Segmentation Good, Over-Segmentation Bad
It's too expensive for marketers to ask everyone if they want to buy the next new thing. They eschew the dragnet for a line baited with the right kind of message that can get a consumer to bite. The trick is to fish in the right spot, which is where market segmentation comes in.
Bundling like-minded consumers into the same groups increases the likelihood of scoring the sale. Thanks to the digital revolution, however, marketers can now comb terabytes of data to develop those segments. Algorithms can find correlations that elude human attention, information that marketers can then use to develop market segments.
That all sounds good, but at what point do marketers develop too many segments? When does the data point in the wrong direction? Experts agree on the basics of segmentation. Where they differ is how to use the technology to keep marketers from over-segmenting.
Triangulate…and keep a human in the loop
“The client's problem is starting with a flawed data set.” says Kent Lewis, president and co-founder of Anvil Media, Portland, OR.
Single data sets are risky because they can provide incomplete or faulty data. One needs to “triangulate” the data set by cross-checking against other data sources, Lewis says.
This is where analysts and data scientists conflict. The analyst can look at the primary data, but data scientists can create multiple searches of data to get a more accurate picture, he explains. The data scientist will know the data and the science, but you still need someone who knows the business and the brand, he added.
Lewis gives several examples. One client used data to craft 10-12 personas, but once someone looked at the data, they saw that three to four of those personas overlapped by 80%. Consolidating those segments would have saved the client money.
Another client ran data through Google's Doubleclick to gain insights into what kind of digital ads to run in which spaces. But the prices programmed into the ad buy were too high, resulting in no placements and zero views of the ad. A human babysitting that project would have caught that mistake, Lewis points out.
It comes down to a human questioning the data. “Machines are great at finding answers.” Lewis said. “[They don't] know how to ask a question.”
Find the target, then aim
“The challenge with this is really how to create segments,” says Matthew Lee, president of San Diego, CA-based BusinessOnline.
Trying to create affinities based on Internet activity is less than reliable. “There is a lot of noise in that data,” Lee points out. You may have gathered a lot of Internet cookies, “but there is no transparency in how they get the data,” he said. “Proceed with caution” because the quality of the segments may vary.
BusinessOnline works in the B2B space, using data to sharpen enterprise-level sales with long, complicated cycles. The segments are pretty small: basically C-level executives. Clients need to find prospects quickly…and cheaply. “Our clients are not going to spend into a black hole”. Lee says. They want a fast return.
“We invested in a data platform” using its APIs to connect with Google, Facebook and LinkedIn. The platform, Data Weld, ports information into a data model to do the analysis. This is not easy, Lee notes, but Data Weld can determine if a segment is promising. The bottom line is that the segment that produces the sales pipeline will be spotted quickly—or not.
“The digital marketing eco-sysem is not simple.” Lee says. “Customers want to spend less and get more.” This puts the heat on marketing managers who have to be good stewards of the campaign, with little room for error and waste, he says.
“We see confusion and frustration with technology,” Lee adds. “We engage with clients looking for us to fix that.”
Check twice, then think twice
“Make sure you identify the segment in a reliable way,” says Dr. Art Markman, professor of psychology and marketing at the University of Texas, Austin.
The goal is to dodge false leads. Run the analysis on half the data set, then run the same analysis on the other half, advises Markman. Some finely-grained segments showing up in the first run-through may be flukes or outliers, and can be discounted if they do not show on the second run-through.
There are two concepts that come into play: spurious and silly, Markman observes. The more finely grained the segment, the more likely it will be spurious. “It's silly if you cut so fine a line that you may not see someone with those characteristics again.” he says. Cross-validation is important, because you want the analysis to be confirmatory, not exploratory, he added. “Otherwise you are misleading yourself.”
While automation can check data twice, only a human can lend insight. “This is why we have PhD programs in psychology,” Markman quips. Understanding human behavior—and the data—is what leads to good insight.
Walking this back to market segmentation, Markman stresses that a segment has to be “not spurious” and must also exhibit future behavior that is predictable. Illustrating this by example, Markman points out “early adopters”—those people who have to be the first on their block to buy a new tech gadget. They will buy an iPod, then a smart phone, then the ear buds, then a pair of Bluetooth speakers. The segment generates repeat sales.
Not knowing human behavior can leave a marketer in the dark. The market segment appears arbitrary—the marketer does not know what the “secret ingredient” is that drives that particular segment. Either the marketer is blamed for not doing a good job when the campaign fails, or he comes to distrust big data.
“Do not treat [the technology] as a black box, throwing [data] into the meat grinder and eating the sausage that comes out.” Markman says. Someone on the team has to understand the causal forces that shape a segment, or “you will never know when you are going off the map.” he adds.
So how much segmentation is too much?
“That's a loaded question,” says Chad Pollitt, author, professor and partner/VP for audience at the Native Advertising Institute.
“Generally speaking, as a marketer, you have to keep persona segments on the low end.” That should be anywhere from one to five; three being ideal, Pollitt says. “With the personas, you craft specific content strategically targeted to those buyer types.”
So far, so good. It sounds like marketing hasn't changed. But then Pollitt dropped the bomb.
“Today, with artificial intelligence, machine learning algorithms, natural language processing and a host of other technologies, we're able to delve into big data.” Pollitt explains. Now it is possible to craft many “micro-segments," relying on AI to do the heavy lifting.
Over time, AI will filter those segments, discarding down to only the most effective ones. “Maybe machine learning determines that only 20 segments are valuable,” Pollitt says. But AI can also overlook LTV—Lifetime Value—of some segments, slices of the market that, while smaller, may generate more value through repeated sales over time.
People will craft the ads that reach those market segments in a language that resonates with those personas, Pollitt notes. IBM's Watson is perhaps the most well-known AI platform, but there is also Salesforce's Einstein. Other platforms beckon.
Pollitt offers Experian Hitwise as an example. Tapping into just a portion of all the clicks at the ISP level, Hitwise uses the data to craft 200 “eerily accurate” personas, right down to showing photos of homes, neighborhoods and TV preferences of “persona families,” just to illustrate the data.
“Introductory marketing jobs are being taken over by AI,” Pollitt says. Algorithms can handle work flows of e-mails and social, predicting which consumers will open and respond to e-mail pitches, and even cranking out personalized messages after comparing them against segment data.
“As long as the outcome and the KPIs reach the expectations of the boss, they are going to keep using it (the technology).” Pollitt says.
And the technology is only going to get better…