Automatic Labeling Service for Incoming Gmail Messages Using AI

Every day, we are inundated with dozens or hundreds of emails from clients, partners, and subscription lists. The problem is not the sheer volume of messages received; it is the time-consuming process of manually sorting through large amounts of email, most of which end up in either the "Important" folder or the "For Later" folder. The use of AI nodes for Gmail Automatic Email Labeling allows users to set rules based on their specific needs and preferences so that incoming emails can be labelled according to those specified rules with virtually no effort on the user's part.

Sounds like it could really help some people, right?

"For the past eight years we have experimented with 43 techniques for automating repetitive tasks, starting with simple scripts and progressing to actual AI. The results have shown us that any task which can be expressed in algorithmic form will be automatable. One example is email labelling. If you are currently doing this manually, you know how laborious and time-consuming it is."

What is Automatic Email Labelling?

Key Terms and Definitions

Automatic email labelling is defined as the automatic categorizing (i.e., creating tags) of incoming Gmail messages. With email labelling, Gmail users do not have to spend any time on the process of categorizing emails because Gmail does it for them automatically.

Gmail uses machine learning technology to process emails automatically. The Gmail service determines where to store an incoming message based on how the system analyses the message's content and other data available to the service. For instance, if an email received from a customer has the word "payment" in it then it is automatically routed to the "Finance" folder, and in the case of someone requesting a consultation, it will go into "Sales."

The beauty of this process is with the use of AI nodes, which are modular blocks that can learn through Machine Learning (ML) and Natural Language Processing (NLP). Unlike regular Gmail filters, that can only identify keywords and sender addresses, AI nodes can understand the content of each email. When a customer writes the following: "I made a payment yesterday and would like to know the status of my order", the automated system instantly identifies both the words "order" and "payment" and then labels them accordingly.

The primary objective of this automation is not merely to be able to sort through your emails, but to make certain that each email will arrive where it can be dealt with as quickly as possible. Hence, this lays the groundwork for the subsequent automation of the next steps, which includes notifying the manager or triggering the payment verification process.

Why Gmail Users Should Care

To be frank, manually handling emails has been a genuine bottleneck, and as the number of emails continues to grow so does the amount of time spent managing them! According to research done by McKinsey, people spend up to 28% of their working hours performing tasks related to emails, which reduces their overall productivity dramatically.

Fifty emails per day is a somewhat attainable goal, however, 200+ emails a day causes major issues, such as overlooking important emails, and email management consuming most of the time in the morning. With automated labeling, it acts as a virtual personal assistant, speedily and accurately sorting through 1,000 emails daily with consistent effort and accuracy.

Moreover, it helps decrease cognitive load. You may find it mentally overwhelming to open a list of a hundred or more emails that have been basically dumped in your inbox. You're going to have to decide what is most important right now, which emails should be set aside, etc. By having the emails organized in folders, it allows your brain to be focused on what is really important.

In industries where speed is of the essence, such as trading or crypto-arbitrage, a delay of just a couple of hours could end up costing you a lot of money, so automating the labeling of the emails can also quickly highlight messages that are urgent (such as price drops, or important listing information). Urgent emails receive a label and end up in a separate folder called "Urgent," and they also trigger a Telegram notification immediately. All other emails go to a standard folder.

Working with an organized email inbox removes a lot of unnecessary decision making, which reduces stress and improves the ability to concentrate on the task at hand.

How does an automated labeling system with AI nodes work?

How Does the Architecture of AI Nodes Work?

An email processing system made up of AI nodes consists of modules that allow you to process email in different stages. The first node is where the email enters the pipeline (i.e. from the Gmail API), the second node is where the text of the email is analyzed by Natural Language Processing (NLP) technology, and the third node is where labels are assigned and notifications sent (if needed).

This allows the pipeline to process the email (from when it arrives to when it is labeled) as defined by the user's needs. All of the AI nodes are self-contained modules that work independently, making it easy to add additional modules or services to the email processing pipeline without changing the entire design.

For example, suppose you receive 200 emails of varying types in a single day. The AI nodes will identify from where the email was sent (i.e. your client, partner, newsletter, etc.), determine what was in the email (i.e. payment, technical support, deal, etc.) and label (i.e. important, not important) the email. The "important" emails will immediately trigger a fast notification in Telegram. Processing time is reduced from approximately two hours down to 20 minutes, decreasing the number of emails that are missed by management. Additionally, as the only items that are seen by management at the end of processing would be the items that are most relevant to the company, there would be no missing items for management.

Additionally, machine learning and natural language processing (NLP) technologies form the backbone of the automated labelling process. Machine learning models are built on transformer architecture and will not only read the words of the email but also understand the meaning behind the email. The model takes the text of the email and breaks it down into tokens, converting those tokens into a numerical vector and then maps those tokens to corresponding categories. For example, if the text says "I paid for my order yesterday, when will I receive my order," the model will immediately classify it into both a Finance category and a Logistics category.

For more specialised topics, such as cryptocurrency or DeFi, the machine learning model is enhanced by additional training on hundreds of labelled training examples to ensure accuracy and to account for domain specific language and terminology. This enhancement is not time consuming and does not require users to have extensive machine learning knowledge, as in the event a model is uncertain, for example the model has a confidence level below 70%, the email will be automatically sent to a folder labelled "requires review."

The benefits of automatic labelling are significant for Gmail users, allowing for increased productivity because Gmail users will save about 1.5-2 hours per day in sorting emails. Colleagues can now dedicate more time for calls, data analysis or other higher-value work. Research out of California has shown that changing tasks requires the best part of 23 minutes before regaining full focus on the next task. Consequently, by eliminating excess task transitions, productivity increases will occur as well. With the label system, employees can quickly see only the emails that are necessary for their work. By establishing their own filing systems, both Tech Support and Sales are able to sort out the information they need without wasting time on irrelevant email correspondence. For example, let's say there's a company with three departments: Sales, Tech Support, and Finance, with all of their emails stored in one central location (i.e., a shared email inbox).

What was done: The company implemented automated labeling of incoming emails that categorized each email by department based on the context of the email; Gmail could sort these emails into their appropriate folders.

What happened: The time it took to process incoming emails was reduced by 60%, and at no time over the next two months was a request for assistance missed.

The ability to sort emails accurately has improved with the implementation of AI nodes, which achieved accuracies of 85%-95% based on the data that was used to train them and is higher than a human's accuracy because a human becomes fatigued over time and makes errors. As you keep adding to the data you are training the AI nodes on, and as you update the models, the accuracy of the predictions will continue to increase. Additionally, AI nodes can assign multiple labels to an email when there is more than one semantic theme within the email that may be of interest to multiple departments—this is something that is challenging to do manually. The number of false positives, or incorrectly labeled emails, is reduced because the AI nodes consider the sender, email context, and historical correspondence with the sender. This allows for less time spent correcting errors while also allowing for improved overall management of email correspondence.

Integrating with Gmail—Technical and Security Aspects

Gmail API Integration

The way you integrate with Gmail is through the official Gmail API using a secure OAuth 2.0 token. By using OAuth, you are giving the application a secure authorization token rather than your password. This entire process can be completed in 2-3 minutes. You will need to create a project using the GCP console, create a client ID and secret key, enter the required information into ASCN.AI, and authorize the application. Each step is clearly explained using the web interface.

The method used to label Gmail emails involves the use of webhooks. Gmail notifies ASCN.AI that there is a new email almost immediately (within 1-3 seconds). Amount of traffic: 250 requests per second per user (a rate that exceeds the average user's use case).

Making sure your information remains confidential and secure is critical. OAuth 2.0 guarantees that you will never have to share your password; you simply provide permission to perform specific actions such as reading or labeling but not deleting. Tokens are stored in an encrypted state, and even if you were to use an API to view them, you would not see the tokens directly.

We only store the minimum amount of data required to support your use of the platform; processed e-mail data is stored only as IDs and labels and processed immediately upon receipt. The data stored is in compliance with EU data privacy regulations (GDPR), thus minimising the risk of any potential leaks. All of your actions on the platform are recorded, including who labelled e-mails and when. Logs are available for the past 30 days and can only be accessed by the account holder who created the log so that they can view their actions as well as correct any mistakes. Two-factor authentication, IP access restrictions for all corporate customers, and continuous audit processes are also provided by this platform.

Use Case Study: Crypto Arbitrage

Problem: An arbitrage team has received e-mails and Telegram notifications regarding the price differences between exchanges for over a month, but has experienced technical issues with the email and Telegram notifications at times.

Solution: All e-mails with the label 'Arbitrage' were labelled as such and Telegram alerts configured to alert when price difference and mention of the exchanges (for example, Bittrex, KuCoin and Binance) occurred.

Outcomes: A total of 43 arbitrage opportunities were identified in sub-30-second response time (prior to implementing). An additional income of more than $6,450 was realised after implementing.

Use Case Study: Crypto Project Support

Problem: A Crypto Project receives more than 300+ inquiries daily to their general inbox (with varying levels of importance). Action: The NLP model (800 examples) was used to identify Critical inquiries; Critical inquiries cause alerting of a specialist.

Result: Processing time was reduced from 3 hours to 40 minutes, reaction time increased to 15 minutes, and 0 negative reviews received over 3 months.

Client Feedback and Results:

"Requests are processed through email and Telegram. Automatic labeling has eliminated missed requests; conversions increased from 12% to 19%."

"I filter and label my personal email as follows: Analytically Letters, Advertisement, Promotions - and save about 1 hour each week."

"We have three areas of focus: Crypto, Automation and Consulting. Automatic labeling allows everyone to see only their labeled emails. Time spent on emails has decreased by 70%."

Frequently Asked Questions (FAQ)

What are AI nodes?

AI nodes are blocks made with Machine Learning, extracting email body text, determining context, labeling content, and generating alerts if needed, using a visual editor and no need for programming. They can easily be customized to suit your specific business needs.

How do I connect my Gmail account to your service?

The process takes approximately 5 minutes. You need to create a project within the Google Cloud Console, receive a Client ID and Secret after completing the OAuth 2.0 Authorization, and then enter those credentials into ASCN.AI. The password will remain confidential. You will be provided with a secure token that can be revoked at any time.

What is your assurance concerning accuracy of labeling?

The accuracy of the system is 85-95% based on the training data and updates to the models. Any email that doesn't meet the model's confidence is sent to a "Requires Review" box so it can be reviewed before responding to avoid problems; the system collects statistics to assist with adjusting the rules.

How Long Does It Take To Set Up?

A basic scenario can be set up within 30 minutes with three categories. For more complex scenarios, where you have fine-tuning and integration, the length of time may extend to two hours; however, using templates will speed you up.

Where Do I Get Support?

There is documentation available; in addition, a chat box is available in Telegram with a response time of one hour. You will receive email support within 24 hours. There is a turnkey option available for enterprise clients based on an agreement.

How to Generate Income With Automation and AI Agents

Automation and AI Agents can assist you in completing repetitive tasks such as sorting mail, responding to frequently asked questions, or collecting data, which can account for 50-70% of your work week. Many people view automation as requiring costly programmers to create a solution, but with no-code platforms like ASCN.AI, you can create a solution and deploy it in a couple of hours without any technical knowledge.

Here is an example: automatic labeling, notifying through Telegram, and logging data into Google Sheets. In less than half an hour, you take three AI nodes that perform these actions and configure them to create a working system. A small team of three to five people can potentially save 10-15 hours per week this way, which could equate to approximately $1,200 per month at a rate of $30 per hour. As a result, you recoup your initial investment of the $79 monthly plan in the first month.

Another way to generate revenue is by selling pre-configured workflows and the setups required to implement these workflows. You may create and sell your solutions by taking advantage of the platform's white-label feature. Solutions may be packaged and sold to clients at a price range of $50 to $200 depending upon complexity, in addition to more clients being served as a result of increased automation. An example of this would be generating a report that previously took three hours to compile but now takes only 10 minutes, thereby significantly increasing the amount of revenue per hour of time worked.