AI has been among the trending topics for quite some time now, and it seems that it will take part in our daily lives as marketers.
If it didn’t already.
And this is a valuable asset to have since it can improve the entire process for both brands and customers alike.
But don’t take my word for it.
For this webinar, we gave the mic to our colleague, Alejandro Díaz Ortiz, Chief Artificial Intelligence Officer here at Creatopy, to tell us more about the vast potential of AI in the marketing department.
I turned the webinar into a written format for those who absorb the information better like this. Otherwise, you can access here the “The Age of AI-Powered Marketing” webinar.
Now, let’s dive into the subject.
A. What’s AI?
There are a lot of definitions regarding AI. Some of them are on a more positive note, while others send shivers down your spine.
To better understand what AI is, we should start by mentioning its two categories.
1. Artificial General Intelligence (AGI)
Artificial General Intelligence is the ability of a digital computer or computer-controlled robot to learn or to perform tasks commonly associated with intelligent beings. When people say that AI represents a risk for humankind, they refer to AGI.
2. Artificial Narrow Intelligence (ANI)
Artificial Narrow Intelligence is a computer’s ability to learn a specific task without being explicitly programmed. Most of the successes created by AI are being achieved by ANI.
In this webinar, Alejandro discussed the ANI, which improves our lives day by day.
Thus, we will refer to ANI when we mention AI in the article.
Five years ago, we did not think self-driving cars could actually exist or that a streaming service like Netflix could recommend what to watch based on previous interaction.
But everything is a reality today with the help of AI. It learned image recognition and labeling, which help radiologists provide the correct diagnostics and even language models such as GPT3.
GPT3 is an autoregressive language model that can create essays, poetry, or news reports almost as good as humans can using just a tiny amount of input text.
All these systems mentioned above used ANI, particularly a set called machine learning.
Machine learning is based on algorithms that can improve automatically through experience and by using data.
Based on machine learning, there are four types of AI.
- Supervised learning
- Unsupervised learning
- Reinforcement learning
- Semi-supervised learning
Let’s take them one by one.
1. Supervised learning
Supervised learning is a mathematical data set model containing both inputs and outputs. Most of the economic value of AI comes from this type of supervised learning.
You know that quantum mechanics has Schrödinger’s cat.
We can explain AI supervised learning with a cat example. After all, aren’t they the best?
So, let’s say you have several images (three, in this case) which represent your input information. Then, you need to build your mathematical model that classifies your images according to a labeled output (e.g., cat and dog), which represents the output information.
Then you train your AI to recognize the images according to the labels.
You train it because you want to arrive at an unlabeled image where you can ask your AI a simple question like: “Is this a cat or not?”. If you train your AI well, it may offer a response with a 90% accuracy rate.
When training your AI, the key to success is the way we represent the information between humans and machines.
“There are a lot of things that AI has to do with no rules.”
For a computer, a cat is no different than any other image. It’s pixels and RGB, where every pixel is a piece of information. What AI is doing is finding what part of the information you are giving correlates better with the fact that you labeled the image with a cat or a dog, as per our example.
To train your AI, you need to offer as many examples as possible, so when you provide an unlabeled image, it will not always give the right answer, but most times, it will offer you the correct one.
Knowing all of the above, how do you take advantage of AI in customer churn prediction, for example?
“Simple, creative questions are your priority.”
Usually, marketers represent their customers based on demographics, psychographics, industry, location, happiness index, usage, response to previous marketing campaigns.
So instead of images, you take all of this information about your customers and transform it into your input information. Depending on the situation, your labeled output for these customers will be “churn” or “no churn.”
Then, similar to the cat example, you train your AI with multiple models to respond to a simple question “Will this customer churn or not?”
When you pass these customer genome databases to your AI, it will recognize all the labeled cases.
How is this useful?
For example, once you know which customers will not churn, you can start a retention campaign or a loyalty campaign.
This supervised learning provides a framework. Depending on what information you want your AI to offer you, you change your input (the customer genome), labeled output, and the question.
The same goes when you want to see if someone will finish a form or abandon the funnel.
This time your input will be the ad tag, the browsing time, site pages visits, heatmaps, responses to marketing campaigns, and the labeled output will be “convert.” Your question will be: “Will this customer convert?”
2. Unsupervised learning
Unsupervised learning aims at finding structure or correlation in data that contains only inputs.
The best way to explain this is with an example.
Let’s say you want to increase your customer engagement with personalized and targeted messaging.
You want to take all your customers’ genomes based on all your information and transform everyone into your buyer persona.
Information is everything from demographics, psychographics, industry, location, happiness index, usage, response to previous marketing campaigns, and more.
This sounds very ambitious, but AI will be able to do it for you. With the right data, you will have a microtarget at the level of every customer, and you will be able to improve your customer engagement at all these levels.
3. Reinforcement learning
Reinforcement learning aims at learning through “agents” in an interactive environment, where rewards and punishments are used to achieve an objective.
4. Semi-supervised learning
This type of AI learning aims at building a mathematical model of data set containing both labeled and unlabeled data.
Now, you may ask, where is the deep learning in all of this?
Here’s an example that will help you understand better.
You can have a pick-up truck running on gasoline, another on diesel, and another one has an electric engine.
Their difference is the engine.
The same goes for deep learning. The difference is in the engine of the models as they have different effectiveness levels and approaches.
These can be:
- Artificial neural networks (deep learning)
- Decision trees
- Support-vector machines
- Regression analysis
But, we won’t run into this subject as we will distance too much from the purpose of this article.
You know that AI is out there, but how exactly do you know if your company is ready to take advantage of all the benefits AI offers you?
Short version: when the marketing department is ready to use data to inform the decision process.
Or, the more extended version: when your company finds its blueprint to help navigate your digital transformation.
The blueprint has to answer five questions:
Transformation starts with people and ends with people. You have to change the mindset, the company’s culture for scale, the alignment, and the collaboration across the company to achieve transformation.
To find if you’re prepared for AI investment, you need to have data.
Tons of data may help, but not all data is equally valuable.
Collecting data per se is useless. You need data that is always related to providing value to your customers.
Also, keep in mind that you need important and accessible data that respects the guidelines regarding data protection.
You have the mindset, the data, but do you have the capacity to extract insights from your data?
Finding the why will help you discover a solution and optimize it. If you use AI, you must remember that your questions must be precise and simple.
Once you have your why and the possible solutions to your problem, you need to decide what to do with them. AI always complements your natural intelligence and your company’s talent. It will show you insights but will never tell you what to do.
5. What’s the outcome
In this final step, you need to analyze if you’ve come to an acceptable solution. Maybe you need more data, or perhaps you need to define what success is for your company.
Now that we’ve scratched the surface of what AI is and its categories, we should also know how to use it in marketing.
1. How do you implement an AI strategy in marketing?
It’s useful to separate all the different AI solutions that marketing has at its disposal in four blocks, from the most basic to the more advanced and valuable.
- Basic stand-alone
Here we refer to the basic consumer automation systems that operate simple tasks and don’t learn from historical data, such as email automation, chatbots, etc.
- Complex stand-alone
In this situation, you have an AI trained on generic data, and your marketing operation sits on top. You query your AI, receive insights and apply what you’ve learned in your marketing strategy. Think about the IBM tone analyzer that interprets the voice’s tone of the customers.
- AI offline
The AI offline is a better solution because it uses the company’s historical data.
By doing so, AI can provide a high level of automation tasks with minimal user input. The AI offline works with deep learning and can provide insights into customer segmentation, customer response prediction, and action recommendation.
- AI online integrated
The most valuable AI solution for your business is having AI embedded in the company-wide data workflows. This means you have a collection of AI doing many things and ingesting data from all the departments.
All the insights you receive from AI can help the marketing or sales departments and even your customers.
Everything will work in a unified framework.
To better understand the AI online integrated, think about Netflix. You don’t download their database because it happens in real-time.
The same goes for your business. You have live recommendations to both customers and marketers. You learn more about what customers like and how to make the product more appealing.
This type of AI integrated online is more and more used in customer relationship management like Salesforce.
2. Where do you start?
Once you know AI is right for your business, you can follow this checklist to have a clear guideline and a strong headstart:
- Simple projects with well-defined objectives and metrics.
- Check your company’s degree of digital readiness and ask for help from the data science/analytics team.
- Consider the possibility of outsourcing this first project.
- Show clear benefits in adopting AI.
Once you see the benefits of AI, you can get into more intricate details.
For example, you can collect more data and upscale your data team, or, why not, you can implement an offline AI project that includes feedback from other departments.
“Remember, AI won’t tell you what to do.”
AI will give you all the information you need if you ask the right questions, but you, the marketer, need to be at the helm.
You control the further actions, iterations, and experiments based on the provided information.
You know what AI is, what it can do, how it can help your marketing strategy, how you can integrate it, but you must keep in mind some do’s and don’ts when it comes to AI.
- Under or overestimating what AI can do
You should have realistic expectations of what AI is. If you invest in AI but don’t use it, it won’t benefit you.
Or, the opposite, you want to create such an unrealistic application that AI is not yet at the level you need it to be.
- Asking the wrong questions
I can’t stress enough the importance of asking your AI the right questions.
You can have all the magic in front of you, but you can touch it only if you ask the right questions.
- Not knowing the value of being right and the cost of being wrong
For those of you familiar with statistics, you might recognize this as type 1 type 2 errors from hypothesis testing.
AI will offer you the right answer most of the time, but it has its accuracy limitations. So, you sometimes need to be prepared for the cost of making a decision based on an incorrect response.
- Always remember about ethics, data privacy, security, and data ownership
“When you collect data, you need to provide value to your customers.”
Data regulations laws, such as GDPR, and many others, shape your data collection strategy.
More regulations are coming in a cookies-less world, and they are based on a few principles: transparency, fairness, purpose limitations, confidentiality, accuracy, storage limitations, integrity, and you have to have someone accountable.
“Start early, fail fast.”
The number of experiments that Amazon, Facebook, Google, Microsoft, and other companies run every year is more than 10.000.
They do this because experimentation allows collecting valuable data that you can offer to your AI to make it smarter.
Good data will offer more precision in your AI answers.
For reaching a high reward, start early and find short-term metrics that are good predictors of long-term outcomes.
Then separate the data relevant to your marketing campaigns and feed this first-party data to your AI to get more precise predictions. Then simply repeat the process.
It doesn’t mean that all the experiments you will make will be successful, but the more experiments you make, the greater the chances of finding the right answers.
- AI marketing strategy rest upon your company’s digital readiness.
- Start simple, get results, improve the quality of your data, then move to more complex processes.
- Experiment early, fail fast. Do this on repeat.
- Benefit the most out of your AI by asking the right questions. Always remember the value of being right and the cost of being wrong.
- AI is there to assist you and accompany your natural intelligence. The final decision is yours.
Audience: I’m interested in revisiting the Digital Marketing Genome- do you have any recommendations on how we can create them for our customers?
Alejandro: Every customer is a row, and every piece of information is a column in a table. For instance, one column could be the landing page through which a customer arrived (you will need a list of all landing pages, and only the one in question will have a 1, the rest will be identified with 0’s), other could be the marketing campaigns have targeted this customer, the outcome (open or not email, click in a link), and so on.
When you think about it, this digital customer genome, constructed in the way I described above, will give you everything you know about her. Since you are thinking at scale, i.e., hundreds or thousands of customers, each and all of them characterized by a digital genome of hundreds of attributes (the columns), you will need an AI system to help you find the correlations between this digital customer genome and the question you are interested in, for instance., what is the probability that a new customer will react positively to a given type of campaign.
We hope we have offered you some valuable insights on what AI does and how it can help you with your marketing strategies to bring you benefits.
If you want to hear what else Alejandro said, you can watch it on-demand here.
Till next time!