Breaking down “artificial intelligence in marketing”
The other day, I was at a meetup of retail leaders in Vancouver.
There were about 20 of us in a room, listening to experts present on the future of the industry. Topics ranged from conversion optimization to omnichannel retail, to digital marketing best practices and digital transformation.
The presentations were strong, but one moment, in particular, struck me.
Near the end of one expert’s deck, he brought up a slide titled “Machine Learning and Retail”. He chuckled and said something along the lines of “you can’t have a presentation today without mentioning ‘machine learning’, so here it is”. With that statement, he quickly moved on and wrapped up. He didn’t actually talk about machine learning at all.
And I thought to myself, that is the definition of a buzzword.
Business and marketing leaders, today, know that artificial intelligence (AI) is a topic of conversation. They know they need to know about it. They know they should plan for it. But over and over again, we see examples of leaders who don’t understand artificial intelligence. Who don’t understand how to plan for it.
At WiderFunnel, we are in the test and learn business. One of our core company values is “curiosity”. The desire to understand why and to provide data-backed next steps is what drives us. So, in today’s post, we are going to start tackling the AI beast, from a strategic marketing perspective.
My goal for this post is to help you:
- Understand the capabilities and limitations of artificial intelligence, right now,
- Understand how you should evaluate and incorporate AI into your organization, and
- Understand how to plan for the future of artificial intelligence
Let’s start at the very beginning: What are we really talking about when we talk about “artificial intelligence”?
Editor’s Note: This post is over 5,500 words. To help you navigate, we’ve included a Table of Contents below.
- What is artificial intelligence…exactly?
- The 3 main artificial intelligence applications (in marketing)
- What does all of this mean for marketers?
- How to evaluate and incorporate artificial intelligence into your strategy
- 1. What business problem can artificial intelligence help you solve?
- 2. What application of artificial intelligence can help you solve your business problem?
- AI and customer retention
- AI and content creation
- AI and personalized recommendations
- AI and marketing experimentation
- 3. What are the risks and biases inherent in AI right now?
- What the future of artificial intelligence holds for marketers
What is artificial intelligence … exactly?
I propose to consider the question: ‘Can machines think?’-Alan Turing in Computing machinery and intelligence (1950)
Artificial intelligence has been an almost mythological concept for years. Visions of sinister, thinking machines from iRobot, The Terminator, and Black Mirror spring to mind. For many of us, AI is synonymous with an exact replica of human cognitive processes. It is both threatening and awe-inspiring.
Over the past two decades, with advances in computer technology, artificial intelligence research has hit its stride. Research breakthroughs have led to technological advances and, in 2017, AI has become a very real possibility. At least, that’s what it seems like. I mean, everyone is talking about and writing about AI.
But the reality of artificial intelligence technology doesn’t quite align with the hype.
In the next section, I’m going to try to break down the buzzwords. To understand what AI is and isn’t, what it can and cannot do, in the context of marketing. Stick with me: Things are going to get a little technical.
Note: There are multiple definitions of artificial intelligence and its applications, so I’ve included a resource library in this post (see below). You should also know that this post will focus on AI as it relates to the field of marketing. Applications in other fields will not necessarily be discussed.
The 3 main artificial intelligence applications (in marketing)
When someone is talking about “artificial intelligence” in marketing, they are rarely talking about true artificial intelligence. More likely, they are talking about one of the three following applications:
- Machine Learning
- Deep Learning
- Natural Language Processing
Ok, great. But what do those terms mean? Let’s dig into ‘em.
The Basics: Algorithms
In math and computer science, an algorithm is “is an unambiguous specification of how to solve a class of problems”. We’re starting here, because algorithms are at the core of artificial intelligence technology. Typically, an algorithm is programmed to take a directed path once it encounters new information (like in the example below).
One of the simplest algorithms is to find the largest number in a list of numbers of random order. Finding the solution requires looking at every number in the list. From this follows a simple algorithm:
- If there are no numbers in the set then there is no highest number.
- Assume the first number in the set is the largest number in the set.
- For each remaining number in the set: if this number is larger than the current largest number, consider this number to be the largest number in the set.
- When there are no numbers left in the set to iterate over, consider the current largest number to be the largest number of the set.
Further reading on algorithms
- “What is Code” by Paul Ford in Bloomberg
- “How Algorithms Shape Our World” by Kevin Slavin, a TED Talk
- “A Beginner’s Guide To Artificial Intelligence, Machine Learning, And Cognitive Computing” by IBM
Machine learning describes algorithms that can learn. There are two main categories of machine learning: supervised and unsupervised. (Though, machine learning can also be semi-supervised.)
Supervised Machine Learning
When a person supervises machine learning, a machine learning algorithm is given a teaching set of data to begin (input) and the possible outcomes (output).
Let’s say the input is a teaching dataset of product images for a sports apparel brand. The machine is tasked with classifying the images by product category.
In this dataset, there are images that are labeled as “running shoe” and images that are labeled as “hiking boot”. The algorithm then defines each image as having or not having the characteristics of the label, until the whole dataset is classified.
The person supervising the machine learning can double check to make sure the algorithm is learning appropriately with each new product image. When the algorithm is shown a new image of a running shoe, it can predict that that product image belongs to that merchandise category based on its training.
Unsupervised Machine Learning
In unsupervised machine learning, there is no teaching dataset. The algorithm explores all of the data it receives (the input) to identify patterns and associations.
If the algorithm is given the sales history for your customer base, it can explore the data to find patterns and similarities among buyers. These patterns and similarities can lead to customized product recommendations for your shoppers.
In both supervised and unsupervised machine learning, the algorithm is able to analyze historical data at a rate far surpassing human ability.
Further reading on Machine Learning
- “What Is the Difference Between Artificial Intelligence and Machine Learning?” By Bernard Marr in Forbes
- “Supervised versus Unsupervised Machine Learning: What Is The Difference” By Bernard Marr in Forbes
- “Avinash Kaushik On Machine Learning And Artificial Intelligence For Marketing” on 6 Pixels of Separation, a podcast
- “The Simple Economics Of Machine Learning” by Ajay Agrawal, Joshua Gans and Avi Goldfarb in The Harvard Business Review
Deep learning occurs when an algorithm is able to process larger datasets and solve more complex issues because it teaches itself the rules.
Deep learning occurs in information layers. Which means that the algorithm builds off of the findings of one layer of information to create a deeper “network” understanding of the data. So, it needs massive amounts of data to truly work.
In one deep learning technique, called clustering, rules are established through a series of questions (called “neural networks”). These questions are true or false, or represent numerical data. And they are asked over and over, until the machine learns all the answers to the questions. The answers become the algorithm’s new rules; these evolve every time the algorithm encounters new data.
When new data becomes available, the algorithm provides another layer of information to the dataset. In this way, the program gets smarter and learns from its mistakes.
With deep learning, it’s not just that it’s really smart. It’s that as the network effect builds in, it can get smarter with every successive generation than the entire sum of all knowledge up to that point. And that is something that, at the moment, is just impossible for humans.-Avinash Kaushik, Six Pexels of Separation
Further reading on Deep Learning
- “But *What* Is a Neural Network | Deep Learning Chapter 1”, YouTube Video
- “Why Deep Learning Is Suddenly Changing Your Life” by Roger Parloff in Fortune
- “What Is the Difference Between Deep Learning, Machine Learning and AI?” By Bernard Marr in Forbes
Natural Language Processing and Generation
Your shoppers are being conditioned to communicate with Siri or Alexa as a tool, without necessarily understanding that these machines filter speech using an application of artificial intelligence called Natural Language Processing.
Part of Natural Language Processing is the ability to understand language. To detect sentiment. To understand the complexities of a question.
Let me explain.
If you ask Siri a question, she first needs to understand what you are asking in common speech, even slang. Once she understands, only then can she search for the answer online, translate it into text, and decide what pitch, tone, and pacing to use when she recalls the information through her own artificial speech.
Just like reading is associated with writing, natural language processing is related to natural language generation. When data is transformed into text, it is called Natural Language Generation.
Note: Machine learning and deep learning can be used in natural language processing, but it is a separate branch of computer science.
Further reading on Natural Language Processing
- “Overview of Artificial Intelligence and Role of Natural Language Processing in Big Data” From Hackernoon
- “Rise of Machine Learning, Artificial Intelligence & Natural Language Processing: Deciphering Intention and Meaning” on Medium
- “What Is Natural Language Processing and How Does it Work?” By Trevor Jackins on Neospeech
Ok, I know that was a lot. But it’s important to understand how each of these applications work, even at a high-level. Because “artificial intelligence” can be an overwhelming concept.
Now, let’s get practical and talk about how these branches of artificial intelligence actually apply to your life as a marketer.
What does all of this mean for marketers today?
Companies are accumulating thousands of records from each customer touch point: What customers search for, how long they hover over a product without purchasing, whether they search on a tablet or a mobile phone, etc.
You can even trace a person’s customer profile to their online footprint, taking into consideration their unique likes and dislikes.
These large sets of data ― commonly referred to as “big data” ― provide precise information about what matters in your business. But the difficulty with big data is making it actionable: How do you sift through and leverage all that information?
Enter Artificial Intelligence.
A brief story…
Under Armour, the sports apparel brand, has been competing for market share since it launched in 1996. But how do you make the most noise when you’re up against Nike and Adidas?
Well, they chose to invest in data.
In 2013, the company began acquiring fitness apps: MyFitnessApp, Endomondo and MapMyFitness.
Each of these apps tracks the daily health, fitness, nutrition, and sleep patterns of millions of users. The company also implemented a new sales platform, integrating transaction data with the lifestyle data they were gathering from their apps. Both information sources combined to create a “single-view of the customer.”
Early in 2016, IBM Watson and Under Armour announced a partnership that would make the massive amounts of data Under Armour was gathering actionable.
This partnership is an ideal marriage of sophisticated artificial intelligence and marketing. IBM Watson is a technology that leverages multiple applications of artificial intelligence: natural language processing, visual recognition, machine and deep learning, and human-computer interaction. It’s called cognitive computing.
In the case of Under Armour, IBM Watson is able to analyze lifestyle data in combination with sales data and the latest academic research to provide customized health and wellness coaching to users.
IBM Watson compares a user’s nutrition, training and sleep information with other members of the community of the same gender, age and activity level. The comparison allows the machine to provide recommendations on nutrition, fitness regimens, sleep quality and length, and even on how the weather forecast will affect a workout.
For example, a man who only sleeps five or six hours a night could receive feedback telling him that other men in his age category and at his fitness level who increase their sleep to seven or eight hours, have a lower body mass index.
The more data provided to the apps, the more meaningful the fitness training and recovery recommendations become. IBM Watson learns more and more with each new user and with each user’s increased participation.
It’s advanced personalization. Under Armour is using cognitive computing to create a one-to-one experience for over 200 million users and counting.
At Under Armour, we believe you need to approach consumers like a hotel concierge who deeply knows his or her guests. A concierge knows all of your preferences and the context. You have an incredible experience because it is highly personalized and memorable.– Paul Fipp, Chief Technology Office at Under Armour. Quoted in Forbes.
Pretty incredible, right?
But hang on. Before you begin lobbying for artificial intelligence at your organization, it’s important to understand that a major investment in artificial intelligence platforms does not make sense for every business scenario, nor every business.
How to evaluate and incorporate artificial intelligence into your strategy
The reality is that many of the tools being marketed as “artificial intelligence” right now are still primitive in form. You may feel an urgency to adopt these technologies, but AI, just like any other technology or tool, is not a magic solution.
Understanding how AI can (or cannot) solve your unique business problem is essential. Ultimately, a person must decide how to leverage the technology. And how. And why.
As advances in AI make prediction cheaper, economic theory dictates that we’ll use prediction more frequently and widely, and the value of complements to prediction – like human judgment – will rise.– Ajay Agrawal, Joshua Fans and Avi Goldfarb, How AI Will Change Strategy: A Thought Experiment
If you’re considering an AI tool, you should ask yourself these three questions:
- What business problem(s) can AI help me solve?
- What application of artificial intelligence can help you solve your business problem?
- What are the risks and biases inherent in AI right now that could potentially affect my business?
1. What business problems can artificial intelligence help you solve?
Artificial intelligence technology can lead to reduced labor costs, optimized production and operations, more efficient timing and delivery, more precise customer personas and journey information, and more.
But in order to leverage artificial intelligence, your data must be up to standard.
You may be ready for sophisticated artificial intelligence (like Under Armour), if…
- You have collected massive amounts of data.
- The data is accessible and ready to be processed.
- You can prioritize an investment in the most sophisticated technology (i.e. IBM Watson).
The majority of companies are not here, yet. Most of us are in one of the following situations:
- My company does not have enough data for sophisticated AI (a startup, for instance)
- My company has data that has not been managed well, so it can’t be processed as is
- My company has data, but internal silos between departments make it inaccessible
- My company’s data gathering processes are not internally automated, so AI analysis would be out of date
Data collection must happen long before any implementation of AI. Don’t be discouraged if your company data is not ready for IBM Watson, though. There may still be areas within your marketing where machine learning platforms can be beneficial.
Another major factor in determining whether or not to implement AI technology in your business your team. Does your team have the knowledge and skills to make AI worthwhile. For instance, if your team doesn’t understand the risks and limitations of AI, they may feed the machine bad data, which would skew all of its results.
And when the AI tool offers many different actionable items based on its analysis of the dataset, your team will ultimately have to decide what to do with that information. Your team’s ability to evaluate the consequences and tradeoffs of each actionable item will be an important skill to foster internally.
AI can assist, but it cannot replace. I think of AI not as artificial intelligence but instead “augmented” intelligence such that the software and technology augments human intelligence and decision-making skills.– Andrew Stephen, Associate Dean of Research & L’Oréal Professor of Marketing at the University of Oxford – Saïd Business School.
The question then becomes less about what the machine can and cannot do, and more about how the machine and the marketer can collaborate. That’s what we will explore in the next section.
Further reading on AI evaluation and implementation:
- “When It Rains Data, Focus On Customer’s Needs, Skills And Good Governance” by Andrew Stephen in Forbes
- “Data Democratization at Airbnb” by Reeport
- “If Your Company Isn’t Good At Analytics It’s Not Ready For AI” by Nick Harrison and Deborah O’Neill in Harvard Business Review
- “AI Is Changing Marketing As We Know It, And That’s A Good Thing” by Andrew Stephen in Forbes
- “The Fatal Flaw of AI Implementation” by Jeanne Ross, MIT Sloan Management Review
- “The Road to Artificial Intelligence: A Case of Data Over Theory” by Nello Cristianini in New Scientist
- “What to Expect from Artificial Intelligence” by Ajay Agrawal, Joshua Gans and Avi Goldfarb in MIT Sloan Management Review
- “Is Your Business AI Ready?” A Report by Fortune and GenPact
- “How To Stop Worrying And Love the Great AI War of 2018” by Harry McCracken in Fast Company
- “The Customer Experience Is Written in Data” Report by Google and Econsultancy
2. What application of artificial intelligence can help me solve my business problem?
The simplest way to think about how artificial intelligence platforms can augment your marketing strategy is to break it down by marketing priority. In this section, you should start to get a picture of how to think about, evaluate, and incorporate AI tools across your marketing spectrum.
We are going to look at how AI might fit into the following four marketing priorities:
- Customer retention
- Content creation
The most important thing to keep in mind as you read through this section is that tools, even artificially intelligent tools, are not magic. They must always support a larger strategy and business goal.
AI and customer retention
If creating a delightful customer experience is an important priority for your business, artificial intelligence tools that facilitate customer service may be an option.
The chatbot is arguably one of the most exciting innovations in customer service in the past decade. With the advent of digital media, the need to be on call for your customers 24/7 has only increased. Online messaging and chatbots have been positioned as a solution.
Early scripted chatbots could provide templated responses to keywords in a customer’s questions. Now, expectations have changed post-Siri, Cortana, and Alexa. Your customer expects to have a conversation with a chatbot, so anything less than an learning chatbot can be frustrating.
A good indication of an AI-powered chatbot is its ability to handle open-ended questions. These bots leverage natural language processing and generation and machine learning to be able to respond to a customer’s unique enquiries.
A scripted bot, on the other hand, will only answer close-ended questions. It can answer frequently asked questions like “What is your return policy?” or “Where are your stores located?” Or even questions about an item or purchase details.
An AI chatbot can track customer behavior and sales so that it can recommend products. And in some cases, it can even provide interesting banter with your customer. But even the most advanced chatbot is still limited in human conversation.
AI and content creation
If you’re facing a need to produce more content, artificial intelligence tools may be able to support your efforts. But they likely won’t solve your problem overnight.
Content creation platforms, like Wordsmith or Quill, produce template pieces, like financial reports, sports updates, local happenings, and business news. These Natural Language Generation tools scour past reports and updates, and create new content based on their learnings.
Media outlets, such as Forbes and the American Press Association, have adopted this type of artificial intelligence tool to produce more articles, quicker. However, an editor still needs to review the content before it’s published for a human reader.
Tools like this will, theoretically, free up journalists and content creators to do more substantial reporting, rather than spending time creating formulaic content. In fact, with the right subject matter, it can be hard to spot the difference between content created by a human and that created by an algorithm.
Where artificial intelligence applications really succeed in content creation is in editorial planning. Machine learning algorithms can find emerging topics to ensure that your content will be relevant to your audience by the time it is posted. They can also help you keep up to date with changes in search engine optimization.
In the near future, this technology may develop enough so marketers can create content that is adaptive to an audience for more personalized messaging. Right now, “smart” content like this is still quite primitive; smart content consists of content suggestions based on customer preferences and messages that adapt slightly to the user.
Other tools that use Natural Language Processing:
AI and personalized recommendations
One of the best-known marketing strategies currently augmented by artificial intelligence is personalization. Many product recommendation tools use machine learning algorithms to sort through customer information.
In a product recommendation tool, machine learning algorithms sort through product images, and compare them to a customer’s purchase history in order to recommend a new product.
The algorithm knows which running shoes a customer bought in the past, for example. It can compare this purchase history with other customers’ purchase history after buying a similar item. The algorithm can then make a recommendation as to which running shoe the customer might prefer next.
Now, let’s say you want to identify the ideal target audience for an upcoming promotion on running shoes. A machine learning algorithm could analyze your data to find out who bought running shoes, when they bought them, what promotions they’ve responded to in the past, and more. Using this data, the algorithm could predict which shoppers will respond to a promotion on running shoes.
The algorithm continues to learn about a customer. So, the product recommendations evolve from basic product attributes to consider more individualized behavior and preferences.
Remember the pile of data Under Armour is producing? A machine learning algorithm could analyze their customer information (including demographic and behavioral data), customer purchase history, and their inventory database to find patterns to help improve marketing and promotions.
Adding an extra layer, a deep learning algorithm could also mine and learn from the lifestyle data being collected by Under Armour’s fitness apps. It would consider aspects like how frequently a person runs and for how long, how serious they are about health and nutrition, and even if they are getting enough sleep.
This information, in combination with sales data, creates that single view of the customer. And this enables personalized recommendations on products and fitness performance.
AI and marketing experimentation
As experimentation becomes the status quo at leading organizations, business leaders must also consider how artificial intelligence can support optimization and experimentation efforts.
Fortunately, experimentation is our bread and butter at WiderFunnel.
For the past decade, we’ve been refining the world’s best process for conversion optimization. And we have argued that optimization teams must balance testing for short-term lift and testing for long-term success.
Companies often put so much emphasis on reaching certain testing velocities that they shoot themselves in the foot for long-term success.– Senior WiderFunnel Strategist, Michael St Laurent
Mike argues that to have a successful experimentation program, you should be running tests to achieve lift and running tests that are purely exploratory.
One 10% win without insights may turn heads your direction now, but a test that delivers insights can turn into five 10% wins down the line. It’s similar to the compounding effect: collecting insights now can mean massive payouts over time.– Michael St Laurent
Of course, this advice can be tough to swallow. Because experiment velocity is often an indicator of success for in-house teams. As is win ratio. Taking the time to run an exploratory test means (potentially) not running a revenue-driving test.
But advances in machine learning technology might be a solution to the problem of time…in the hands of a capable strategist.
For example, tools like Sentient Ascend enable marketers to try more ideas within the same timeframe as an A/B test, leveraging genetic algorithms.
Genetic or evolutionary algorithms are loosely inspired by Darwin’s “survival of the fittest” principle. Essentially, the algorithm is working towards an optimized page by promoting elements that perform better, and demoting those that perform worse.
It then goes on to “breed” new variations by combining proven best performers together. This goes on and on through multiple generations. Eventually, the algorithm arrives at the optimized combination of every available variable.
The exciting aspect [of tools like Sentient] is that you can test through changes that you otherwise would probably never put together. So it automatically helps you push the boundaries.– Michael St Laurent
But, as great as this all sounds, a tool like this is limited to optimizing based on the concepts it is given. Humans still need to ideate and control the variables that are fed into the tool.
Sticking to a Darwinian metaphor, a strategist is responsible for programming all of the “genetic mutations” (Variations) that are to be evaluated. The genetic algorithm is then responsible for working towards the optimal point, given the variations at its disposal.
Unbounce also recently unveiled its Landing Page Analyzer. Leveraging 8 years worth of landing page data and machine learning algorithms, the tool analyzes your landing page and provides recommendations as to how you could improve your conversion rate.
While the tool is predictive―and constantly getting smarter―the recommendations are still just that: recommendations. They should become another source of experiment ideas. Or, if you have a low-traffic website and can’t A/B test, a tool like this can provide a starting point for making improvements.
“‘Artificial intelligence’ platforms allow optimizers to experiment and gain insights with greater speed. However, the idea that these tools can discover business insights and increase business revenue on their own is far from reality, or at the very least, very premature.
The human element is still required in order to ‘define the rules’ by providing inputs on hypotheses, consumer insights, consumer behavior, human psychology, etc.―all of which are crucial to optimization.
Based on those inputs, these tools then allow faster discovery of the positive combinations, as well as uncover in-depth trends and patterns within segments and cohorts that previously would’ve taken a huge amount of human data analysis to uncover.” – Nick So, Director of Strategy, WiderFunnel
3. What are the risks and biases inherent in AI right now?
Artificial intelligence is opening up doors for marketers, freeing up time and expertise. But if your marketing team doesn’t fully comprehend how to work with the machine, they may unknowingly sway the machine’s outcome.
In this section, I’m going to explore the risks and biases inherent in AI right now.
The limitations of artificial intelligence today
Some of the hype around AI is based in truly extraordinary advances. It’s true that deep learning algorithms can learn with every new layer of information processed by a system, gaining insights at an exponential rate. But it’s important to recognize that the machine does not really understand as we do.
Here’s what I mean.
This past August, Facebook employees decided to test the artificial intelligence capabilities of their chatbots. They assigned two chatbots, Bob and Alice, the task of making trades: trading balls, hats, and books.
If you don’t remember reading about this, there was A LOT of media hype around the fact that Facebook shut down the project. People were asserting that the chatbots were speaking to each other in a secret language, and things got very science-fiction-like.
But here’s what really happened. In the experiment, researchers provided each chatbot with different rewards using a teaching dataset. But they were rewarding the chatbot’s ability to make a trade, not to communicate a trade as humanly as possible.
So, the chatbots started speaking in a language code. Think of it as shorthand they were using to expedite the trade. In this shorthand, the chatbots used words repeatedly to represent numerical values of trade. If they wanted to trade nine balls and hats, for example, they repeated the words “to me” nine times.
“I can i i everything else …….” – Bob, Facebook chatbot
“balls have zero to me to me to me to me to me to me to me to me to me” – Alice, Facebook chatbot
Facebook shut the project down because the experiment wasn’t working. At least not in the way they wanted it to.
They did not create a chatbot that could reason, converse, and negotiate like a human being. They did make strides towards getting the chatbots to compromise — a key tactic of negotiation — through further testing, though.
But you can see how the input dictates the outcome.
And this is a key factor to consider in implementing artificial intelligence. Human beings can affect and influence the outcomes. In this next section, we look at how the dataset we give to the machine can show a bias.
Bias in artificial intelligence
One of the most significant flaws of artificial intelligence technology right now is a vulnerability to bias. Especially if the dataset that is given to the algorithm demonstrates a bias itself.
In 2016, Microsoft created a Twitter chatbot named Tay meant to converse with Millennials. Within the first 24 hours of its release, people coordinated to feed the tool biased messages, essentially training Tay to accept racist and sexist statements as a “rule” of conversation.
The public was outraged. Microsoft pulled Tay offline and issued an apology acknowledging the vulnerability in this type of AI tool.
But even in situations where artificial intelligence tools do not experience a coordinated attack like Tay, the algorithms are likely to incorporate the bias of their creators and the input data. Because machine learning relies on historical data to make predictions for the future, it can reinforce past biases.
Joshua Gans, co-author of Prediction Machines and Professor of Strategic Management at Rutman at the University of Toronto encourages business leaders to “understand these potential biases and be aware of them when making AI adoption decisions.”
It is important to think critically about your dataset(s)―the information that the algorithm’s findings will be based upon. Start by considering the following questions:
- What is the input dataset?
- What is the nature of this dataset?
- Is this dataset complex enough?
- Are the results indicating that we need more data?
- Is the person choosing which data is fed to the algorithm affecting the result?
- If the algorithm isn’t successful, what are the consequences of those failures?
Further reading on bias in AI:
- “How we transferred our biases into our machines and what we can do about it”, You Are Not So Smart Podcast, Episode 115
- “Forget Killer Robots—Bias Is the Real AI Danger” by Will Knight, MIT Technology Review
- “Your Artificial Intelligence Is Not Bias-Free” by Daniel Newman in Forbes
What the future of artificial intelligence holds for marketers
By far the greatest danger of artificial intelligence is that people conclude too early that they understand it.― Eliezer Yudkowsky, Author and AI Researcher
The danger of artificial intelligence is that we can both overestimate its power today and underestimate its capabilities tomorrow. Right now, artificial intelligence in the marketing field is largely predictive, but we don’t know what the future will hold.
Advancements in the field of AI are happening quickly. If you want to be on the cusp of this technological change, here are three recommendations:
First of all, look at your processes and procedures for data governance. When an artificial intelligence tool does make sense for your strategy, will your data will be ready? Think of how Under Armour is integrating their information for a single view of their customer. Are your decisions data-driven and ultimately customer-centric? How can you position your organization for future data analysis?
Secondly, foster an internal interest in artificial intelligence research, even if it seems out of scope for your business. You can achieve this by:
- Hosting lunch-and-learn or Q&A sessions with a guest expert
- Sending your marketing team to specialized AI conferences
- Sharing articles and information (start with this one!) internally to start conversation
- Subscribing to reputable journals and magazines to keep up-to-date on relevant research
- Hiring people who are deeply interested in technology on your marketing team
Lastly, experiment with artificial intelligence tools that can help you solve a business problem. Having a “data-driven”, “test-and-learn” culture means you are willing to research and test possible solutions, and then measure effectiveness once you’ve implemented the tool.
This field is vast and we have only scratched the surface of “artificial intelligence in marketing” in this post. I’d like to invite anyone with feedback, resource recommendations, follow-up questions, or comments to leave your thoughts in the comments section below. This conversation will be ongoing.