Marketing AI Agents101: What You Need to Know Right Now

Are you a marketer and feel like you have data figured out because you have a customer data platform (CDP)? Sorry, but you have to start all over again. I say that half-jokingly, as the truth is marketing has entered a new era of agentic AI. CDPs helped us bring our data together to create a unified customer profile. Now agents can incorporate not only CDP data but also data not typically associated with a CDP – and act on it. But this requires marketers to take a beginner’s mindset to understand AI agents and the data requirements to make them work. Think of this blog as a marketing AI agents 101 to help you get the basics down.

The basics of AI agents

I may have scared you with that intro, so let’s take a step back. No, you don’t need to be an expert on AI agents, but you do need to know the basics. The human needs to teach the agent — it doesn’t just learn on its own. Think of the agent as a super-smart intern. It’s very capable, but you as the marketer have to inform it with good data to help it do the job right. 

You’ve probably heard of Salesforce Agentforce. Agentforce takes AI from being reactive to proactive. Imagine being able to rely on an agent to assess a situation, reason, and then take appropriate action, especially on fairly tedious tasks. A digital process that can feel like “pushing paper”, but one that took some initial assessing, is ripe for being assigned to an Agentforce agent. 

Interactions that can take many shapes or forms and result in a variety of actions can be managed by an agent to give a customer a more personalized experience or reduce the workload on internal teams. This is because Agentforce agents, unlike LLM chatbots, take on a role and then act on it while respecting human-defined guardrails.

Let’s first take look at five traits that make an Agentforce agent:

• Role: An agent’s purpose. This defines the job to be done and the broader goals the agent should achieve on your team.
• Knowledge: The data an agent needs to be successful. This could include company knowledge articles, CRM data, external data via Data Cloud, public websites, and so on.
• Actions: The goals an agent can fulfill. This is the predefined task an agent can execute to do its job based on a trigger or instruction. For example, it could run a flow, prompt template, or Apex.
• Guardrails: The guidelines an agent can operate under. These can be natural-language instructions telling the agent what it can and can’t do, when to escalate to a human, or could come from built-in security features in the Einstein Trust Layer.
• Channels: The applications where an agent gets work done. This can be your website, CRM, mobile app, Slack, and more.

To help ground this, let’s look at an example of WhatsApp assisting customers with product recommendations, questions, purchases, and returns. As marketers, we know that engagement via WhatsApp moves customers through their journey with urgency. And the customer journey doesn’t fit neatly into the roles defined within an organization. 

Marketing needs to work with Sales, Service, and Commerce to provide an end-to-end customer journey that meets individual customers’ needs. The agent has a role — a job to be done — which is to move the customer to the next step on their journey via WhatsApp. It’s not realistic for a human to respond to every inbound chat at every second of the day or night. However, an AI agent can.

Brands are increasingly using ads that directly connect to WhatsApp. Yes, it might make sense to connect with a human if there’s a clear indication of conversion, or worse, attrition or negative sentiment. But there often is a reasonable solution available that AI can handle. 

For example, let’s say a customer asks a question that can be answered by consulting the FAQ on the website. An agent can not only answer it efficiently but automate one step further. It can take an appropriate action, such as sending the person product information, offering them a product bundle or promotion, entering them into a nurturing journey, or even setting up a call with an associate. Once it’s determined that the customer isn’t in a critical state, the agent can help the customer and take action to move them to the next phase of their journey.

Agents are a logical evolution of AI. They’re intelligent actors specific to situations within the context of a company. They require company-specific knowledge and access to internal tools and processes to make them effective. This builds on recent developments within AI Large Language Model (LLM). Chatbots such as ChatGPT can answer broad questions but are based on static, public data. They can’t answer questions specific to a company or those that rely on timely information. 

Retrieval Augmented Generation (RAG) enables extending the base chatbot LLM model and adds relevant, often proprietary knowledge that is more current to the LLM and augments the output. Both LLM and RAG are foundational components within an Agentforce agent. The agent needs to be informed by a specific model to get the appropriate response in order to determine what to do next. Agents, generally, aren’t going to be very effective if they are reliant on a general purpose, foundational model.

Marketers need to think about what the agent needs to do it’s job. It’s different from even a year ago. LLMs and foundation models are really about capturing all the data so you can get great natural language responses on virtually any topic. The more data the better. However,  LLMs don’t have the knowledge specific to the business. RAG helps to bring in the needed context, such as information that is related to recent product offerings or customer service issues. 

Now take this a step further using an Agentforce agent. The agent’s reasoning is greatly augmented. The agent has access to customer data typically found in Data Cloud. Most important, the agent can act using the Atlas Reasoning Engine to learn from the information provided, evaluate pros and cons, predict outcomes, and make logical decisions. 

The agent can assess the situation with near real-time understanding of the customer’s context and the latest information within an organization. Plus, the agent can populate fields in the customer record, including flagging for likeliness to buy. It could also trigger a new journey or automation, create a service ticket or send a link to engage a human, or create an opportunity for a sales rep to call. 

AI agents will fundamentally change how marketers work. We will have new capabilities, especially in working across organizations to move customers to the next step of the journey. It’s up to us to ensure a  strong data foundation that will guide these very smart “interns”. But if done right, marketing will lead the charge for more efficient workflows, improved personalization at scale, and better customer experiences in the years ahead.