learn and familiarize themselves with these concepts. Stay up to date on AI by checking out the Salesforce Admins Blog and visiting the admin AI resource page.

Resources

• Salesforce Admins Site: AI Page
• Salesforce Admins Blog: Be an AI-Minded Salesforce Admin
• Salesforce AI Research Blog: The Age of Conversational AI

Note: This post was updated on November 6, 2023, with the latest information and resources.