Thanks to advances in the Natural Language Processing (NLP) subfield of machine learning, it is now possible to rank text, such as online reviews and social media posts, by favorability. We can even predict the feelings – such as anger, fear, or excitement – that are likely expressed by the author. We can subsequently extract all the text in which a person, brand, or product was mentioned to compile a reputation index, and suggest ways to improve it. This is made possible by deep learning advances such as recurrent neural networks with attention mechanisms.
2 - Customer personalization
The more you learn about a customer, whether it is behavioral data such as click pattern, or descriptive data such as age, gender, and hobbies, the more machine learning models can devise a personalized experience to them. Customers have higher than ever before and this trend will not slow down any time soon, so delivering a high quality customized experience and support should be on the priority list of every marketer. Another area of customization is custom pricing: accurate risk models allow the company to quote the best price to a customer based on their situation and history, increasing the likelihood of converting a lead.
3 - Lead scoring
Several classical lead scoring tools exist and can be purchased or developed in house. They often rely on human intuition and experience of what constitutes a good lead and whether they are marketing qualified leads, sales qualified leads, or dead ends. The main drawback of these tools is their generic nature. As seasoned professionals know, no two markets or industries are the same, and customer psychology changes greatly with context. Neural network-based models cut the guesswork and find hidden patterns in the data that are often impossible for a human to identify, and provide a more accurate scoring. Most importantly, these predictive learn patterns from YOUR data, not somebody else's, and therefore give you the best and most accurate predictions for your leads.
4 - Customer churn and lifetime value modeling
Customer lifetime value and churn modeling go hand in hand. The churn risk of a customer measures how likely they are to stop interacting with the business while customer lifetime value is a prediction of how much revenue they are likely to generate for the business if they remain a customer. Machine learning can model both of these quantities to allow the marketing and sales teams to intervene with high churn risk, high lifetime value customers and bring them back with incentives. The retention strategy can even sometimes be automated.
5 - Customer segmentation and discovery
Unsupervised algorithms such as K-means clustering or DBSCAN can identify patterns in your data and reveal how your customers are clusters by factors such as age, income, address, interests etc... The more information you obtain about your customers, the more accurate the clusters will be, which is a general rule in machine learning. You can then label those clusters and customize your approach to their needs and expectations.
6 - Recommender systems
Propensity models serve to upsell and cross-sell customers during their online purchases. 35% of what consumers buy on Amazon and 75% of what they watch on Netflix come from dynamic product recommendations (Source). Recommender algorithms can also be used to optimize message accuracy while targeting customers, thus reducing marketing waste. The latter would recommend ways to approach customers and topics that might be of interest to them.
7 - Chatbots and virtual assistants
If you've recently interacted with virtual assistants such as Siri or Alexa, you might notice how far they've come. In fact, the future of marketing might just be full of highly sophisticated chatbots and virtual assistants. The field of natural language processing and voice recognition have evolved by leaps and bound in the past few years and their progress has shown no sign of slowing down. What's more, AI giants such as Google and OpenAI provide pre-trained models for a reasonable price or even sometimes free.
8 - Minimizing marketing regret
If you're a marketer, you're certainly familiar with A/B testing. The issue with A/B testing is that a lot of opportunities are lost while testing. With algorithms such as multi-armed contextual bandits and reinforcement learning, the opportunity loss - or marketing regret as it is often called – is minimized, as the algorithms naturally sample the better options more often.
9 - Text extraction and summarization
Another exciting application of natural language processing is the ability to automatically extract and summarize text. Applications of this can be the quick processing of news articles after a major product launch to gauge the market reception and correct the course of the launch if necessary.
10 - Marketing mix optimization
Often, marketing portfolio mixes are based on experience and intuition. While those can work remarkably well, machine learning can once again take out the guesswork and provide optimized solutions. Algorithms will typically look at previous marketing spend on various channels (online CPC, CPM, radio, TV, etc...) as well as sales, and output an optimized allocation of funds to each channel to maximize return on investment.
11 - Computer vision
Image recognition is used by some companies to detect when their branded collateral is posted on social media or in blogs. A famous study by Gumgum and Miller Lite utilized both text analysis and computer vision to analyze millions of user-generated content (on social media, blogs and such) to uncover ways to connect with their consumers and promoters. In the end, they revealed that they found 1.1 million posts, 3.2% of them were images with no relevant text, meaning they were found by the power of computer vision alone.