Artificial intelligence, or machine learning, is on the rise. From Siri to self-driving cars, new frameworks are becoming accessible to the mainstream, and products are growing increasingly "smarter." As massive quantitates of data are being plugged into predictive systems, this leads to a question of quality: How can we ensure the programs we put in place will understand the data we throw at them? Geoff Roach with Chief Outsiders argues that no matter how smart our machines become, humans will always need to be one step ahead.
"Even 5-10 years from now as these tools get more powerful, it doesn't release us from the need to be critical. People get so hung up on technology, because of the relative newness of marketing being so data driven, you have to be able to ask, 'Does this answer make sense?'"
Machine learning can be especially useful to marketers, from pricing to positioning to streamlining data readiness. The challenge is figuring out how to use machine learning to maximize the opportunity for analysis. Marketing departments depend on data that comes in drastically different formats, and we need to be smarter about how we look at that data, and what we do to analyze it.
Geoff gives us a classic example of how machine learning can help interpret data. "When I'm presenting, I will ask a room full of people to write down the address '10 North 1st Street,’ then I ask them to read back, letter by letter, exactly what they wrote. Some people use abbreviations, some write out each word. There are so many variations, and the issue is, if you have an analytics program trying to figure out where people live, it's not going to recognize that all those variations refer to the same place. We need to improve the data so we can make better decisions."
High-level marketers of the past spent inordinate amounts of time preparing data to be entered into smart programs, thereby negating the time they were supposed to be saving. Geoff explains, "They'd look at the data and the outliers, and find all this stuff that wasn't working, so they'd have to go back to their excel spreadsheet and run it again and again. They weren't really running analytics they much as they were trying to get the data to work."
As new programs come along and start looking at the data in different ways, the machine learning gets smarter, taking the burden of data prep out of people's hands - and putting it into the machine's. At the same time, programs like Tableau, Alteryx and Paxata are trying to take the data prep out of the IT shop and put it into the marketer's hands.
ML tools to improve analytics
The intersection between machine learning and analytics is the ability to make decisions based on the interpreted data. Google Analytics, for example, is often difficult to use if you don't know exactly what questions to ask, and what data parameters to use to get what you want out of it. Adding a few machine learning tools to the program could make it easier to use, and jobs like SEO specialists - roles that are paid to interpret the data - could be simplified.
"There's a difference between data and information," says Geoff. "If you take the data and put it in context, suddenly it becomes useable. In the case of Google analytics, I was working with a company whose homepage had a bounce rate of 75%. That didn't seem good, but what's the context? Is it bad? Is it good? We don't know. We had a piece of data, and with a little more context from smart machines, we can know if it is an acceptable bounce rate."
Machine learning is changing the way companies handle pricing and inventory decisions, in small businesses and major retailers alike. For the first time, machine learning is leveling the playing field, making ideas and techniques that were only available to major retailers accessible to everyone.
"The heroes of that are Walmart, they were doing big data before big data came along," says Geoff. "They had 80 weeks of sales data on every SKU in every store. They were trying to develop techniques on geographical pricing, but also answering questions like, 'It's getting late in the season, should we try to sell all our bathing suits, or send them all to Florida?' Those are the kinds of questions ML should be answering."
Still, Geoff advises caution.
"It's a means for people to make better judgment. Testing ideas are getting easier and faster, so we can do more of it, but it still needs to make sense. There needs to be a place for a human to double check and see what's going on. The tools need to tell you how they arrived at their decisions, so you know if you can depend on it."
A fool with a tool is still a fool
The concept of machine learning isn't new, but now that it's gaining traction, we need to remember that a machine is still just a machine - it doesn't have the ability to empathize or think conceptually. Machines can crunch massive amounts of data at lightning speed, but only humans have the ability to read important customer cues and subtleties.
"Artificial intelligence has been around for decades," Geoff adds, "but what has made it accelerate is the ability to throw lots of memory into a box to make some of this stuff work. The technology has gotten cheaper and faster. The downside is, people rely too much on the technology without thinking of it critically. Don't forget to look inside the box and see what choices it's making, what it's doing. People can't forget that."
The bottom line
Machine learning is already automating many routine tasks within marketing departments. As it becomes a more integral part of our working environment, CMOs will be called upon to develop a basic understanding of how to interact with these programs. Geoff offers the following suggestions for CMOs stepping onto the new ML landscape:
Working with machine learning systems is a way for us to make better decisions - not delegate decisions to a piece of technology. Smart technology will improve at a staggering rate in the next few years, and it's more important than ever that we use discretion, common sense and critical reasoning - qualities only a human can have.