The next BriefingsDirect big data innovation case study interview investigates how Localytics uses data and associated analytics to help providers of mobile applications improve their applications -- and also allow them to better understand the uses for their apps and dynamic customer demands.
To learn more about how big data helps mobile application developers better their products and services, please join Andrew Rollins, Founder and Chief Software Architect at Localytics, based in Boston. The discussion is moderated by me, Dana Gardner, Principal Analyst at Interarbor Solutions.
Here are some excerpts:
Gardner: Tell us about your organization. You founded it to do what?
Rollins: We founded in 2008, two other guys and I. We set out initially to make mobile apps. If you remember back in 2008, this is when the iPhone App Store launched. So there was a lot of excitement around mobile apps at that time.
We initially started looking at different concepts for apps, but then, over a period of a couple months, discovered that there really weren't a whole lot of services out there for mobile apps. It was basically a very bare ecosystem, kind of like the Wild, Wild West. [Register for the upcoming HP Big Data Conference in Boston on Aug. 10-13.]
We ended up focusing on whether there was a services play in this industry and we settled on analytics, which we then called Localytics. The analogy we like to use is, at the time it was a little bit of a gold rush, and we want to sell the pickaxes. So thatâs what we did.
Gardner: That makes a great deal of sense, and it has certainly turned into a gold rush. For those folks who do the mining, creating applications, what is it that they need to know?
Analytics and marketing
Rollins: Thatâs a good question. Here's a little back story on what we do. We do analytics, but we also do marketing. We're a full-service solution, where you can measure how your application is performing out in the wild. You can see what your users are doing. You can do anything from funnel analysis to engagement analysis, things like that.
From there, we also transition into the marketing side of things, where you can manage your push notifications, your in/out messaging.
It means not only analyzing these things, but making meaningful interactions, reaching out to customers via push notifications, getting them back in the app when they are not using the app, identifying points of drop-off, and messaging them at the right time to get them back in.
An example would be an e-commerce app. You've abandoned the shopping cart. Letâs get you back in the application via some sort of messaging. Doing all of that, measuring the return on investment (ROI) on that, measuring your acquisition channels, measuring what your users are doing, and creating that feedback loop is what we advocate mobile app developers do.
Gardner: You're able to do data-driven marketing in a way that may not have been very accessible before, because everything thatâs done with the app is digital and measurable. There are logs, servers -- and so somewhere there's going to be a trail. Itâs not so much marketing as it is science. We've always thought of marketing as perhaps an art and less of a science. How do you see this changing the very nature of marketing?
Rollins: Everything ultimately that you are doing really does need to be data-driven. It's very hard to work off of just intuition alone. So that's the art and science. You come out with your initial hypothesis, and thatâs a little bit more on the craft or art side, where you're using your intuition to guide you on where to start.
From there, you have to use the data to iterate. I'm going to try this, this, and this, and then see which works out. That would be like a typical multivariate kind of testing.
Determine what works out of all these concepts that you're trying, and then you iterate on that. That's where measuring anything you do, any kind of interaction you have with your user, and then using that as feedback to then inform the next interaction is what you have to be doing.
Gardner: And this is also a bit revolutionary when it comes to software development. It wasn't that long ago that the waterfall approach to development might leave years between iterations. Now, we're thinking about constantly updating, iterating, getting a feedback loop, and condensing the latency of that feedback loop so that we really can react as close to real-time as possible.
What is it about mobile apps that's allowed for a whole different approach to this notion of connectedness and feedback loops to an app audience?
Mobile apps are different
Rollins: This brings up a good point. A lot of people ask why we have a mobile app analytics company. Why did we do that? Why is typical web analytics not good enough? It kind of speaks to something that you're talking about. Mobile apps are a little bit different than the regular web, in the sense that you do have a cycle that you can push apps out on.
You release to, letâs say, the iPhone App Store. It might take a couple of weeks before your app goes out there. So you have to be really careful about what you're publishing, because your turnaround time is not that of the web. [Register for the upcoming HP Big Data Conference in Boston on Aug. 10-13.]
However, there are certain interactions you can have, like on the messaging side, where you have an ability to instantly go back and forth. Mobile apps are a different kind of market. It requires a little different understanding than the traditional approach.
... We consume the data in a real-time pipeline. We're not doing background batch processing that you might see in something like Hadoop. We're doing a lot of real-time pipeline stuff, such that you can see results within a minute or two of it being uploaded from a device. That's largely where HP Vertica comes in, and why we ended up using Vertica, because of its real-time nature. Itâs about the scale.
Gardner: If I understand correctly, you have access to the data from all these devices, you are crunching that, and you're offering reports and services back to your customers. Do they look to you as also a platform provider or just a data-service provider? How do the actual hosting and support services for these marketing capabilities come about?
Rollins: We tend to cater more toward the high end. A lot of our customers are large app publishers that have an ongoing application, letâs say a shopping application or news application.
In that sense, when we bring people on board, oftentimes they tend to be larger companies that arenât necessarily technically savvy yet about mobile, because it's still new for some people. We do offer a lot of onboarding services to make sure they integrate their application correctly, measure it correctly, and are looking at the right metrics for their industry, as compared to other apps in that industry.
Then, we keep that relationship open as they go along and as they see data. We iterate on that with them. Because of the newness of the industry it does require education.
Gardner: And where is HP Vertica running for you? Do you run it on your own data center? Are you using cloud? Is there a hybrid? Do you have some other model?
Running in the cloud
Rollins: We run it in the cloud. We are running on Amazon Web Services (AWS). We've thought a lot about whether we should run it in a separate data center, so that we can dictate the hardware, but presently we are running it in AWS.
Gardner: Letâs talk about what you can do when you do this correctly. Because you have a capacity to handle scale, you've developed speed, and you understand the requirements in the market, what are your customers getting from the ability to do all this?
Rollins: It really depends on the customer. Something like an e-commerce app is going to look heavily at things like where users are dropping off and what's preventing them from making that purchase.
Another application, like news, which I mentioned, will look at something different, usually something more along the lines of engagement. How long are they reading an article for? That matters to them, so that they can give those numbers to advertisers.
So the answer to that largely depends on who you are and what your app is. Something like an e-commerce app is going to look heavily at things like where users are dropping off and what's preventing them from making that purchase.
Gardner: I suppose another benefit of developing these insights, as specific and germane as they might be to each client, is the ability to draw different types of data in. Clearly, there's the data from the App Store and from the app itself, but if we could join that data with some other external datasets, we might be able to determine something more about why they drop-off or why they are spending more, or time doing certain things.
So is there an opportunity, and do you have any examples of where you've been able to go after more datasets and then be able to scale to that?
Rollins: This is something that's come up a lot recently. In the past year, we have our own products that we're launching in this space, but the idea of integrating different data types is really big right now.
You have all these different silos -- mobile, web, and even your internal server infrastructure. If you're a retail company that has a mobile app, you might even have physical stores. So you're trying to get all this data in some collective view of your customer.
You want to know that Sally came to your store and purchased a particular kind of item. Then, you want to be able to know that in your mobile app. Maybe you have a loyalty card that you can tie across the media and then use that to engage with her meaningfully about stuff that might interest her in the mobile app as well.
"We noticed that you bought this a month ago. Maybe you need another one. Here is a coupon for it."
That's a big thing, and we're looking at a lot of different ways of doing that by bringing in other datasets that might not be from just a mobile app itself.
We're not even focused on mobile apps any more. We're really just an app analytics company, and that means the web and desktop. We ship in Windows, for example. We deal with a lot of Microsoft applications. Tying together all of that stuff is kind of the future. [Register for the upcoming HP Big Data Conference in Boston on Aug. 10-13.]
Gardner: For those organizations that are embarking on more of a data-driven business model, that are looking for analytics and platforms and requirements, is there anything that you could offer in hindsight having traveled this path and worked with HP Vertica. What should they keep in mind when they're looking to move into a capability, maybe it's on-prem, maybe it's cloud. What advice could you offer them?
Rollins: The journey that we went through was with various platforms. At the end of day, be aware of what the vendor of the big-data platform is pitching, versus the reality of it.
A lot of times, prototyping is very easy, but actually going to large scale is fairly difficult. At scale, you have to know what each technology is good at, and how you bring together multiple technologies to accomplish what you want.
That means a lot of prototyping, a lot of stress testing and benchmarking. You really donât know until you try it with a lot of these things. There are a lot of promises, but the reality might be different.
Gardner: Any thoughts about Verticaâs track record, given your length of experience?
Rollins: They're really good. I'm both impressed with the speed of it as compared to other things we have looked at, as well as the features that they release. Vertica 7 has a bunch of great stuff in it. Vertica 6, when it came out, had a bunch of great stuff in it. I'm pretty happy with it.