February 6, 2013
Q&A with Ira Helf, chief analytics officer, JWT North America
One of our 10 Trends for 2013 is Predictive Personalization: the idea that brands will be able to predict customer behavior, needs or wants—and tailor offers and communications very precisely—as data analysis becomes more cost efficient, the science gets more sophisticated and consumers generate more measurable data. Ira Helf, who oversees JWT North America’s marketing analytics practice, has particular expertise in developing and deploying quantitative analytical models, especially in the areas of segmentation systems, response models and customer profitability. He talked to us about how Big Data is changing the analytical process and the potential for these new data sets to help brands build more personalized experiences.
“Big Data” is a hot topic these days. How would you describe it in layman’s terms?
Big Data is a fairly new term. We’ve always had a lot of data, but Big Data seems to be the term that’s used now when including a lot of social media data. The big difference for me, as somebody trained in statistics and data analysis, is that this new set of data is not structured. We’re used to things that are sets of numbers. It was transaction data and so on. Now we have user-generated content that people are typing in, so it doesn’t fit into the traditional mold of analytics.
We haven’t really settled the issue as an industry, which is to answer “How do you classify that user-generated content?” The misclassification is mindboggling. I was pointed to a site just before the election that was trying to predict election results based on positive statements about Obama versus positive statements about Romney in the media. I looked at the statements and the error rate was probably about 50 percent.
There is a big problem in what is called Big Data in terms of how to use that in the context of a traditional statistical analysis. The second part of it, which will flow out of a solution to the first, is what does it really mean, anyway? What does it mean if somebody says, “I really like Chips Ahoy cookies [on Facebook]”? Does it have any effect either on that individual’s behavior or on the people who read that? We don’t know yet.
If we go back to the more traditional kinds of numerical data, the point is still valid that every year we’re getting more and more of those data. So in an area where we can do more analysis, we are definitely working with a larger set of data that comes in more quickly. If you go back decades, we’re accustomed to surveys that used to come in via the mail. Now not only do we have surveys that come in via the Web, we also have all kinds of other click activity that comes in almost immediately.
But what they call Big Data gets a little confusing, because when they look at all the zettabytes of data that are coming in, a lot of that is coming in as social media.
And mobile activity as well?
You’re absolutely right. The mobile component of that is growing.
What are some of the factors driving Big Data—are computers getting faster and more powerful?
Inevitably, the computing power increases. What hasn’t changed too much is the analytical tools that we use against that—which is not to say there isn’t new software coming out all the time, but basically the things they do are still based on statistical procedures that were designed in the 1930s and earlier. It is shocking to think, but there’s a certain part of this that really hasn’t changed. Which means the basic computational devices that we use are more or less the same. We have more data to work with. It’s newer data, and there are more people that know how to use it, though this last bit is growing slowly.
There is at least the thought of making some of these procedures more accessible. So whereas in the world that used to be dominated by the traditional programs such as SAS and SPSS, there’s now many different software packages, some of which claim to be more user friendly, so that somebody who has never really been trained in it can run data more easily. There’s a little bit more accessibility because of those software platforms. There are more programs at universities to train people, and even the people who aren’t going out to be a statistician.
Some are predicting that in another 5 or 10 years, data analysis will be widespread throughout enterprise, and it will be a key skill to have.
That’d be great. I hope so. It’s hard for me to envision that because I still get too many people whose eyes glaze over when I put a spreadsheet in front of them.
How does it change things when you have more real-time information—that is, measuring the past versus measuring things as they’re happening?
It does require some different infrastructure, but the biggest thing for us is that it makes what we do more accurate. I would in many cases take quicker over more data. Say I was going to predict your need for a new automobile: If I have data that’s three months old, that’s not as good as what your search behavior was yesterday.
The fact that things are coming in more real-time makes the predictive models more accurate, and that’s a great advance. When we went from postal mail surveys to Internet-based surveys, we gained a lot of accuracy because we could get those answers sooner. You can take it even further now and say, “I know what you were doing on the Web five minutes ago.” I as a marketer can interject a message based on that, and that’s very powerful.
And not just what you were doing on the Web. Taking it a step further, “I know that you’re standing outside my store, and you usually get coffee this time of day.”
Yes, locational data. It’s the same idea in supermarkets now that we know what aisle you’re in. We can send a text to your mobile saying, “We have a sale on Domino sugar or whatever, and we know you’re in the baking aisle.”
That will feel a bit scary or invasive to many consumers though?
People are very concerned about that. I have a bias because this is what I do for a living, but from a marketer’s perspective, we’re not trying to invade your privacy. We’re actually trying to give you something that’s more relevant to what you’re doing in your life. But since people aren’t so accustomed to it, you can really scare people and freak them out. You can say, “I know you were interested in a certain kind of jeans yesterday while shopping on the Gap’s site.” Wow—it’s jeans; it’s not really a big deal. But it’s spooky.
It can be spooky when your online activities are suddenly being showcased.
It’s hard for people to perceive that there’s still anonymity. We as marketers really don’t know who you are; we don’t know where you live. All we know is you were shopping for jeans on the Gap. It’s hardly an insight into your life. But it just feels like somebody’s watching you.
Do you think that will be a major hurdle to overcome before this type of marketing becomes more widespread?
I think it will. Certain cultures are more concerned about it. Europe is notorious for this issue. They don’t feel like you should be able to observe those things. So whether we ever change that culture or just reduce the activity to meet the cultural norm, we’ll see. But I do think it will be important.
Have you seen many creative uses of Big Data at this point?
It’s a funny thing about the way it’s used. Even as we get more accurate in terms of the timeliness of the data and the amount of data, when it’s used well, it means that you did something smart about it. And in a certain way, that is not any different than it might’ve been 20 or 30 years ago, even though we might not have known much about prospect information but rather customer information.
Imagine at a bank, they had your transaction data. They knew how much you had in what kind of accounts. Armed with this information, is there some way that they could change your experience versus my experience because of what they knew? They might offer you a home equity line of credit or something like that; there are certain products they could serve up. But if you go a little deeper in terms of the experience of the brand, could they really differentiate that? In a lot of ways, they couldn’t. We’d both walk into the bank branch, if we were to go into the branch, and we would have the same teller experience, even though we might be two very different kinds of customers.
So to me, until our clients, and the agency itself, can manage the tailoring of the experience based on what the predictive analytics finds, the predictive analytics become less valuable. The classic example has been for a long time Amazon. Amazon does try to tailor the experience. They’ve taken customer data and been able to offer up different product suggestions. I worry about how we can do that in a broader sense when you move away from customers and you talk about prospects.
Even on Amazon, my first time on the Web, say, they don’t know anything. They’re going to start to get information, because maybe I’m going to shop around a bit and they might see what I click on. But in what way can they tailor my experience after that? And certainly if you move outside a digital brand, an e-commerce brand—let’s take a bank again. I’m shopping for a checking account. They don’t know what my assets are or what types of accounts I need. Are they really able to make that prospecting experience different?
British Airways has a new data-based program, “Know Me,” focused around offering highly tailored customer service. They’re saying that when they get it right, the customers love it, but if they go a little too far, it starts to creep people out.
The airlines have been very good in general. The benefit of the frequent flyer programs have been that they’ve gotten information about customers at an individual level. That’s been both explicit and implicit. The explicit part meaning that if you’re a really good customer, you go in a different line. So there’s a different experience of the brand and hopefully something that’s more pleasant and works for retention. They can tell even without a picture pulled from Google whether you’re super good or just regular good. And so they did differentiate the brand experience more implicitly, too, in that way. The airlines are on a good track in that way, using information in a way that isn’t too creepy.
Certainly, I expect the airline to know whether I’m a good customer. In fact, that’s an interesting twist to the whole thing: If you avoid the creepy information but stick with the things they should know, one of the changes on the customer side is that consumers now expect you to know a certain amount. If you’ve used OpenTable 100 times to book reservations at a particular restaurant, you’re surprised if they don’t give you a good table, if they don’t know you’re a good customer. There’s a little bit of a give and take going on in this.
I liken it to when my grandmother shopped in Brooklyn in the days before supermarkets. It was the days of small stores. When you would go into a store, the proprietor would actually go and get your groceries. He knew everything you bought. He knew what brands you liked, how much Swiss cheese you wanted and all that stuff. Then things changed. In a more general sense, everything became more anonymous, and suddenly proprietors didn’t know. Now we’re back to a point where there is a way for a company like Fresh Direct to know what products you like, and it’s oddly like the mom and pop store that used to be a long time ago.
So what are some of the roadblocks standing in the way of companies trying to take advantage of this?
We’ve touched on many of them. One is understanding how you change not only the product but really the brand experience according to what you know. There’s still a fairly substantial roadblock in terms of people knowing what to do with the data.
There’s a roadblock that’s people-oriented in terms of getting people on your team that not only can tabulate the numbers but also can help you with the insights and the actions, and that’s a problem on my side of the industry—people that are number-crunching and not willing to push it further. We need to come to some agreement between the people that are non-analytical and the people that are analytical. How do we make the analysis more impactful?
Where do you see this going in the next three to five years?
Let me say where I hope it will go. I really do hope it opens marketing up to not just the idea of predictive analytics but the bigger idea of testing and learning. There are two reasons behind that. One is the data and the second is just the digital revolution that’s already here but still has way more to go. It’s only going to get bigger and better.
There is still a bit of a mentality from people who have come from the era of television that it’s all about creating this big TV ad, and you live with it for six months. Then you find out if it works or didn’t work. And I hope we will change philosophically. We still have the TV ad, but the campaign is bigger than the TV ad—if I can make the messaging on the website better, if I can make the email communication better, if I can instill the right kinds of brand journalism into social media, if I can do all those things on an ongoing basis and do it methodically. I may keep 80 percent [of the budget] for the mainstream but keep that 20 percent out and always try something, because we have the ability to do that now.
And maybe it’s a little bit more budget, which is an issue these days. Hopefully by doing it on an ongoing basis, we can get to a point that we make back the money we invested in learning. That’s what I would love to see. I think that’s doable in five years. I don’t think it’s a big shift from where we are, but it’s amazing to me that there’s so much concern about getting the first version out there.
It’s not an easy thing to do. There are a lot of moving parts. But to include in that [budget] the other 20 percent, which is, “How am I going to build into the system ways that I’m always learning something more?” That’s what I’d like to see.
It’s just old-fashioned to do a big campaign, let it run for six months to a year, and then say, “OK, we need new creative.” First of all, you don’t even know really if you have wear-out, but let’s say you do. What elements do you keep? What do you change? That capability’s really there nowadays.
Which industries do you think this will impact the most?
This is sometimes a hope, but I feel that the packaged goods industry is one of the ones that have the longest distance to go. A lot of the other industries we work with have transactional data, and we’ve been able to mine it for a long time. But because the nature of that industry is such that you have the manufacturer, then you have the distributor, you have the supermarket or the store itself and then the consumer, it’s been harder for all that to be linked together.
We do have some manufacturers that are already setting up small e-commerce sites for niche products you can buy directly from them, but the vast majority of purchases are happening in this different kind of set of silos. If we can come up with ways that link them together—an example being somebody like a Fresh Direct, where they immediately have all the data digitally. That’s a place where I hope we’ll see a lot of progress over the next few years because they haven’t been as much the beneficiaries as other industries that have been more direct and digitally oriented.
I’ve read that companies aren’t really sharing information right now but that for everyone to benefit, we need to tap into the wisdom of the crowd.
That’s true. They’re very protective of the data because it’s a very valuable asset. And they use it to manage their businesses, so they don’t really want to open it up. But I think you’re absolutely right—that as a category, they will all benefit by sharing. But it’s back to the obstacles.