At a recent dinner, I was asked, “how could a marketer get started with machine learning and AI?”. Plenty of vendors offer specific solutions, but what if we just wanted to dip our toes in the water? What if we even just wanted to know where the water is?
Let’s look at how to plan and build our first machine learning/AI project with the AI/Machine Learning Lifecycle:
Identify A Problem Of Scale
Before we choose a technology or platform, choose a marketing problem of scale that we want to solve. What marketing challenge do we face that artificial intelligence is well-suited to solve? Artificial intelligence and machine learning solve problems of scale best.
For example, we cannot reasonably read through 10,000 articles in a day to understand tone or sentiment; machines can. We cannot physically watch more than 168 hours of video in a day, because we run out of time; machines can. What problem of scale do we face that, if we solved it, would fix a problem in our marketing or give us an insight we currently lack?
Find a Machine Learning Platform
Once we’ve identified our problem, we should choose a platform and technology to help us address the problem. Machine learning solutions tend to fall in one of several technology categories:
- Speech Synthesis
- Pattern Recognition
Let’s say we want to understand the sentiment and tone of our social media replies and comments. This falls in the field of language, so we need to look for language solutions.
Of the four major providers of end-user AI software, which have language solutions?
- Amazon AWS does not have a dedicated language solution currently
- Microsoft offers its Text Analytics API
- Google offers its Cloud Natural Language API
- IBM offers its Watson Alchemy Language API
Choose whichever provider your developers are most comfortable working with; all offer SDKs and example code to get started quickly.
Design Our Desired Outcome
Before we begin coding, identify what outcome we want:
- What will the output be?
- How will the software work?
- Who will be able to use it?
- What will we do with the end results?
Having clear answers to these questions ensures a more usable outcome.
Pilot A Use Case
Our developers will build a thin, lightweight pilot, a use-case which creates the outcomes we designed. This may take hours, days, weeks, or months depending on the problem we’re trying to solve.
We’re aiming for what’s known as the MVP – the minimum viable product. This is software that allows us to test our core assumptions, our desired outcome, without necessarily having all the bells and whistles of a finished product. Imagine building the engine and powertrain of a car without the polished interior or amenities.
Assess The Results
Once we’ve built our pilot, we test the results. Does our software solve the problem we’ve asked of it? Did we achieve our desired outcome? If we did, or if we at least see significant progress towards our outcome, we proceed. If it looks like we’re simply not going to achieve our goal, then we might shelve the project.
Iterate to Improve
Assuming we’ve made progress towards our goal, we work with our developers to iterate, to add features or fix bugs, until we have a stable, functional piece of software. As with development, this step may take days, weeks, or months to do, but it’s an essential part of the process to ensure we create a polished solution.
Deploy in Production
Once we’ve worked out all the bugs and our machine learning solution does exactly what we want it to do, we roll it out. This might be to our team, our company, our customers, or the world at large. Our machine learning solution needs support now, from instructing and coaching users to fixing unforeseen bugs.
Plan for the Next Version
After our machine learning solution is deployed, we start thinking about version 2:
- What else could we do?
- What would we have done differently?
- Has our problem changed in the time it took to develop it?
- Have our technology platforms added new features and functionalities with which we can improve our solution?
Get Started With Machine Learning and AI Now!
The steps to building and deploying machine learning and AI aren’t complex; you may recognize them as the software development lifecycle, or even more broadly as the scientific method.
The key to leveraging the power of machine learning and artificial intelligence is to start somewhere. Start anywhere. Try to solve any business problem of scale we have with AI and machine learning, but start. The longer we wait, the more opportunity we grant our competitors to develop a solution that leaves us in the dust.
Want to read more like this from Christopher Penn? Get updates here:
Get your copy of Marketing Blue Belt!