Creating Simplicity in the Workplace with NLP

Author: Matt Paterson
Last modified: June 7, 2021

Challenge:

If your business has a general email address such as info@yourBusiness or help@yourBusiness, then you probably know how much work can go into answering customer emails in a timely manner. You may even pay an employee or 10 to answer emails all day long. If this is the case then you know all too well that there are days when that number of emails is just too high for your team to handle in a reasonable amount of time, or within your Service Level Agreement (SLA). Have you ever missed a very important message due to it being buried in the queue? Ever wish you had a way to identify the customer complaints so that you could focus on correcting those issues immediately? That might be about to change. 

Benefits:

  • Reduces labor costs by efficiently routing communications to the appropriate staff
  • Labels emails as “Positive”, or “Negative”, or “Refund Request”, etc for easy attention
  • Can summarize long emails so that an agent can phone customers when appropriate
  • Infinitely scalable — no need to add levels of management just to handle your email queue
  • Better Reliability — when the right people are looking at the right emails, you’ll catch every important detail and no immediate issues will fall through the cracks
  • Has the ability to create a customer-feedback forecast that automatically predicts the number of staffers that you’ll need to address the level of contact that you are comfortable with

Cloud Brigade’s new SmartDispatch Solution is able to:

  • Use Natural Language processing to screen every email as soon as it comes into your queue
  • Label it according to the relevant department
  • Identify and label it as positive or negative feedback
  • Prompt an auto-response in case customers need a higher-touch care

——-

About/Background

Natural Language Processing, or NLP, is all the buzz in the ML/AI tech news, but it is not a new technology. Every time you call a company that has an automated voice operator on the phone, or use an automated chatbot online, or see google’s predictive text in your gmail account, you are using NLP. So why then, with all of this NLP in the market, do you still have to hire staff to answer your customer emails?

One big reason is personality. While Siri might seem like a complex presence, AI lacks a human touch, and really that’s a good thing. Your customers know when they are interacting with a chatbot as opposed to a real person. Beyond that, though, you know from being a customer yourself that when you create a personal bond with a human being, you are exponentially more likely to purchase from that company again and to recommend their services to your friends. 

So how do you reconcile this? How do you create a personal bond with as many of your customers as possible without paying all of your profits away to an over-sized customer service department? The answer just might be Cloud Brigade’s SmartDispatch Solution.

SmartDispatch uses NLP to screen every email in near real time as it comes into your queue, labeling each email with its level of importance, its topic, and its tone. SmartDispatch then directs each one to the right person or team so that you don’t have refund questions sent to your tech staff, or website bug notifications sent to your accounting team. With SmartDispatch, you save time and reduce or eliminate the added stress of reading and re-reading emails that you’ll just have to send to somebody else.

Business Challenges

  • Irresolvable Complexity – The human mind can read text and conceptualize feelings and importance, whereas until recently, computers could not do this
  • Improving Systems and Processes – Smart Dispatch allows you to use AI to re-route the important emails to the people who are best suited to respond to them
  • Skills & Staffing Gaps – No longer must you train every customer service employee on every aspect of your business before they handle a single customer request–now they can be experts in a specific area and answer more queries with precision and confidence
  • Antiquated Technology – Although email once revolutionized how we can interact with our customers, the use of SmartDispatch will make old email look like the Pony Express

Solution and Strategy

Cloud Brigade is no stranger to finding the best technologies for your business. We can build a solution from scratch using python libraries such as Natural Language Toolkit (NLTK) or Textblob, but often we can get a head start with another product that might be a better fit for your use-case. As an AWS Partner, Cloud Brigade has experience pairing the right technology with the right problem, and Natural Language Processing has a number of options. A small-scale solution might best deploy AWS Lex to build a chatbot on your website, for example. A larger-scale integration may include incorporating SmartDispatch into your own branded CB-Abi solution (Cloud Brigade’s AI Business Intelligence Solution). While most projects will fall somewhere in the middle, we are adept at finding existing solutions in the market and piecing the right products together by inventing the data pipelines, user interfaces, and dev-ops solutions that are required to connect your current email manager or CRM (customer relationship management software) to the technology of tomorrow.

Where it Began

Before I came to study Data Science and Computer Science I was a Sales Director in charge of about 50 people in two call centers. Part of my job was creating a forecast for how many customer service employees that we would need to hire and train based on the expected number of phone calls and emails that we would receive in the coming months and years. One of the factors that went into this forecast was the number of emails that an employee was able to answer each hour, and our Service Level Agreement, or SLA. The SLA was 24 hours, which meant that every email we received had to be answered within 24 hours of its receipt. You may in fact have this same situation in your company. 

If you do, then you also know how easy it is for an email that came in at the end of the work day to get pushed off until tomorrow. It’s still “in SLA”, you’re not outside of a reasonable response time, but maybe the email is of a much higher importance to your business than the average missive. Maybe it’s a user or member who is ready to cancel their account and trash you on social media, and they just need a quick response to cool off and be won back to your favor.

But how do  you know, without opening each email and reading it, if that email coming in at 5:01 has the potential to affect your sales for the next month? How long does each email take to read? How can you read it without responding, and how long does that take?

That’s where NLP with SmartDispatch from Cloud Brigade comes in. 

This Is No Ordinary Project

Building Natural Language Processing (NLP) solutions is not as easy as making a predictive sales forecast or predicting home prices based on the square footage of a home. Before we could tackle this issue, it was important to build NLP models from scratch, see the pitfalls and the intricacies, and learn how to customize and temper the results for better interpretation. For example, take an email such as:


“I bought a microphone and holder from you one month ago and it sounds terrible. Im happy to try another product but id like my money back”

Some off-the-shelf NLP products may not have enough Hyperparameter Tuning (a process of teaching your model the intricacies of your dataset or real world use case) mark this email as a “Positive” email because it contains the words “happy”, “like”, and “try”. Clearly we can see that this customer is willing to stay with this company, but the word “terrible” really ought to throw up a flag. 

Also, from context, a trained NLP model will be able to route this email to the refunds department, or even to a sales representative who can find a better quality product or a new microphone as maybe the particular one that the customer received was defective. While early iterations are much more accurate at predicting sentiment than choosing between several departments, we are now in an era where technology advances almost as quickly as we can dream it. Once perfected, SmartDispatch could automatically process this refund if you so choose, as maybe this is a new customer or a repeat customer that has never received a refund or discount. The automation frees up labor costs and time that your employees can spend being more productive. These are just some of the levers that your management can pull, and that Cloud Brigade will work hand-in-hand with you to perfect as we push your solution to your team.

Smarty-Pants Technology

We can build a full, specialized Customer Relationship Management software solution (CRM), or a specialized user-interface for your email stream. However, there are great solutions out there for CRM already from companies that do CRM and only CRM. If you want a fast solution without a full custom build-out, we can do that too. Using services such as Amazon Comprehend and Amazon Lex (The AI product that powers Alexa), Natural Language Processing can be done in hours instead of weeks. There are also other products in the AWS Marketplace that we can employ in this buildout to decrease turnaround time, and thus costs.

With the help of some data engineering, we can create a pipeline that pulls each email into a workflow allowing it to be analyzed for sentiment (she loves us, she loves us not), topic (product, shipping issues, or ‘the way that I was talked to on the phone’), and department and level (is this a question for Julia the CTO, or for Kai the Director of shipping and fulfillment or for Maria in Customer Service?).

So what does this mean? This means that we can program a dispatcher that lives inside your computer and sends the right emails to the right people with a quick summary of what’s going on so that your staff saves time and feels less stress in their day-to-day. What’s better for your customers than a happy customer service team?

PHOTO CREDIT: https://www.freepik.com/free-photo/people-typing-computers_6802098.htm

A Successful Integration and a Technology Handoff

One thing we have learned through our work in Machine Learning is just how time consuming and intricate each step can be. A lot of times, the work of building up the infrastructure and lining up the data so that the different parts can all communicate with each other are actually more challenging than creating and using Artificial Intelligence. This product is no different. The live software must be programmed to scan every email in near real time, there must be a logic that diverts each email to the right employee (and that can also be done using Machine Learning), the data pipeline must be programmed with the proper error handling and notification systems, and the data reports must be connected to the alert systems with bullet-proof security. However once the data science has been fully integrated with the CRM or custom email interface, very little regular maintenance is required to keep the system working.

The final step will be to package the Machine Learning model with an API connecting your model with your CRM and/or email server.

Opportunities

While this project is designed to reduce costs and improve customer service provided by your business, it is not a niche product.  From identifying questionable copy before your marketing or sales team releases it, to monitoring social media sites for mentions of your business, SmartDispatch can be repurposed, scaled out and adapted easily. We can also build you a chatbot! We look forward to the final product being useful in companies of all sizes and needs.

WHAT’S NEXT

If you like what you read here, the Cloud Brigade team offers expert Machine Learning as well as Big Data services to help your organization with its insights. We look forward to hearing from you.

Please reach out to us using our Contact Form with any questions. 

If you would like to follow our work, please sign up for our newsletter.

About Matt Paterson

Having lived in five different states and with over 12 years of experience in sales, entrepreneurship, machine learning and data science, it’s no surprise what Matt enjoys most is learning about new things. With a BA in English from the University of New Hampshire, Matt began building his professional foundation by surpassing expectations in sales and leadership roles before starting his own small business.

As the numbers flowed, Matt saw the importance and opportunity around this data, and opted to obtain his second degree in Computer Science. Matt’s highly technical engineering skills allow him to analyze and forecast data using SQL, Python, Scikit-Learn, Tensorflow, PyTorch and Django, while also looping in his creative problem solving and people/project management abilities to deliver solutions that people are able to use and enjoy.

When Matt’s not geeking out over data at work, he’s learning how to play the piano, running on the beach, reading fiction, or identifying exoplanets with Neural Networks and deep-diving in to other outer-space exploration topics.

CONTACT US

Would you like more information, or have an idea we can help become a reality?

Contact Us
Email Us
Call Us 831-480-7199
Newsletter Subscription