Yeah, sure — Big Data. We get it, right?
We all know the digital age is producing huge amounts of data about consumers and their behaviour. And, sure, we know that anybody who’s in the marketing and advertising business — like local media companies — needs to get good at it. Right?
Not that we’ve quite learned how to do it yet. But surely we know — don’t we? — that we simply must master it to benefit both ourselves and our customers? And we’re working on it, right?
Well, I am. I hope you are, too.
Why? Because somebody is going to bring Big Data to Main Street. If it’s not us, Big Data will be the next big wave of disruption in our advertising and marketing business. It’s guaranteed to whittle down our local media ad revenues still further.
I’ve blogged about the huge opportunity and threat of Big Data for local media companies four times in the last 13 months. If you’re a regular reader, you may be thinking, “What, again!?”
If you’re not a regular reader, I strongly recommend you catch up on Big Data and its local media possibilities.
But I can’t stop there. I keep digging deeper to learn more about what Big Data can do and how we can master its potential for ourselves and our customers. And I keep learning.
For the last couple of months, I’ve been digging into predictive analytics (PA), a narrower niche in the vast expanse of Big Data. It’s the sharp cutting edge that is making Big Data even more powerful.
It started when a colleague recommended a book by Eric Siegel. He said it’s the easiest path to understanding what PA is and why it’s important. The title is Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die.
Siegel is a data scientist who seems to be making a career of helping the rest of us understand what PA is. I’ve been plowing through his book and taking multiple side trips on the Web to learn more as I go.
Believe it or not, Siegel actually makes PA fun. In fact, before you read any further, you should stop right now and check out his rap video.
What is predictive analytics?
PA uses masses of data about customer attributes, behaviours, and life changes to predict which individuals are most likely to take a desired action or manifest a certain outcome.
For example: Who is most likely to buy a product or service? To list a home for sale? To buy a print or digital subscription? To develop a certain medical condition? To default on a loan? To commit a terrorist act?
In advertising and marketing, PA is rapidly emerging as the source of the next competitive edge. When likely buyers can be accurately identified, they can be marketed through highly targeted channels, cutting marketing costs and reducing wasted sales time.
And PA can also forecast which next step in marketing is likely to work best with any given potential buyer — e-mail, phone call, direct mail, or even banner ad A versus B.
This is marketing with a laser beam instead of a mass-media shotgun.
How do they do it?
Now that so much data exists about everyone, the best way to know who’s most likely to make a certain kind of purchase is to examine the deep data about those who have already made that purchase.
A data scientist builds a software programme that can sort through all the data on the past buyers — characteristics (e.g., age, income, education, family status, etc.), behaviours (Web searches, credit card purchases, clicks on ads, etc.), and life changes (marriages, divorces, child-bearing, college graduation, etc.).
The software can examine far more variables and combinations, far faster, than any human ever could.
The software boils the data down to those data points, or combinations of data points, that appear to be the best predictors of that particular type of purchase.
Then the scientist turns the model on the far, far larger number of individuals who have not made that purchase recently. The software cranks through these prospects, scoring each on his or her likelihood to make the desired purchase.
And the model, once built, keeps working — more data, more computations, greater accuracy. That’s what they call “machine learning.”
Those who score highest are the best prospects.
For example, I talked recently with the CEO of a small company that’s doing PA for colleges and universities. His clients are happy to spend as much as US$100,000 a year for PA to reveal the few thousand high-school grads most likely to apply to their colleges or universities. They can focus their marketing dollars and efforts on those kids, instead of trying to market to the millions of kids who graduate every year.
How is PA different?
PA takes an important step beyond the most common uses of Big Data in marketing. Compare the PA approach, for example, to the targeting that’s becoming so common in digital programmatic ad buying.
In a typical programmatic ad programme, the buyer identifies certain Big Data attributes that seem likely to correspond with the proclivity to buy — age, income, digital searches, location, and so on. She then sets up a bid on the ad exchanges so the ads fire when individuals matching those characteristics click into Web pages.
This approach may draw on pretty much the same pool of consumer data as would be used in PA. But it’s built only on broad guesses about the attributes of likely buyers. Predictive analytics is based on the far more powerful data about who has already made the purchase.
Where to apply it
PA can be used in a vast range of fields including medicine, national security, finance, insurance, and more — anywhere the ability to predict certain behaviours has concrete and valuable benefits.
For media and marketing, most applications of PA are about identifying the best leads, or about determining which marketing actions will work the best to convert a lead.
For local media, I’m thinking the most promising verticals are real estate and auto sales. Who is most likely to list a home for sale in the next 12 months? Who is most likely to buy a car in the next 30 days?
These are high-ticket purchases where a strong lead is very valuable — enough to justify the cost of PA.
PA could also prove worthwhile in other local marketing verticals, too. But at Morris Publishing Group, we’re working to find PA solutions in real estate and automotive first.
How to get started
First, hire a data scientist or two, plus some support staff.
Then buy the software and analytics tools you need to build PA models, and then license all the data you’ll need. You should be able to do that for about half a million bucks or so.
Get real, Gray. Who can afford to do that?
Well, some large media companies could, and maybe some are. But at Morris, we’re planning to walk before we try to run.
We’re talking to some start-ups that are already doing PA for the real estate and auto industries. Our pitch to them is that our sales teams can give them feet on the street in our 11 media markets.
We already have great relationships with thousands of prospects — real estate agents and brokers and car dealers. These customers would welcome next-generation marketing solutions from us. Most small PA start-ups can use this kind of help on the sales side.
I won’t name any of these possible partners right now, because we don’t have the business relationships worked out yet. But when we do, I’ll be happy to share.
Predictive analytics applies, too, on the audience side. We’re talking with Blueconic and Cxense about using their tools to increase audience engagement on our Web sites.
Predictive analytics is the next wave, and you’ll be hearing more and more about it — in our business and throughout the business world.
As Eric Siegel’s video says, “I can predict your avenue, just give me all your information … Provide me the data to improve, and I’ll apply the computation.”
And my favourite line: “I love it when you call me Big Data.”