Artificial Intelligence is powerful and will open doors we don’t even know exist at the moment. But there are some steps you can take today to ensure that you are making the most of this new capability and are not left behind.
As this is such a nascent area, it serves to begin with a brief description of Artificial Intelligence and Machine Learning. My assumption is that you have read a dozen different definitions and should be curious about which applies to this particular essay.
Table of contents
So how do we define AI?
As an umbrella term, Artificial Intelligence is advanced computer science that brings us computer vision, natural language processing, and chatbots. For marketing, it brings us customer service conversations that include the question, “Are you a bot?”
It also brings personalized, dynamic content, robots in hardware stores, and the ability to order blue pens by pressing Staples’ Easy Button and simply asking for them.
What role does machine learning play in AI?
Machine Learning provides the engine underneath AI. In a nutshell, machine learning is the art of building systems that can change their mind. Rather than writing strict logic for strict execution, a machine learning system weighs the available data to make a decision about a given goal. Which ad should we show? How should the bot respond to this question? When it gets new data, it changes its previous weighting to accommodate the new reality.
Machine Learning is categorized into Supervised, Unsupervised and Reinforcement Learning. A data scientist will choose a Supervised Machine Learning algorithm when the answer is already known (Is this a picture of a cat?), an Unsupervised Machine Learning algorithm when the answer is unknown (What do my best customers have in common?) and Reinforcement Learning when there is no absolute answer (What’s the best ad to show this type of person right now?).
We can correct the machine when it’s wrong about that cat, go out and find prospects who look like the customer segment it identified, and allow the machine to improve our return on ad spend over time.
The gap between the practical applications of this new tech and the heavy math behind it is quite wide. Aside from diving into a new Ph.D. program head first, how does a marketing executive ensure solid footing, quick comprehension and a competitive edge on the leading edge?
- Brush Up on Your Statistics
- Keep Abreast of Developments
- Experiment – Push the Limits
- Embrace Fuzzy Numbers and Failure
- Be a Change Agent
Brush Up on Your Statistics
Math, as most commonly taught in schools is tortuous and interminable. As a result, most people happily declare, “I’m not a math person!” That’s as unfortunate as people coming out of school saying, “I hate Shakespeare!” The gym coach can wreak a lot of havoc in a classroom. Even if your teacher was sub-par, a high school stats book will be much more helpful than a dense Wikipedia entry.
This is not to suggest that you’ll have to perform impressive feats of math in your head, but you do want to have rational conversations about data and the analysis of that data.
You need to know how to drive a car, the rules of the road, and where you want to end up. You don’t need to know how the car works, but when they say there’s a problem with your transmission, you should know enough to be worried about the cost of fixing it.
If the concepts of sampling, statistical significance, and linear regression are outside your comfort zone, take some time to cozy up to them.
Keep Abreast of Developments
When listening to actual data scientists talking about how all this stuff works, you quickly realize the freshness of this science. It is all so new that they will invariably refer to a canonical paper written in the past three months. If you are determined to dive in and learn ML from top to bottom, you have made a career decision, and not picked up a hobby. Things are moving so fast that you should not dwell on the specifics, but on the concepts.
Listen to podcasts like This Week in Machine Learning and AI to get a flavor of just how nascent and complex this line of research really is. Get to know the language, but only well enough to order off the menu and ask for directions.
That way, when there are articles about Google building Machine Learning machines that are teaching themselves how to build Machine Learning machines, you can read them without a glossary.
Experiment – Push the Limits
In marketing, Machine Learning systems are very good at rote, ‘cognitive’ tasks:
- Finding patterns
- Finding outliers
- Learning from new information
AI marketing tools
There are all sorts of start-ups creating all sorts of AI and ML tools as virtual personal assistants, data-driven decision making, organizing and analyzing visual assets, content creation, and general marketing platforms. Here’s A Marketer’s Guide to AI and 45 AI Marketing Tools to Get Started With.
Bring one or two of these into your workplace and try them out. Explore, experiment, exploit. Eventually, you will find some more useful than others, but for now, learning is the key.
These tools are very narrowly focused and really only so-so at their jobs. However, if you can train a generic Machine Learning tool to do a job as poorly as a human being, it’s still better, cheaper, faster than hiring a human. No breaks, no sick time, no pension and, at least for now, no collective bargaining (joke). Given time and enough data, it will learn to do the job better than a human. There are plenty of well-paid people spending a great deal of time doing rote tasks who can be freed up to do something more creative.
Experiment with tools for:
- Lead scoring
- Meeting scheduling
- Personalizing content
- Inbound e-mail sorting
- Social media monitoring
- Programmatic advertising
- Creating social media messages & ad copy
Embrace Fuzzy Numbers and Failure
When you work with humans who weigh the facts and then change their minds, you count on them to explain why their opinion shifted. Psychologists will tell you that people make emotional decisions and then rationalize their judgment when asked.
Machine Learning systems are based on rationalization, without the benefit of emotions. Rather than check with their gut and determine that they should not eat sushi from a food truck on a hot, Summer day, an ML system tries to correlate every data point at its disposal to come to a mathematical decision. The reason for its decision is math, not logic.
Data scientists will explain just how opaque that process is. Women, between the ages of 18 and 34 are more likely to be interested in a new, floral scented deodorant, says the agency that did focus groups, online polls, and social media research.
The machine says that people who score a) 53.256% more this way, b) 45.980% more that way, and c) have exhibited one of these 273 behaviors in the requisite time frame are more likely to be interested. What do those numbers mean? When there are 50, 100, or 1,000 attributes under consideration, the meaning is mathematical, not understandable by a human. And yet, the results are impressive.
Therefore, you must get used to the idea that probability is your friend. This is very different from accounting. There are no binary choices anymore, there are only shades of gray.
Start by changing how you talk about computers. Take a tip from Zoni Nation on GitHub and never say never. Say, “chances are slight,’ or ‘it’s highly unlikely’.
As much as we crave certainty, nothing in real life is aside from death and taxes. Machine Learning is comfortable with fuzzy numbers and probabilities and has no trouble at all with learning from its mistakes, at scale.
Good judgment comes from experience, and experience comes from bad judgment.
– Rita Mae Brown
Humans are far less forgiving. Fortunately, humans are far more intelligent. We have the capacity to bring all of our accumulated knowledge to bear on any given problem, while that machine can only work on a given task with what little data it is given at the moment.
Computers are incredibly fast, accurate, and stupid. Human beings are incredibly slow, inaccurate, and brilliant. Together they are powerful beyond imagination.
– Albert Einstein
Be a Change Agent
Being an effective marketing executive depends on executive decision making. Solid executive decision making depends on experience and best practices. You can figure out what those best practices should be and take the lead, or wait for everybody else to figure it out and then try to catch up.
Leading is better. Because of Artificial Intelligence and Machine Learning, judgment and boldness will be more valuable than experience and a solid track record.
There are likely to be one or more groups in your organization playing around with artificial intelligence in their spare time. They are attending conferences. They are mucking about with training datasets on cloud-based systems. They are building rudimentary apps. This is certain because artificial intelligence and machine learning are just so fascinating that everybody who works with software and/or data thinks testing some form of advanced, self-learning system is far more interesting than binge-watching the latest must-see television.
The first and most important step is to find the experimenters in your firm and get them excited about using their superpowers for marketing. These methods are in much wider use in product development, manufacturing, fraud detection, healthcare, and so on. Marketing simply hasn’t been on their radar.
Once your data scientists are enthused about the enormous amount of available marketing data and are convinced that they understand the goals of marketing, they will start testing. Make sure you stay in the loop so you can deftly nudge them in the right direction while keeping abreast of your firm’s growing capabilities.
You might start by training a machine learning system to suggest cut-and-paste paragraphs to your customer service reps, personalize some degree of content on your website, or choose the right time of day to send out e-mails. This will be a time-consuming and somewhat frustrating experience. That’s okay. It’s how we learn. After much testing and training, it will be time to open the floodgates and let the teenager drive the car solo. Slightly risky? Yes. Nerve-wracking? Decidedly. Necessary for the competency development? Absolutely.
Change management is the art of convincing others. The comforting thing about AI is that it can—and should—be done piecemeal. You should not try to boil the ocean. You don’t have to get the entire board, C-suite, vice presidents, directors, and managers in a room and “Explain the Future,” as was necessary with a digital transformation.
John Kotter’s 8-Step Process for Leading Change is important when changing an aircraft carrier’s course:
- Establish a sense of urgency.
- Create the guiding coalition.
- Develop a vision and strategy.
- Communicate the change vision.
- Empower employees for broad-based action.
- Generate short-term wins.
- Consolidate gains and produce more change.
- Anchor new approaches in the culture.
In “The Rise of Cognitive Work Redesign,” Tom Davenport looks at implementing AI thematically. He points to suitable business process contexts, including:
- Where there is a knowledge bottleneck (e.g., medical diagnosis and treatment in rural areas).
- Where the required knowledge is too expensive to provide broadly (investment advice, and perhaps even college education, are examples).
- Where there is too much data or analysis for the human brain to master (programmatic buying of ads in digital marketing is a great example).
- Where there is a need for consistently high decision quality (the best examples are insurance underwriting and credit decisions in banking).
- Where regulatory pressures require a more informed process flow (again, investment advice is an example).
Working with an architect, a kitchen designer, a job foreman, and a city inspector is required to remodel your kitchen. Yet, all require very different tactics and modes of conversation.
Working with a data scientist is different from working with classically trained IT technicians, database designers, or programmers. Creating AI systems is not a case of defining requirements, generating designs, waiting for construction, grinding your teeth during testing, pulling out your hair during debugging, and holding your breath during launch. Creating AI systems is more of a dance. For that, it’s good to know your partner.
The most important message to deliver to your organization is that AI is not being brought in to replace people, but to work with them and help them work together better.
Your job is to get comfortable with statistics, keep on top of changes in the technology from about the 30,000-foot level, learn by trying things and see how they go, get used to shades of gray and uncertainty, and be the one leading the charge.
These may be baby steps for now, but eventually, your bot will thank you for it.