The onset of big data sowed the seeds of an interesting phenomenon that most companies are faced with – "we're drowning in data but lacking on insights”. Digitalization has led to massive amounts of data being generated at great speed, volume, and variety, making them too complex for a human agent to grasp. According to McKinsey, 80% of data scientists’ time is spent organizing data to make it fit for analytical purposes. Even more so, 55% of companies have a predominantly manual approach to data aggregation, resulting in high error rates. Using Natural Language Processing to extract value from underutilized pools of data can give your organization a competitive edge and open untapped growth opportunities in the fields of customer service automation, social media listening and brand recognition.
Integrating Artificial Intelligence (AI) into business operations has long been perceived as one of the most cost-effective ways of optimizing repetitive processes, getting a handle on underutilized data pools, and focusing on core revenue-generating functions. But is it the best AI investment for your company? Which machine learning (ML) techniques would ultimately make a difference in your KPIs? Well, let me answer these questions whilst shedding some light on the fastest-growing field of AI – Natural Language Processing (NLP). Since 2019, the field of NLP has been experiencing exponential growth, spurred by the rapid rise of self-supervised learning, the advent of the Transformer Model, and the Bidirectional Encoder Representations from Transformers (BERT). According to the Statista report, the market growth of the NLP sector is predicted to reach an astounding $43.9 billion by 2025. So where does all the hype come from? Is there actual money to be made in this technology, or is it just the usual case of fear of missing out (FOMO)?
First, let’s get the definition straight. Natural Language Processing is a branch of machine learning and AI which deals with human language and, more specifically, with bridging the gap between human communication and computer comprehension. One of the main reasons why NLP is so important to businesses is that it can be used to analyze large amounts of textual data, e.g., social media comments, customer support queries, online reviews, and sales reports. Some of the most common techniques involve question answering, machine translation, information extraction, and sentiment analysis. Sounds straightforward, right? It might seem so initially, but to develop a solid NLP algorithm, you need to have a linguist, a computer scientist, and an AI expert all in one room. Now, that’s the round table discussion I’d like to look at. Essentially, various computational methods are put together to make sense of written and oral texts –large volumes of them.
However, making sense of human language, aka unstructured data, has an added level of complexity due to syntax, semantics, and various grammatical structures. How many times have you struggled with which tense to use or with the correct conjugation (hello to German speakers)? Essentially, unstructured data can be thought of as qualitative (words), and structured data as quantitative (numbers). To be deciphered by an algorithm, your data inputs should be turned into structured query language (SQL). That is why, in order to create a solid NLP algorithm, you need to first build an NLP pipeline with pre-processing steps that can convert words into numbers. This is what enables an algorithm to understand us. The construction of an NLP pipeline would thus have to involve several pre-processing steps to turn our words into numbers and make an algorithm understand us. A typical pipeline consists of the following:
· Tokenization: breaking down text into single words
· Part-of-speech-tagging: subdividing the text into nouns, verbs, adjectives, etc.
· Stemming: reducing words to their root forms
· Removal of stop words: eliminating articles, prepositions, or words with no contextual value
Having conducted all the above steps, you now have a basis to work with and input for your algorithm.
If you are to jump on the NLP bandwagon, there's quite a menu to choose from. It will all depend on your business needs and the data volumes you're working with. The new generation of Transformer language models, e.g., BERT, GPT-3, and ELMO have unlocked some novel applications ranging from market segmentation to automation of customer support tasks. Let’s take a look at some examples.
“Hi Alexa, …”
While most of us relate the world of voice assistants to the likes of Amazon’s Alexa or Microsoft’s Cortana, there is much more to it. NLP models can convert an input voice into text and vice versa. This is one of the most potent expressions of deep learning, set to revolutionize the generative type of neural networks. The ability of these techniques to create a human-like voice by utilizing only a few audio inputs can underpin a fully-automated support call center, for instance. Here’s a case to consider: you are a multinational enterprise with customer support centers dispersed around the world. Imagine creating a single chatbot in say, English and then layer NLP-based translation for other countries where your company is operating. Automation of customer support functions is made possible with the real-time human-machine conversation, leaving your staff more time to focus on superb experiences and unique human touch in areas where it matters. Think about the added value that can be created if you are to eliminate all the tedious parts like sorting and routing incoming emails or typing in the same Facebook reply to a disgruntled customer for the hundredth time.
Let’s look at two examples of how you could apply NLP algorithms and their benefits.
Automated market analysis and benchmarking
A vital prerequisite for developing a sound business strategy is understanding your customers – how do they feel about your brand? What are their pain points? What makes them tick?
Although holistic market research goes a long way in this process, getting a full picture is subject to an asymmetry of information. NLP can help you bridge this asymmetry by actively “listening” to your target audience and drawing insights from multiple relevant sources. Be it social media, blogs, news websites – you name it – the NLP model can crawl the web and aggregate only relevant data, depending on pre-determined parameters.
Want to delve deeper into the market research? Consider NLP-powered benchmarking. You feed the algorithm with relevant parameters – competitors' names, industry trends, etc. – and the model automatically puts together a report on what has been found. Marketers and strategists can then make well-informed decisions on how to position your brand. Impressive, right?
As the old marketing adage says, “listen to your customer and you shall be rewarded”. Enterprises usually have large quantities of customer-generated data sitting idly by – product reviews, emails, chatbot data, social media comments, etc. How can you make sense of this valuable data coming from such distinct sources?
Sentiment analysis and content classification are used to identify patterns and understand the tonality of customer feedback (positive, negative, or neutral). Based on the data you’re trying to extract, the algorithm will search through the web – like Facebook, Twitter, TripAdvisor - and then classify aggregated comments with a positive, neutral, or negative value. This methodology is called opinion mining and, if applied over time, it can revolutionize the way you approach branding, advertising, and marketing. Having customers' opinions as a backbone of your communication not only allows you to adjust the service delivery but also to decrease the churn rate and make your customer heard.
Should I jump on the bandwagon?
So, how can you use different NLP applications to make better use of your data to customize customer experiences and increase customer service efficiency? That will, of course, depend on your current business strategy and on how NLP can support you in realizing it. There is a full catalogue to select from, ranging from sentiment analysis, pattern recognition or automatic chatbot translation. If your enterprise is inundated with large quantities of incoming customer messages and you’re still processing them manually, there’s a high likelihood of you not replying to all of them. Next thing you know, your brand reputation is dampened and the Net promoter score figures ring alarm bells. Guess what, NLP-empowered process automation wouldn’t allow this to happen – you would be able to respond to every single message in the most customized way for your customers.
Having hands-on experience with leading tech and telecom companies globally, we see that AI and ML will continue to shape key dynamics across different sectors of the global economy. Understanding the value these solutions can bring to your customer service, branding and marketing can be critical to your competitive advantage. We can support you in exploring how NLP can make the best use of your data and applying that to your marketing and communications initiatives.
Zlata Kolesnyk is a Strategist at Seventy Agency with extensive AI/ML and business process optimization experience from leading a tech start-up as well as working for ICTs-focused NGO.