How Mixmax Uses AI to Automatically Reschedule Sequences: Engineering Behind a Single Checkbox

The Problem

Creating a carefully crafted email sequence can be a labor-intensive task, but it can be rendered frustrating when one or multiple of your recipients respond with an out-of-office (OOO) message. Suddenly, follow-up emails in the sequence are likely to land in an inbox that goes unread, wasting both time and effort. This is a problem for multiple reasons:

• The recipients notice that their out-of-office auto-reply was not respected and got flooded with emails anyway
• In order to avoid that it was a manual step for our users to reschedule the sequence stage for those recipients

But What If We Could Automate That Process?

At Mixmax, we have tackled this problem head-on. Our feature automatically reschedules the next stage in a sequence for recipients who are out of the office. The user interface is elegant and deceptively simple — a single checkbox on the sequence settings page, but the engineering behind it is interesting and more complex.

The Challenge

While recognizing that an email is an auto-reply is straightforward at first, it's merely the beginning. The real challenge lies in determining whether that auto-reply constitutes an OOO message. It involves interpreting natural language responses to extract critical information, such as the recipient's return date.

Our Solution

To resolve this issue, we leverage the power of AI. Here's how it works in a nutshell:

1. Email Classification and Preprocessing: We determine based on email headers that an email is an auto-reply, but filter out those that are not relevant before processing them using an LLM. For example Google Calendar invites and Slack notification emails are often auto-replies, but are not relevant for our use case. Additionally, a significant portion of an email consists of unnecessary data that isn't relevant to our needs, such as layout elements / HTML tags. By preprocessing the emails before analyzing them with LLM we make sure that our solution is efficient, both in terms of performance and costs.
2. Natural Language Processing Using Large Language Model: After classification, we utilize LLM to analyze the content of the email. This allows us to accurately determine if the given auto-reply email is in fact an OOO message, and extract the return date specified by the recipient.
3. Automatic Rescheduling: Once we determine that a recipient is OOO and retrieve their return date, our system automatically reschedules the follow-up sequence stages. This capability ensures that emails reach the inbox when they can be read — maximizing engagement.

The Technology Behind It

To power this feature, we have harnessed modern cloud technologies:

• AWS Lambda: This serverless computing service allows us to scale seamlessly as we process large volumes of auto-reply emails. Instead of managing servers, we can focus on refining our AI-driven solutions.
• OpenAI Integration: We've integrated an advanced language model that help us interpret human-written auto-reply emails. The ability to understand context, and extract relevant dates ensures surprisingly high accuracy in determining OOO statuses.
• Event-Driven Architecture: Our solution employs an event-driven architecture. By using an EventBus to publish and subscribe to OOO-related events, we ensure that components of our system can react to changes, allowing for quick updates of sequences based on the recipient's availability, in a fault tolerant manner. This decoupling of system components also improves maintainability and scalability.

A Single Checkbox, Big Impact

Each month, we reschedule sequence stages for several thousands of recipients due to them being OOO, ensuring that emails reach their intended audience at the right time instead of being sent into the void. The simple and straightforward UI conceals the engineering and AI models working behind the scenes to deliver this functionality.

Interested in joining the team? Visit Mixmax Careers.