Optimize Your Organisation’s Email Marketing with GenAI Agents

Introduction

Congratulations! You run a successful business. Through your web pages, social media campaigns, webinars, conferences, free resources, and other sources, you collect 5000+ email IDs daily. The next obvious step is to send them an email based on the source and user behavior of these users, with the goal of converting them to paid customers. With generative AI agents, marketing managers can reduce the team’s bandwidth for creative tasks by building autonomous email systems without worrying about the volume of emails to be sent. This article will guide you on how you can implement GenAI agents in your company to automate email marketing.

Overview

• Understand how traditional email marketing systems work and what their challenges are.
• Learn about the different GenAI agents you will build for the smooth execution of autonomous email marketing.
• Understand the importance of Human Intervention and feedback while testing your agentic System.

Traditional Email System

Let’s understand the traditional email system first. Imagine you have a vast internet presence, which includes the organization’s website with multiple web pages, blogs, free downloadables, etc. And you are constantly pursuing expanding your online presence with webinars, online conferences, and access to free e-books and product trials. All of this results in a vast collection of email IDs. A generic traditional marketing system involves:

• Keeping a proper record of the source of the email ID (sign-ups, webinars, downloadables) and tracking their behavior (which web pages they browsed, which resources they downloaded, the webinars they attended, etc).
• Based on the above point, the marketing team segments the customers based on similar properties and sends customized emails to each segment.
• Now, based on whether the response of the email (if a person has opened the email, the next email is sent, etc.), either a reminder email or a customized email (based on the activity on the first email) is sent. This cycle continues until the lead is irresponsive or becomes a paid customer. At this step, we will also be analyzing the performance of the email using engagement metrics like click-through rate, open rate, etc.

Sounds tedious and crowded, doesn’t it? However, with generative AI, we can build autonomous agentic systems that will do what is necessary within the rules and boundaries set by the organization. Let’s have a brief understanding of agents.

What are AI Agents?

Just like you hire a travel agent to delegate the planning of your trip to free up your time to do more productive/relevant stuff, AI agents are the modern-day tools that let you build complex autonomous systems to perform tasks, repetitive or otherwise.

An AI agent can understand the environment and the circumstances it is presented with. It can make decisions based on the environment without any human intervention, and perform tasks based on these decisions.

The critical task while building agentic systems is not creating the agents but defining the structure of the autonomous agentic system, the number of agents to make, etc. We will understand this while understanding how to build an autonomous email system.

Email Marketing Optimization Using GenAI Agents

Let’s look at how you can transform your marketing team and bring absolute efficiency to their workflow.

Here are the four steps you can follow to build an automated email marketing campaign using generative AI agents.

1. Planning the Agentic System
2. Creating the Agentic System
3. Performance Evaluation with Human Intervention
4. Canary Deployment and Feedback Implementation

So, let’s go through the steps mentioned above one by one.

Step 1: Planning the Agentic System

The very first step is to build the structure for the agentic system. The structure should primarily involve the following:

1. Identifying the Tasks

An efficient system should have the right number of AI agents where no agent is overworked or underutilized. The best way to determine the number of agents is to determine the tasks involved and the number of people required to perform these tasks to achieve the goal traditionally.

In our instance, we have four tasks to perform if we were to follow the traditional email marketing system:

1. Writing the email
2. Editing the email
3. Planning the campaign
4. Publishing/sending the email campaign

2. Finalizing the Number of Agents

Once we have the tasks ready, the next step is to assign the tasks to the agent. Ideally, one agent should have one primary task. But this is not a hard rule; an agent can also take up multiple small tasks. But in our case, let’s stick to one task for each agent. Therefore, we will create four agents:

1. Writer
2. Editor
3. Planner
4. Publisher

3. Designing the Interactivity Between Agents

The next step is to establish the interactivity among agents. In our case, the flow will look like this:

1. The Writer agent fetches user details for each email ID and writes an email using an LLM.
2. The Editor agent takes the email from the Writer agent and edits it per the set guidelines.
3. Based on the subject and the content of the edited email, the Planner agent plans the campaign and sets instructions.
4. The Publisher agent collects the instructions from the Planner agent and the edited email from the Editor agent and sends the email as per the instructions.

Once we have clarity on the above three points, the next step is to create the agentic system.

Step 2: Creating an Agentic System

In this step, we will construct the four agents that we designed in the planning stage:

1. Writer

The Writer is the first agent in your autonomous system. Once an email ID is added to the database, the Writer agent reviews the source and the recorded behavior. It then writes an email based on this using the preferred LLM you have connected with. The Writer agent will also incorporate the rules for words you cannot use and other relevant parameters per the brand guidelines.

While writing the email, it can refer to the previous emails you sent with the same behavior in the traditional setup to prevent itself from hallucinating and try to improve the email to increase the chances of improving the open and click-through rates.

Note: Since the planner uses the data collected, you must ensure your data collection is accurate. Otherwise, you will witness garbage in and garbage out.

2. Editor

Next, the Editor agent comes into action. The editor agent will check the email written by the writer agent. You can ask the Editor to check the grammar of the written email, if it aligns with the plan by the Planner agent, if there are any unparliamentary words, etc.

Note: The Editor needs to balance thoroughness with efficiency. Over-editing can delay the campaign, while under-editing can result in errors slipping through. Ensuring the agent’s rules are well-defined is critical.

3. Planner

The Planner agent reviews the edited email from the Editor agent. It reviews the email content and subject line and also accesses the user’s history to understand the right time to schedule the email.

Once the first email is sent to the newly signed-up person, the email and its performance are saved in a pre-determined database. The Planner agent can then plan round 2 of the email for the same person.

4. Publisher

The publisher agent publishes the campaign through the emailing software of the company based on the instructions from the Planner agent. The Publisher agent will also collect the report on the performance of the email and add it to the database. This will then be collected by the Writer agent to determine what the follow-up email should be.

This concludes the process of creating the agent. The next step is to test it internally on sample data.

Step 3: Performance Evaluation with Human Intervention

You can either generate synthetic data or simply generate some emails by picking some samples of diverse behavior.

Once the email is generated from the sample data, the marketing manager and team or the concerned people in the organization must read and check the email. This has to be done with great attention to detail, as any changes in the agent must be made based on their feedback before deploying it to production.

We will have to go back to step 2 based on the outputs of the test and modify the agents until we get the required result. This step is absolutely crucial and cannot be avoided, no matter the amount of iterations it takes to get the best possible result.

Step 4: Canary Deployment and Feedback Implementation

Once the internal testing is complete, the next step is to deploy it. The best strategy to follow here is canary deployment, where we release the new version of a tool gradually to minimize any disruption and allow real-time testing on a subset of users.

If you have a small database, you can deploy the agentic system to 20% of the model, and if you have a database in millions, you can deploy it to even 1%. At this stage, we analyze the engagement metrics, like CTR, open rate, etc, to see if they are improving on this set. If not, then you need to modify your agentic system again and see if there is any improvement in its performance.

Note: Even at this stage, you cannot call the system autonomous, as it still requires constant human intervention to ensure that the agents improve based on the performance of the emails sent.

Once we are satisfied with the performance of the sample set, we can gradually increase the subset of the audience on which the model is deployed. With each increase in the subset, we will have to check its performance and tweak the agents accordingly until we get the best performance on our email campaigns.

At the end of this exercise, we will have a fully functional AI agentic system for email marketing that not only generates emails but also records the performance of the emails. This performance record is taken in as feedback by the Writer agent while writing the follow-up emails.

Conclusion

Optimizing your organization’s email marketing campaigns with GenAI agents allows for personalized, efficient, and scalable communication with your audience. By implementing an autonomous agentic system, you streamline the email creation and distribution process, enhancing customer engagement while reducing manual effort. With the right setup, performance monitoring, and iterative improvements, your AI-powered email campaigns can significantly boost open rates and click-through rates, leading to higher conversion rates and overall business success.

If you want to know more about AI Agents, checkout our GenAI Pinnacle Program.

Frequently Asked Questions

Abhiraj Suresh

My name is Abhiraj. I am currently a manager for the Instruction Design team at Analytics Vidhya. My interests include badminton, voracious reading, and meeting new people. On a daily basis I love learning new things and spreading my knowledge.