Email management is a bigger challenge every year. In 2019, business email accounted for more than 128.8 billion emails sent and received per day, according to the Radicati Group. Adding to the challenge, many emails never make it to the right business account because they are sent to bulk accounts like info@company.com or sales@company.com. Regardless of where the emails land, the average full-time worker spends 28% of the workday reading and answering email, according to a McKinsey analysis. That amounts to a staggering 2.6 hours spent each day.

Corporate email continues to rule in the business world, but the deluge is impairing productivity, not to mention becoming unmanageable from a corporate perspective. TechJury cites that at least five out of 10 emails were spam in 2019, but spam isn’t the only problem. Oftentimes, the right email doesn’t reach the right person within the organization quickly, resulting in operational delays and rising costs.

The latest worldwide developments regarding confinement and, in some countries, the obligation to work remotely from home, have caused organizations and consumers alike to more heavily rely on email. This has generated a tsunami of emails for organizations used to other means of connecting with prospects and customers.

In the financial services, for example, these requests and inquiries used to be handled via physical call centers, many of which are now closed. Customers and prospects are often directed to send their questions per email, which results in a huge increase in the number of written requests to be processed.

Organizations without an automated process for handling incoming messages are at a disadvantage. First, there are the hard-dollar costs to pay staff to triage messages, and then soft-dollar costs of losing prospects and customers due to slow response time. Even those who have implemented some form of automation are concerned about the cost of improperly handling the messages due to challenges with natural language processing.

As a minimum, a simple classification and routing system would help speed up the response time by forwarding the requests to the correct department.

Too Little Business Value Extracted from Emails

Emails represent an extremely rich and essential source of business information. However, extracting insights and relevant information is difficult and time-consuming. As the number of emails continues to grow, it becomes even more difficult, time intensive and expensive to realize the business value.

The main issue is that emails fall into the category of unstructured data that is text heavy. This results in irregularities and ambiguities that make it difficult to automate the process.

Most current technologies for email classification fall into two broad categories. Rule-based approaches classify text into organized groups by using a set of handcrafted linguistic rules. These rules instruct the system to use semantically relevant elements of a piece of text to identify relevant categories based on its content. Essentially, the system counts the number of word appearances in the text and subsequently classifies the content into predefined categories.

This approach has several disadvantages. For starters, these systems require deep knowledge of the domain. They are also time-consuming, because generating rules for a complex system can be quite challenging and usually requires a lot of analysis and testing. Rule-based systems are also difficult to maintain and don’t scale well given that adding new rules can affect the results of the pre-existing rules.

Instead of relying on manually crafted rules, text classification systems based on Natural Language Processing (NLP) use statistical modeling. These approaches are usually more accurate than rule-based approaches; however, they involve building large-scale statistical models that are trained through machine learning (ML) using massive amounts of training data.

In many instances, sufficient amounts of training data are not readily available. The process is slow, costly and complex, involving several iterations of data preparation, feature engineering, model training, hyperparameter tuning and evaluation. Even if more data is thrown at the model to become more accurate, there is still about 20% of text-based information that the model cannot precisely categorize.

Additionally, most standard ML-based systems cannot explain why they made a certain decision. This “black box” effect is a substantial drawback in light of regulations such as the EU’s General Data Protection Regulation (GDPR), which stipulates that consumers have a right to know why and how corporations made decisions affecting them. Another issue with "black-box" models is they are hard to improve. If their predictions are wrong but there is no way to explain why, it is difficult to know how to fix the model so that it delivers the correct output.

Semantic Folding is Improving the Bottom Line

Now, there is an out-of-the-box solution for mining emails for business content and classifying them, that comes with standard filters and pre-built classifiers. If customized filtering is needed, it can be done with a fraction of the training data, as well as much higher accuracy and efficiency compared to other models. It is based on Semantic Folding, an innovative approach to Natural Language Understanding (NLU).

Semantic Folding is a procedure for encoding the semantics of natural language text in a sparse distributed representation called a semantic fingerprint. This unique approach provides a framework for analyzing unstructured data such as emails similar to how the human brain processes language data.

Semantic Folding is inspired by and based on the principles of cerebral processing and neuroscience. By applying neuroscience tenets discovered by scientist and author Jeff Hawkins to the realm of NLU, forward thinking technologies have been designed that can identify and analyze language faster and more accurately. Using Semantic Folding, an Artificial Intelligence (AI)-driven algorithm can quickly and accurately categorize emails. The result is an intelligent, efficient way to analyze emails so organizations can tap the value they contain, even across multiple languages.

Organizations that can figure out how to efficiently categorize and segment emails can improve their bottom lines, reduce costs, improve productivity, streamline and automate business processes, and serve customers better. The potential results are tremendous, as well as applicable to any business function. For instance, organizations can do the following:

• Detect and surface only relevant emails having to do with RFP and bid responses, budgets, forecasts and reporting for the Finance Department.
• Assist sales and marketing by extracting emails regarding requests for information, requests for quotes, market research, advertising and customer satisfaction.
• Surface emails about scheduling production, ensuring product quality and minimizing production costs to streamline manufacturing.

A Case Study in Customer Service

Customer service is among the most compelling use cases for email classification. For example, an international transportation company saw the benefits of an intelligent, AI-based semantic classification and routing solution to optimize email processing in customer centers. The company was coping with hundreds of thousands of emails sent every day in 35 countries in multiple languages. Teams were wasting huge amounts of time manually sorting out the emails that required action.

An AI-based Semantic Folding approach was created to scale a secure web service that that ignores irrelevant emails and automatically routes business-relevant emails to the appropriate regional center, depending on the language used in the message. Only 50% of the emails that made it through the spam filters were relevant to a customer service case. Many of the non-relevant emails were automated confirmation messages providing updates about the status of shipments, such as a container left at a certain port or a container that was not loaded onto the intended ship.

With the new solution, which was deployed in just eight weeks, customer service representatives were able to ignore irrelevant emails and focus on solving requests more efficiently based on case-relevant communications. The company greatly improved its responsiveness, leading to higher customer satisfaction and retention, not to mention the increased motivation of its customer center agents, who could focus on more rewarding activities. The solution required very little training data, resulting in tremendous cost savings. It has the potential to save the company €4.9 million annually.

The Good News

Although email is a faster, better alternative to paper documents and faxing, it can have a direct impact on corporate bottom lines by distracting workers from role-relevant tasks to spend hours daily dealing with unimportant messages. Considering this drag on workplace efficiency, it is time to adopt solutions for categorizing massive volumes of email in near real time using a powerful, low-cost, accurate email classification solution.

The good news is that this type of solution can also be applied to a broad range of messages, including instant messages, tweets and blog posts. The results are well worth it, including lower costs, higher productivity, improved customer satisfaction, and increased efficiency and responsiveness for all functional areas of the organization.