The world of marketing is changing, the axis tilted with the advent of the internet and the growth of marketing sub-disciplines that include digital, inbound and social media marketing; Artificial Intelligence has spurred a similar sea change, and as a result, marketing professionals need to educate themselves as we enter the age of machines.
This is the first blog in Humans For AI’s AI in Marketing — 101 series, to be followed shortly by our full course which will lay the foundation for marketing professionals to build upon in becoming “AI Savvy”.
What You Will Learn
If you are a marketing professional you have come across some or all of these of these terms used ever more frequently.
- Conversational Marketing using Chatbots
- Personalization using Recommendation Engines
- Content Marketing Optimization using Machine Learning
- Predictive Targeting using Customer Segmentation
- AI-powered Competitive Intelligence
- AI enabled Programmatic Advertising
- AI Journalism and Automated Narratives using NLG (Natural Language Generation)
- Customer Journey using Contextual Relevance
- Voice Marketing and AI
But before we get into exploring applications of AI in marketing, we should start off with some basics.
Understanding AI and Machine Learning
Artificial Intelligence is a term initially coined in 1956 and rooted in psychology to understand and imitate how the mind works. However with an exponential increase in computer processing power combined with the widespread dissemination of smartphones across the human populace over the last 60 or so years, AI is now most commonly associated with computer science.
What is Artificial Intelligence and Machine Learning?
“If AI is a rocket, the engine is algorithms and the fuel is the database.” Matthew Grob, CTO, Qualcomm
Definition — Artificial Intelligence (AI)
A simple definition of AI is the capability of a machine to imitate intelligent human behavior.
AI can learn, reason, problem-solve, and plan. AI can recognize speech, images, and language, interpret complex data, and even be programmed to move objects (used in robots and self-driving cars)
One of the ways that machines can do all this is using the concept of machine learning, a core discipline in the AI body of knowledge.
Definition — Machine Learning
So how do machines really learn?
Machine Learning can be defined as
“An approach to achieve artificial intelligence through systems that can learn from experience to find patterns in a set of data… and apply it to new data it has not seen before”.
There are two phrases to concentrate on in this definition: “learn from experience” and “finding patterns”.
So in simplistic terms — machine learning is involved in the discovery of patterns.
An Example of Machine Learning
Customer segmentation using a clustering approach for an insurance company.
For example, based on given a data set, we want to divide customers into different segments and machine learning helps us do this through patterns.
To discover a pattern (in order to segment), machine learning uses an algorithm (set of rules). There are different types of algorithms to solve different type of problems in machine learning.
In our current example, we can say the following:
Using clustering algorithms for a database of insurance customers and creating a model using 16 customer features, and ten insurance features, we get a result set that shows the customers of the insurance company are divided into three groups labeled as ‘profitable customers’, ‘potentially profitable customers’, and ‘disinterested customers’.
If you want to dive deeper into understanding machine learning, here is an interesting article on “The Science of Making Machines Learn”
The Process of Machine Learning
A popular Machine Learning approach is the seven-step process:
The Seven Step Process for Machine Learning
- Gathering Data — This first step is very important because the quality and quantity of data that you gather will directly determine how accurate your predictive model will be.
- Data Preparation — where we load our data into a suitable place and prepare it for use in our machine learning training.
- Choosing a Model — There are many models that researchers and data scientists have created over the years. Some are very well suited for image data, others for sequences (like text, or music), some for numerical data, and others for text-based data.
- Training — Often considered the bulk of machine learning — the training. In this step, data is used to incrementally improve a model’s ability to predict
- Evaluation — Once training is complete, it’s time to see if the model works, using Evaluation. This is where that dataset that we set aside earlier comes into play. Evaluation allows us to test our model against data that has never been used for training. This is meant to be representative of how the model might perform in the real world.
- Hyper Parameter Tuning — At this stage, you may want to see if you can optimize machine learning. This done by fine-tuning your parameters
- Prediction — Prediction, or inference, is the final step where we look to answer questions we’ve sought via machine learning, I.E., where the value of machine learning is realized.
If you want to explore further, here is a very popular blog post with a detailed explanation of the above step-by-step process.
Choosing the Right Algorithm
On choosing a model, the right machine learning algorithm becomes an important aspect of the model design. Some of the popular algorithms in marketing use cases are regressions, classification, clustering, logistic regression, k-nearest neighbors.
To get more insights into algorithms here is an interesting read The modern marketer’s guide to machine learning algorithms.
The Significance of Training Data and Evaluation
Training a data set is a very critical and a unique activity as a part of model creation which requires domain expertise. This is one area that the domain experts and the data scientist work together and a powerful reason why domain experts need to understand machine learning fundamentals.
Hyper parameter Tuning and Prediction is a feedback loop to help ensure accuracy in your predictions.
Most AI-driven marketing tools and solutions do not require you to go in depth into the statistical model as they provide easy interfaces to work with your data. Yet, it is important to understand what goes on “under the hood” in order to have some confidence in the predictions that are generated, and for this reason, an overview of machine learning concepts is critical.
Introducing Artificial Intelligence as a Plethora of Technologies
AI can be viewed and dissected from various perspectives. One is viewing AI as an umbrella with a set of technologies under it.
Artificial Intelligence as a Group of Technologies
The following technologies are a part of AI:
● Natural Language Processing (NLP) and Text Analytics — NLP uses and supports text analytics by facilitating the understanding of sentence structure and meaning, sentiment, and intent through statistical and machine learning methods
● Natural Language Generation (NLG) — Producing text from computer data
● Image and Video Analysis
● Computer Vision — Computer vision focuses on techniques and models for acquiring and analyzing images in order to understand objects and scenes in the real world
● Speech Recognition — Transcribe and transform human speech into a format useful for computer applications
● Robotics — the discipline of study which deals with building physical machines that move within an environment with a degree of autonomy
● Robotic Process Automation — Using scripts and other methods to automate human action to support efficient business processes
● Swarm Intelligence — is the discipline that deals with natural and artificial systems composed of many individuals that coordinate using decentralized control and self-organization
For more details here is a great read from Forbes: Top 10 Hot Artificial Intelligence (AI) Technologies.
Introducing Artificial Intelligence Based on Approach to Learning
Though machine learning is one of the most popular forms of learning in AI, there are others which are used. In recent times deep learning and reinforcement learning are finding many applications.
● Expert Systems — algorithms that apply a series of if-then rules to make sense of structured inputs, in the manner of a linear decision trees
Example: self-serving checkouts, ATMs.
● Machine Learning — Algorithms that learn underlying statistical patterns from training data (often labeled), leading to an ability to make predictions for novel data.
Example: Workforce Recruitment, Fraud detection
● Deep Learning — A type of machine learning algorithm, an “artificial neural network” with many layers through which data passes to spot sophisticated patterns.
Example: Translation, Health Diagnosis
● Reinforcement Learning — Programming Approach that uses the feedback mechanism to improve algorithms.
Example: Chatbots, Customer Recommendations
● Transfer Learning — Programming approach that reuses the knowledge underpinning an algorithm in one domain to develop algorithms in another.
Example: Training Autonomous Vehicles, Natural Language Processing
For more details, access this publication on Automation and the Future of Work from thersa.org.