How to Use NLU to Delight Your Customers
NLU lends itself naturally to improving the customer experience. Your customers produce a large amount of data through interactions with your website. This data can be immensely useful to you.
- How NLU works
- How NLU is opening doors in the CX world
- Analysis and insights on customer feedback
- Automatic support ticket organisation
- Employee satisfaction analytics
- Social media sentiment
- Where NLU is just not quite ‘there yet
Above is a typical example of a conversation taking place between a human and a chatbot that’s been enhanced with Natural Language Understanding technology. As you can see, we expect a lot from our chatbots. Not only do they need to understand all the grammatical rules of the English language, but they also need to be able to decipher human speech as it is spoken -- warts (colloquialisms, abbreviations, typos) and all. That’s a lot to ask of a machine.
Natural Language Understanding, or NLU, is a subdiscipline of Natural Language Processing, or NLP. Where NLP refers specifically to the range of tools machines use to process human speech, NLU has a narrower purpose: getting the machine to understand what the text it’s processing really means. Machine reading comprehension involves things like categorising texts, gathering news, and of course, analysing content.
Machine learning is what allows machines to learn from large amounts of data without being specifically taught. The machine studies the data and, from it, learns how to best perform certain tasks, improving exponentially as it studies more and more data.
We humans produce a lot of data -- to the tune of 2.6 quintillions (which is a 1 with 18 zeros) bytes daily.
There are four types of machine learning: supervised learning and unsupervised learning, in which the machine either learns an algorithm from a teacher-like training dataset or has no training dataset to guide it and must deduce the patterns in the data on its own. Then there’s reinforcement learning, where the machine receives a reward/penalty for a particular action in response to the observed environment.
Finally, there’s deep learning, which uses neural network architectures (specific sets of algorithms) that mimic the structure and learning patterns of the human brain. Deep learning teaches machines by example and therefore requires large amounts of data. Luckily, we humans produce a lot of data -- to the tune of 2.6 quintillions (which is a 1 with 18 zeros) bytes daily. That’s probably why deep learning has now surpassed humans on tasks like classifying images, and why it just keeps getting smarter with each passing day.
The good news is, NLU lends itself naturally to improving the customer experience. Your customers produce a large amount of data through interactions with your website, engagement with your brand on social media, and email correspondence, to name a few. This data can be immensely useful to you, but there’s no way you could possibly make anything out of an unstructured mountain of data. That’s where NLU comes in.
There are four key ways that NLU can be leveraged to improve your customer experience.
Your customers leave your feedback across a range of channels, from customer support exchanges, app store reviews, review sites, online surveys, and more. Imagine if you could scrape all that unstructured data and run it through a machine to sort through it, detecting patterns, and finally organising and presenting it in a human-friendly way (charts and diagrams, primarily).
Considering a human agent can only process up to 1,000 data sets of feedback per day, getting help from a machine is essential for companies.
That’s precisely what NLU can offer. Using the technology, you’ll be able to quickly and easily see trends in your customer feedback, and use it to make decisions on what changes you can implement that will most likely have the largest impact on your customers' happiness.
What's more -- NLU is getting pretty good at learned associations. So good, in fact, that this is what our Chief Science Officer at Ultimate.ai, Jaakko Pasanen is most excited about in terms of AI developments in 2020. He explains:
"You feed [the neural network] text like who is uncle, child, sister, and then it learns to create associations in this family tree that it has never seen, and remember those associations. In a way this is the first step towards representing associations."
As machines continue to improve at being able to understand and predict associations on their own, their ability to read and understand chunks of customer feedback with increasing accuracy.
Considering a human agent can only process up to 1,000 data sets of feedback per day, getting help from a machine is essential for companies with any kind of substantial customer base that wants to understand their data.
Traditionally, customer service agents have had to spend hours upon hours routing incoming support tickets to the right agent or team, giving each ticket a contextual topic tag. But now with significant advancements in NLU, machine learning technology has proven able to do this job automatically. The machine studies historical customer service data already tagged with keywords (the more it studies, the more accurate it will become), and learns how to automatically identify, tag, and route tickets based on their corresponding category and priority.
Look at how Uber is leveraging this tech. The ride-sharing giant’s robot-customer-support-ticket-sorter goes by the clever name of COTA (Customer Obsession Ticket Assistant). Not only does COTA read and sort incoming conversations, but it adds in three potential solutions on each of its hundreds of thousands of tickets that surface daily on Uber’s platform that plans over 400 cities worldwide. Amazingly, COTA gets this process right over 90 percent of the time. The agent then selects the appropriate solution and personalises it for the customer. Uber has reported increased ticket resolution times and increased customer satisfaction.
Unhappy employees cost money in many ways: not only will dissatisfied employees be far less productive and efficient on the job, but they’re also far less likely to go out of their way to make your customers happy. On top of that, the more employees you lose, the more money you have to spend on recruitment, hiring, and training. So, it’s worth keeping an eye on the pulse of your employees to see what improvements you can make within your company to improve their satisfaction and employee retention rate.
With NLU technology, it’s possible to create surveys full of open-ended questions, since the machines are not just reading the text literally, but will work to analyse and understand the text as a whole.
In the past, traditional surveys have forced employees to fit their answer into a multiple-choice box -- even when it didn’t. But now, with NLU technology, it’s possible to create surveys full of open-ended questions, since the machines are not just reading the text literally, but will work to analyse and understand the text as a whole, allowing your employees the freedom to really let you know what they’re happy with -- and what they’re not. The NLU tech can then read, decipher, and analyse this data -- no matter how high the mountain of responses -- and present it to you in a comprehensive way, allowing you to address issues like employee burnout before they begin.
Did you know that on an average day, over 500 million tweets are released into the Twitterverse? If you already track your brand mentions on social media, and you probably do, you’ll know that platforms like Twitter offer a treasure trove of customer feedback and interaction. But how could your team ever find the time to declutter this data? Enter NLU once again, to save the day. With NLU technology, it’s possible to sort through all your social media mentions and messages, and automatically identify whether the customer is happy, angry, or perhaps needs some help -- in a number of different languages.
Social media sentiment analysis can be incredibly important and beneficial for brands. Just look at Nike. When the company hired NFL Quarterback Colin Kaepernick as its new spokesperson, the media decreed it a marketing disaster. Overall sentiments for the brand on social media, the press said, showed overwhelming disappointment in the brand. It’s a good thing Nike didn’t listen to the media, though. It turns out that between the lines of all the negative sentiment was a wave of positive sentiment for the brand that actually resulted in an increase in sales of over 30 percent for the sporting manufacturer.
While NLU technology is leaving us breathless with the speed of its development, it’s not perfect -- not yet, anyway. It still struggles to tell homographs apart, for one. Homographs are words that are spelled the same but have different meanings (like ‘you guys rock!’ and ‘it was heavy as a rock’) -- and choosing the correct meaning of a homograph is something that we humans can do easily by simply paying attention to the context of the sentence. Meanwhile, NLU still struggles.
It also struggles with text that contains multiple sentiments. Normally, you see, NLU can easily tag a sentence positive or negative, based on the keywords contained. But sometimes, a sentence will contain multiple sentiments, making the NLU’s job significantly more difficult.
Take a look at these two sentences:
- “For the love of God, I can’t begin to express how disappointed I am in my new gaming console.”
- “I wasn’t too thrilled with the sprinkling system when I first installed it, but after a while, it grew on me.”
As you can see, NLU doesn’t exactly have it easy. It’s just a good thing that it isn’t likely to complain or get tired. As the technology matures, we can bank on the fact that we’ll be asking more and more from it. And with good reason, because this clever technology has the ability to completely transform the customer service and customer experience landscape as we know it.
Stay in the loop
Subscribe to our email newsletter. No spam, just occasional insight from our experts.