Programmatic IO session: Best practices for implementing a CDP
Watch the recording of mParticle's workshop at Programmatic IO San Francisco covering key CDP use cases and the best practices you should follow for implementing a CDP into your tech stack.
[00:03] Hi everyone, I’m David Spitz, CMO at mParticle and today we’ll be talking about best practices for implementing a CDP. You probably have all read about CDP I’d assume over the last 12 to 18 months. There’s been a lot written about them. Ad Exchanger came out with their first installment and their guide last week, which I recommend to all of you. There’s a lot of literature from Gartner and Forrester on the topic we have our own guide, which you are welcome to download using this link. I’m all these things sort of cover what is a CDP, why do they matter? But to be honest, nothing so far has talked about how to implement a CDP and Ah, who should implement a CDP and we’re going to try to address that today for the first time. So let’s get started. I’ll give you a little primer on what is a CDP, uh, just to, just to set the stage.
[01:03] It’s a customer database managed by the business, not by the it department, which associates groups have known users together into segments or other sorts of meaningful groups in particular. These databases are great at ingesting data from lots of external systems and since syndicating that data to many more external systems for insight and a execution and then closed looping that data back into the database so you can think of is sort of as a marketing middleware, which is a concept that makes all the sense of the world. I’m not entirely new it departments and marketing teams at large companies have been trying to hack things together and build customer data platforms, lowercase customer data platform, uh, for many decades now. Uh, so, so why all the excitement? This is an image from AdExchanger article talking about the CDP sort of rising like the phoenix and destroying all other sorts of marketing technologies are or something like that.
[02:07] So what, what’s so exciting about CDP is I think, I think what’s different about them compared to everything else you know, at this conference or that you encounter in conferences like this is that [inaudible] are really more broadly customer data management appeals to a very broad set of stakeholders. So rather than a particular function within, say the marketing organization or the media department, we have people whose imaginations are really inspired by the idea of working together, cross functionally to create just breakthrough customer experiences at whatever cost. So that’s, that’s quite different from the last decade or so of marketing technologies that have been very splintered to one functional group at a time. So we see CDPs engaging with teams across digital marketing, customer experience, customer analytics, customer service, IT and engineering and much, much more to do what exactly is needed to bridge the gaps.
[03:12] One prime example would be just bridging the gap between web analytics and mobile APP analytics, making those things work together. So they’re not two separate islands or silos. You heard of the last presenter talk about tying a media attribution to customer lifetime value, so you really know the true value of the install that say you’re, you’re acquiring a connecting paid media to see around these are all at different examples of CDP use-cases. There’s many more which will cover a during the presentation. So where do you find a CDP? I said historically you would look around and try to find some group internally that was building this for you, or maybe you would work with a consultancy like accenture or deloitte to try to build one custom for you. I’m really over the last several years, a productized customer data platforms have started to come to market and they generally fall into one of these three different buckets.
[04:13] A, let’s start with the one in the middle that probably people here are most familiar with. A lot of the companies that historically went to market as tag managers focused on web data collection and a routing a web data to different, different sources of have come forward and said, no, actually we’re a CDP. I think that, with a few exceptions, they generally don’t meet the definition of storing persistent data in a data warehouse, enabling you to connect across multiple external systems. They’re very web and advertising focused, so I’d say generally by definition they’re, they’re not a CTP, but then that leaves these other two, two sets on the one side, there’s the cloud apps that are focused on maybe [inaudible] or, uh, some, some other type of analytics solution and a lot of them have come forward and say like, we’re collecting all this data, we’re persisting the data, we’re matching ideas together.
[05:07] We’re, we’re going to be a CDP. And then on the, on the other side of this chart, there’s the pipes and my company and particle sometimes is derogatorily referred to as one of the data pipes. And really we’re focused entirely on getting the Apis to work as best that they can, uh, collecting data natively from, you know, everywhere that touches the customer journey and providing just enterprise grade data management. We don’t do any execution. We don’t, we don’t send emails, we don’t serve ads, we don’t give you analytics reports. We’re trying to be the best, you know, data plumbing company, if you will, uh, in the world. And so, which one of these, our CDP or which one’s right for your business? I think, I think that’s still an open question, but, well, I will say is that it comes down more to philosophy than to, to product Ed.
[05:54] So you have to ask yourself who wakes up every day thinking about how do I get more data out of this, this place and into the hands of the people who need to use it no matter what business model, uh, those companies or whether that person’s part of my company or someone else’s company, how do I get data out of the system? And our customer success, people are actually incented to get data out into the hands of the people who need it. So, so our, our product people are, our engineers are thinking about that every day. They’re also thinking about how do I get more data in from, you know, Roku devices and x-box and whatever, you know, at Alexa, whatever the next frontier of data collection is, you know, how do we get data into from those important sources and that’s what they worry about every day and how do we make sure that we’re protecting customers a privacy rights as well in, in the process.
[06:40] So it really comes down to more of a matter of focus. And I’ll borrow from a Peter drucker and say, culture eats strategy for breakfast, a justice, which is to say it’s less about the product and, and really maybe more about the philosophy or the culture of the company in terms of what is, what is the true essence of the CDP. So now let’s get into how do you get started with it, with the CDP, uh, we have a five step process and it starts with understanding your, your current state. And a lot of people come to us with a map that looks something, something like this. So they say, Oh, here’s my current architecture. I have this integrated with that. And I have this integrated with that. And these are all different sorts of points. Integrations. Some people look at this chart like, oh, that’s actually pretty good things.
[07:24] Or things are integrated together. The reality of the situation is usually much, much worse. Mary Meeker,in her Internet trends report, said that the average company has 91 cloud based marketing application. So you need a piece of paper much bigger than this slide to display the true reality of what the average marketing stack looks like. Much less if you want things working together in a real time. The, this, this type of integration method is just not, it’s just not going to cut it. So this is a good place to start when you’re starting to assess your current state is just to understand what sort of, what sort of tools and analytics needs do you currently have around your business. But it’s not a good place to finish. We recommend that people also think deeply about the customer journey. And I think not just about, you know, where are you expect people to go.
[08:13] But also with the hacks that they are there, they are doing to get there and how data can better to support and, uh, enable that, enable that journey. So once you thought about the current state, now it’s time to think about where you want to go and what’s the, what’s the end state, and aligning your vision is one, just aligning all those stakeholders that we talked about in the organization, but it’s also a aligning in terms of the what you need to do and why, why you’re going to do that. And thinking deeply about the connection between the customer data strategy and your businesses strategy. So I’ll give you three examples of that from, from customers that we’ve been working with on, on exactly this topic. The first example is a overstock.com. And in November, the CEO of overstock on the company’s earnings call was talking about how the economics of their marketing function was just broken and they need to totally rethink how they went about customer acquisition, why he cited one particular competitor, not Amazon, but another well-funded competitor.
[09:21] That was just spending an irrational amount on search engine marketing. So keywords that used to talk used to cost seven cents are now $2 and it just, it made no sense. This company, their competitor was losing money, but it was driving up, you know, all the, all the keywords at the same time. Their organic traffic was a declining, like many, many companies’ were, around that time. So the customer acquisition model that had served them well for the last decade plus was, was no longer working, so they needed to think, rethink the model and a lot of that starts with personalization and making sure that the people who arrive at the site have the best possible experience and are the most likely to buy and buy, you know, fill their basket size to the fullest and making sure that the site site experience what was excellent, but it didn’t stop there, uh, using a CDP.
[10:12] They, a wrapped together not only the site site personalization experience, but also the media experience, uh, through integrating with their dsp as well as the email and a push messaging experience. So all these things were working in tandem with data being sent in and out of a central repository with a single customer view for the purpose of driving better economics on their marketing spend. So that’s one very common CDP use case makes all this, makes all the sense in the world, but it’s very different from use case number two, which is a postmates. So how many of you are familiar with a postmates or used postmates? OK. So, uh, they’re more than just a food delivery company. They like to say like, you’re not just getting food delivered to your door, they’re a logistics company today actually, they announced a partnership with Walmart that there’ll be doing delivery for walmart groceries, which will reach 40 percent of the United States.
[11:11] And that’s part of their vision to be a sort of one part ups, which is this all purpose logistics company. But then on the other side, one part, the Disney company, they want to provide what they call a magical experience. So logistics meets magic, if you will. And how does the CDP help help them with that while they’ve centralized all their data, meaning, you know, all their touchpoints with the customer, from the product analytics. So what’s happening, you know, in the app and on the site to the marketing touch points and even through ZenDesk, which was their customer service platform,and they looked at all this stuff together and they thought, well, if we could spend an incremental dollar, I’m on the customer experience, where should it go? And doing these analyses, usually you find it’s not marketing that needs that incremental dollar.
[11:59] They invested more in product redesign, parts of the UI, and invested more in the customer service experience. They also, by looking more more deeply, they ran a couple of experiments and found the more that they’ve invested in marketing sometimes actually that created more customer service calls, more, more problems because there weren’t enough delivery people to deliver on time, et cetera. So it was actually damaging the customer experience in some cases. In other cases, they found that part of their customer service recovery process when someone had a bad experience. And they called up the help desk, they could then serve them an ad or send them an email, which actually made them more happy to be a customer postmates than ever before. So it’s this orchestration of different functions within the company, uh, enabled through the integration of data that’s really driving postmates, use of the CDP, starting with insights leading to better execution, all for the purpose of customer experience or long-term customer growth and satisfaction.
[13:01] So that’s example number two. And the third one, you know, equally valid, equally common is SeatGeek. SeatGeek, if you’re not familiar with them, is a secondary ticketing business and they came into business in a very crowded market, but they realized that there were still a lot of inefficiencies. So through data and through speed and agility, they were going to make ticket purchasing just much more efficient and make it cheaper for the end customers to buy secondary tickets. So it’s essentially, it’s a margin game. And so when they looked at their customer data strategy, they thought, you know, we need to keep our margins in line, we need to make sure that we’re not over investing on it costs and engineering and customer development where we don’t have to. So we’ve been working with them for the longest of all, all three of these examples, a first with just a deploying new tools into the app.
[13:58] So if you’re familiar with the app world, it’s often like open heart surgery to try to install a new analytics or attribution or a messaging, a SDK into your app. It’s very, very difficult. Using a CDP, you’re able to just switch on these integrations server-side without having to get all these engineers and, and developers a coding on the app, uh, and, and basically slowing down your speed to market. Uh, over time they started testing out new providers, uh, just switching on, switching off different services that they wanted to test. And, uh, most recently, uh, there was actually an article in ad exchanger about this a couple months ago. They started using the CDP to build suppression lists, so they don’t want to ever run an acquisition campaign against people who’ve already installed the APP. That’s just kind of obvious, but it’s also a, you know, a process complexity to try to continue to really sink all the identifiers of people who install the APP with snapchat, with facebook, with all the, with all the channels that they’re running, their ua campaigns using the CDP, that list is st automatically. And they found that before doing this they were wasting 40 percent of their impressions as just simply by, you know, sinking these lists, suppressing out those Lou users, they were able to save millions of dollars in media costs. So three examples, three very different examples of a CDP in action, now I’m going to turn it over to Paul, who will tell you about how you implement those visions.
[15:32] Thanks David.
[15:34] So even before you talk about architecting a CDP, it’s valuable to step back and say, well where does the CDP fit within an overall marketing architecture. And I think Forrester has a pretty good framework. They call it the modern customer experience architecture and really just four tiers to it. The first tier is the client here, so that’s actually the channels where your customers are actually going to interact with the brand. That’s going to be your website. Any apps that you might have on mobile apps or Iot apps and that sort of thing. Sitting beneath the client here is the delivery tier, so that is going to be tools like site personalization tools that have. I’m sure a lot of you in the room have used at some point in time that really get the content to the client tier.
[16:14] I’m sitting beneath the delivery tier is the aggregation tier, so that’s basically saying, OK, the data, how do I deliver the data up to the delivery tier so that the delivery tier can get a customer experience to the client here and CDP sit within the aggregation tier. Sitting beneath the aggregation tier is the services here and that’s really where the data is minted, if you will, that this is going to be some of your purchase data. This could be data from email opens and and so forth. So we’re focused on the aggregation tier in today’s session and, uh, you know, to make some good notes about key things you should look for in your aggregation tier or your CDP. The first is that it should be flexible, really shouldn’t enforce rules like everything has to be within a pixel or tied to some overall id or anything like that.
[16:59] It should be able to get any sort of data that’s tied to a user in any sort of format and be able to get that out to delivery tier, the second and third event driven and asynchronous or almost like the push and the pull model. Event-driven means when something happens, the aggregation tier must immediately look to deliver that data to the delivery tier. But asynchronous meaning that whenever the delivery tier needs to pull data, it should be available as well, not based upon a user action or anything like that. Rather the delivery or making the request to the, the aggregation tier and of course high performance at incredible scale. That kind of explains itself, so when you’re thinking about designing your aggregation tier, or maybe another way to think about this is when you’re thinking of selecting a CDP, what are the key things you need to keep in mind?
[17:45] The first is the CDP has to have a holistic approach, uh, to data. So what we mean by that is that, uh, there’s use cases where you’re going to need to stream raw data in real time, uh, but there’s also use cases where you’re going to need segmentation, right? I don’t need this in real time, but I need to send a segment of users to some sort of platform. So you have to have that holistic approach and not say, all right, this is only a orders is only a segment or it’s only real time. You have to be able to do all of those things. The next two things go hand in hand. A combining first and third party data. So that means whether it’s your data that’s from your crm system, a behavioral data from a website or some sort of app or third-party data that you have access to the aggregation tier must have the ability to manage that. But besides that it has to be able to resolve identity and this is perhaps one of the most important things. All those things that I just mentioned about first and third party data, a lot of that data is going to be tied to various different types of it. Sometimes you have cookie IDs, sometimes you have emails, sometimes you have mobile device IDs. Your aggregation tier needs to be able to manage all of those identities and get the right data against the right identity downstream.
[18:53] Yeah, a control and filter is also important. That’s, you know, especially with all the talk about GDPR, I’m sure there’s some session for GDPR requirements at this conference today. But in a CDP you need to have the ability to control and filter what data goes where and under what conditions. So your ESP is pretty useless if you don’t send an email address, but a lot of your paid media tools, it’s a no notice in any PII there. So you need to make sure that your CDP has the ability to control exactly which data points are going. We talked already about audience segments. Finally, and this is perhaps one of the more powerful features of a CDP is transformed data. So this could be things like if you’re a retailer and you have additional product information that’s perhaps not available to your client when it comes to your CDP, can I enrich that product view or that adds a card with additional product data that’s not available to the client. Or you know, we know it’s a real world and implementations don’t always go as planned. So can I use a rule so programmatically say that, oh well, in this app I get USA and my other app, it says United States, I want to change both of them to us so that my data is consistent when I send that downstream. So you need to have the ability to transform data.
Once you’ve designed the aggregation tier and the next step in creating your CDP architecture is to really map the relevant services. Are the inputs and outputs a your. Are we getting data from a mobile app, from a website, from some internal CRM system, and then where is that data going to go? Are we sending it to marketing automation tools that might send push messages or emails? Are we going to send it to an ad server for paid media analytics and so forth, but perhaps this example’s better illustrated through a reference architecture if you will. And this is actually one of our customers, a few skated architectures and I’ll give you some background in terms of how they worked with us. So really when they. When we first started working with this customer, the first thing is that we’re in scope where all we’re going to implement Sdk in Ios, android and the Amazon fire tv APP, and their first use case out of the gate was getting data into their analytics tool, in which case was adobe, formerly omniture analytics.
[20:59] So off the out of the gate, we did that with that behavioral data already in the CDP, however that quickly opened the door for them to leverage that same data for other channels. The next one they pursued, there was marketing or automation and they use Urban Airship for that to send emails and, and push notifications. And really that was a key inflection point for this customer because that’s when they enabled. And David alluded to this earlier, so the bi-directional nature of integrations within a CDP. So yes, the behavioral data was coming from iOS, Android and Amazon Fire to the marketing automation tool. But also we were getting the campaign data about which push messages and emails to users open back into the CDP, oh guess what? We had implemented the connection to the analytics tool before. So those email and push opens started going to the analytics tool.
[21:48] So then that same story repeated itself with attribution. So they, they use Kochava for their attribution to say, OK, well we’re going to send the behavioral data that are my goal events for my campaigns to coach Java. Great. So that’s getting their Kochava to provide the great reporting that they do and attribution. Let’s get that post back data about the installs, you know, which campaign the user converted on, what network and so forth, back into the CDP. Guess what? Now of course it’s going to my analytics. We did that already. I can also layer that and use that for intel to help my push my email messages because I’ve connected the marketing automation tool. So I think you can see the pattern developing, right? There was additional data that they were getting from video engagement from a convenia feed. They were also activating that sort of on the publisher side or they use the crux or salesforce as their dmp.
[22:34] So let’s get the behavioral data into the DMP and leverage that for advertising facebook and so forth. So you know, before you make any of the connections that I spoke about just before, uh, the one thing and we take a lot of time to do, to set mparticle is really how do we map our data and our ideas and what we do and we call this exercise data planning and, and it’s really something about sitting down and saying, OK, we had all those channels that you saw on the previous slide, iOS, Android, Roku website, and so forth. What does the data that we’re going to be getting from those channels? And let’s, let’s create a plan to say, all right, you know, user action by user action in any one of those channels. When a user does this, here’s what we’re going to call that event.
[23:19] And here’s the metadata we’re collecting a around that event. And that really is an exercise. And David alluded to this before, right? CDP is hit across the entire organization. So that data planning exercise requires you to get the right folks from the organization in the room. You want to get your marketing team in the room, your analytics team, your product team. You might have your agency team at teams in charge of monetization in the case of this publisher and say, OK, hey team, what are the APIs that you’re working towards? What types of campaigns are you running? What do you need to see here and in your analytics, and really get everyone on the same page about data and nomenclature and create a plan for that, and you should do that for every single channel that you, whose data you want to enter into your CDP.
[24:01] And then alongside that, and we alluded to this earlier, you need to create a strategy for your identity, which is super important. All those different channels will have data against different identities. Sometimes it’s a cookie. It, sometimes it’s a device id, email address. Uh, sometimes it’s a huge facebook it, you see some of those examples over there and map, OK, well I’m getting perhaps an email address from the website and the mobile app create. I have made that connection. I can tie that information to one user identity. And then you propagate that in the manner that I described earlier on the other side. The other importance of identity kind of goes back to a couple like regulatory environments, right? So a publisher, if they’re a video publisher, like the one that, uh, we had shared before a Hipaa, right? The video privacy act is, is key there. So if a user has not logged in, then I can’t associate any of that anonymous behavioral data with a logged in profile. So your CDP needs to be, have the ability to do that, give you the control to say that when I’m not logged in versus when I’m logged in, that should be one user or that should be separate users and not let the API dictate that. Let your business rules dictate that and an implement that in your CDP.
[25:13] So, you know, I said a lot of stuff and there was a lot of good use cases and stuff on the slide. But I think the thing that, you know, everyone thinks that, you know, sees when they look and think about this is that, you know, there’s a lot there, right? And it’s not something where you implement a CDP and five months you’re at that stage, right? It’s, it’s a journey that takes a lot of time to do that. So what we always recommend our customers do, and we work with them to do, is that let’s get some quick wins, right? So in the previous slide, when I talked about how we work with the publisher, the quick win for them was getting the data into the analytics tool, right? That was, all right, let’s get a thorough data collection model going. The data’s in the analytics tool so that allow us to get a deploy analytics a little bit faster.
[25:53] That allowed us to deploy a Java a little bit faster, get data to my DMP faster first quick win. Once I started doing that, once we started doing that, then the next step was OK, I could unify the insights. I was getting that feed data from the marketing automation tool about what emails users were opening, what pushes users were opening, getting the attribution data from coach back into the CDP. So I was able to unify insights as those connections get made. That’s when you really work your way and you’re truly orchestrating marketing at least the way we think marketing should be orchestrated, where I am taking all of these data points and user actions across all these channels and using to drive marketing across various channels.
Latest from mParticle
How Marks & Spencer drove revenue growth with mParticle's real-time segmentation and personalization
mParticle’s real-time data insights and segmentation fuel Joe & The Juice loyalty program growth