(Watch) I joined Ryan Joe, AdExchanger Senior Editor, to talk about Salesforce’s acquisition of Krux and how Salesforce is thinking about artificial intelligence, the role of data management, and applications for better marketing.
How Beacons Might Alter The Data Balance Between Manufacturers And Retailers
As Salesforce integrates DMP Krux, Chris O’Hara considers how proximity-based personalization will complement access to first-party data. For one thing, imagine how coffeemakers could form the basis of the greatest OOH ad network.
For years, marketers have been talking about building a bridge between their existing customers, and the potential or yet-to-be-known customer.
Until recently, the two have rarely been connected. Agencies have separate marketing technology, data and analytics groups. Marketers themselves are often separated organizationally between “CRM” and “media” teams – sometimes even by a separate P&L.
Of course, there is a clearer dividing line between marketing tech and ad tech: personally identifiable information, or PII. Marketers today have two different types of data, from different places, with different rules dictating how it can be used.
In some ways, it has been natural for these two marketing disciplines to be separated, and some vendors have made a solid business from the work necessary to bridge PII data with web identifiers so people can be “onboarded” into cookies.
After all, marketers are interested in people, from the very top of the funnel when they visit a website as an anonymous visitor, all the way down the bottom of the funnel, after they are registered as a customer and we want to make them a brand advocate.
It would be great — magic even — if we could accurately understand our customers all the way through their various journeys (the fabled “360-degree view” of the customer) and give them the right message, at the right place and time. The combination of a strong CRM system and an enterprise data management platform (DMP) brings these two worlds together.
Much of this work is happening today, but it’s challenging with lots of ID matching, onboarding, and trying to connect systems that don’t ordinarily talk to one another. However, when CRM and DMP truly come together, it works.
What are some use cases?
Targeting people who haven’t opened an email
You might be one of those people who don’t open or engage with every promotional email in your inbox, or uses a smart filter to capture all of the marketing messages you receive every month.
To an email marketer, these people represent a big chunk of their database. Email is without a doubt the one of the most effective digital marketing channels, even though as few as 5% of people who engage are active buyers. It’s also relatively fairly straightforward way to predict return on advertising spend, based on historical open and conversion rates.
The connection between CRM and DMP enables the marketer to reach the 95% of their database everywhere else on the web, by connecting that (anonymized) email ID to the larger digital ecosystem: places like Facebook, Google, Twitter, advertising exchanges, and even premium publishers.
Understanding where the non-engaged email users are spending their time on the web, what they like, their behavior, income and buying habits is all now possible. The marketer has the “known” view of this customer from their CRM, but can also utilise vast sets of data to enrich their profile, and better engage them across the web.
Combining commerce and service data for journeys and sequencing
When we think of the customer journey, it gets complicated quickly. A typical ad campaign may feature thousands of websites, multiple creatives, different channels, a variety of different ad sizes and placements, delivery at different times of day and more.
When you map these variables against a few dozen audience segments, the combinatorial values get into numbers with a lot of zeros on the end. In other words, the typical campaign may have hundreds of millions of activities — and tens of millions of different ways a customer goes from an initial brand exposure all the way through to a purchase and the becoming a brand advocate.
How can you automatically discover the top 10 performing journeys?
Understanding which channels go together, and which sequences work best, can add up to tremendous lift for marketers.
For example, a media and entertainment company promoting a new show recently discovered that doing display advertising all week and then targeting the same people with a mobile “watch it tonight” message on the night of it aired produced a 20% lift in tune-in compared to display alone. Channel mix and sequencing work.
And that’s just the tip of the iceberg — we are only talking about web data.
What if you could look at a customer journey and find out that the call-to-action message resonated 20% higher one week after a purchase?
A pizza chain that tracks orders in its CRM system can start to understand the cadence of delivery (e.g. Thursday night is “pizza night” for the Johnson family) and map its display efforts to the right delivery frequency, ensuring the Johnsons receive targeted ads during the week, and a mobile coupon offer on Thursday afternoon, when it’s time to order.
How about a customer that has called and complained about a missed delivery, or a bad product experience? It’s probably a terrible idea to try and deliver a new product message when they have an outstanding customer ticket open. Those people can be suppressed from active campaigns, freeing up funds for attracting net new customers.
There are a lot of obvious use cases that come to mind when CRM data and web behavioral data is aligned at the people level. It’s simple stuff, but it works.
As marketers, we find ourselves seeking more and more precise targeting but, half the time, knowing when not to send a message is the more effective action.
As we start to see more seamless connections between CRM (existing customers) and DMPs (potential new customers), we imagine a world in which artificial intelligence can manage the cadence and sequence of messages based on all of the data — not just a subset of cookies, or email open rate.
As the organizational and technological barriers between CRM and DMP break down, we are seeing the next phase of what Gartner says is the “marketing hub” of interconnected systems or “stacks” where all of the different signals from current and potential customers come together to provide that 360-degree customer view.
It’s a great time to be a data-driven marketer!
Chris O’Hara is the head of global marketing for Krux, the Salesforce data management platform.
Salesforce unveiled its Einstein AI platform this year, baking predictive algorithms, machine and deep learning, as well as other data analysis features throughout its Software-as-a-Service (SaaS) cloud. Einstein is essentially an AI layer between the data infrastructure underneath and the Salesforce apps and services on top. The CRM giant is no stranger to big money acquisitions, most recently scooping up Demandware for $2.8 billion and making a play for LinkedIn before Microsoft acquired it. The Krux acquisition gives Salesforce a new, data-driven customer engagement vector.
“We’re working to apply AI to all our applications,” said Eric Stahl, Senior Vice President of Marketing Cloud. “In Marketing Cloud, Krux now gives us the ability to do things like predictive journeys to help the marketer figure out which products to recommend. We can do complex segmentation, inject audiences into various ad networks, and do large-scale advertising informed by Sales Cloud and Service Cloud data.”
As Salesforce and Krux representatives demonstrated Krux and how it fits into the Marketing Cloud, the data management platform acted more like a business intelligence (BI) or data visualization tool than a CRM or marketing platform. Chris O’Hara, head of Global Data Strategy at Krux, talked about the massive quantities of data the platform manages, including an on-demand analytics environment of 20 petabytes (PB)—the entire internet archive is only 15 PB.
“This is our idea of democratizing data for business users who don’t have a PhD in data science,” said O’Hara. You can use Krux machine-learned segments to find out something you don’t know about your audience, or do a pattern analysis [screenshot above] to understand the attributes of those users that correlate greatly. We’re hoping to use those kinds of signals to power Einstein and do things like user scoring and propensity modeling.
The Einstein Journey Insights feature is designed to analyze “hundreds of millions of data points” to identify an optimal customer conversion path. In addition to its Krux-powered Marketing Cloud features, Salesforce also announced a new conversational messaging service called LiveMessage this week for its Salesforce Service Cloud. LiveMessage integrates SMS text and Facebook Messenger with the Service Cloud console for interactions between customers and a company’s helpdesk bots.
The more intriguing implications here are what Salesforce might do with massively scaled data infrastructure like Krux beyond the initial integration. According to O’Hara, in addition to its analytics environment, Krux also processes more than more than 5 billion monthly CRM records and 4.5 million data capture events every minute, and maintains a native device graph of more than 3.5 billion active devices and browsers per month. Without getting into specifics, Salesforce’s Stahl said there will be far more cross-over between Krux data management and Einstein AI to come. In the data plus AI equation, the potential here is exponential scale.
The term “real time” is bandied about in the ad technology space almost as heavily as the word “programmatic.”
Years later, the meaning of programmatic is finally starting to be realized, but we are still a few years away from delivering truly real-time experiences. Let me explain.
The real-time delivery of targeted ads basically comes down to user matching. Here is a common use case: A consumer visits an auto site, browses a particular type of minivan, leaves the site and automatically sees an ad on the very next site he or she visits. That’s about as “real-time” as it gets.
How did that happen? The site updated the user segment to include “minivan intender,” processed the segment immediately and sent that data into a demand-side platform (DSP) where the marketer’s ID was matched with the DSP’s ID and delivered with instructions to bid on that user. That is a dramatic oversimplification of the process but clearly many things must happen very quickly – within milliseconds – and perfectly for this scenario to occur.
Rocket Fuel, Turn and other big combo platforms have an advantage here because they don’t need to match users across an integrated data-management platform (DMP) and DSP. As long as marketers put their tags on their pages and stay within the confines of a single execution system, this type of retargeting gets close to real time.
However, as soon as the marketer wants to target that user through another DSP or in another channel, user matching comes back into play. That means pushing the “minivan intender” ID into a separate system, but the “real-time” nature of marketing starts to break down. That’s a big problem because today’s users move quickly between channels and devices and are not constrained by the desktop-dominated world of 10 years ago.
User matching has its own set of challenges, from a marketer’s ability to match users across their devices to how platforms like DMPs match their unique IDs to those of execution platforms like DSPs. Assuming the marketer has mapped the user to all of his or her device IDs, which is a daunting challenge, the marketer’s DMP has to match that user as quickly as possible to the execution platform where the ads are going to be targeted and run.
Let’s think about how that works for a second. Let’s say the marketer has DMP architecture in the header of the website, which enables a mom to be placed in the “minivan” segment as soon as the page loads. After processing the segment, it must be immediately sent to the DSP. Now the DSP has to add that user (or bunch of users) to their “minivan moms” segment. If you picture the internet ID space as a big spreadsheet, what is happening is that all the new minivan moms are added to the DSP’s big existing table of minivan moms so they are part of the new targeting list.
Some DSPs, such as The Trade Desk, TubeMogul and Google’s DBM, do this within hours or minutes. Others manage this updating process nightly by opening up a “window” where they accept new data and process it in “batches.” Doesn’t sound very “real-time” at all, does it?
While many DMPs can push segments in real time, the practical issue remains the ability of all the addressable channels a marketer wants to target to “catch” that data and make it available. The good news is that the speed at which execution channels are starting to process data is increasing every day as older ad stacks are re-engineered with real-time back-end infrastructure. The bad news is that until that happens, things like global delivery management and message sequencing across channels will remain overly dependent upon how marketers choose to provision their “stacks.”
The Future Is Dynamic
Despite the challenges in the real-life execution of real-time marketing, there are things happening that will put the simple notion of retargeting to shame. Everything we just discussed depends on a user being part of a segment. I probably exist as a “suburban middle-aged male sports lover with three kids” in a variety of different systems. Sometimes I’m an auto intender and sometimes I’m a unicorn lover, depending on who is using the family desktop, but my identity largely remains static. I’m going to be middle aged for a long time, and I’m always going to be a dad.
But marketers care about a lot more than that. The beer company wants to understand why sometimes I buy an ice-cold case of light beer (I’m about to watch a football game, and I might drink three or four of them with friends) and when I buy a six-pack of their craft-style ale (I’m going to have one or two at the family dinner table).
The soda company is competing for my “share of thirst” with everything from coffee to the water fountain. They want to know what my entry points are for a particular brand they sell. Is it their sports drink because I’m heading to the basketball court on a hot day, or is it a diet cola because I’m at the baseball game? The coffee chain wants to know whether I want a large hot coffee (before work) or an iced latte macchiato (my afternoon break).
This brings up the idea of dynamic segmentation: Although I am always part of a static segment, the world changes around me in real time. The weather changes, my location changes, the time changes and the people around me change constantly. What if all of that dynamic data could be constantly processed in the background and appended to static segments at the moment of truth?
In a perfect world, where the machines all talked to each other in real time and spoke the same language, this might be called real-time dynamic segmentation.
This is the future of “programmatic,” whatever that means.
[This originally appeared in AdExchanger on 8/31/2016]
It’s a hoary old chestnut, but “understanding the customer journey” in a world of fragmented consumer attention and multiple devices is not just an AdExchanger meme. Attribution is a big problem, and one that marketers pay dearly for. Getting away from last touch models is hard to begin with. Add in the fact that many of the largest marketers have no actual relationship with the customer (such as CPG, where the customer is actually a wholesaler or retailer), and its gets even harder. Big companies are selling big money solutions to marketers for multi-touch attribution (MTA) and media-mix modeling (MMM), but some marketers feel light years away from a true understanding of what actually moves the sales needle.
As marketers are taking more direct ownership of their own customer relationships via data management platforms, “consumer data platforms” and the like, they are starting to obtain the missing pieces of the measurement puzzle: highly granular, user-level data. Now marketers are starting to pull in more than just media exposure data, but also offline data such as beacon pings, point-of-sale data (where they can get it), modeled purchase data from vendors like Datalogix and IRI, weather data and more to build a true picture. When that data can be associated with a person through a cross-device graph, it’s like going from a blunt 8-pack of Crayolas to a full set of Faber Castells.
Piercing the Retail Veil
Think about the company that makes single-serve coffee machines. Some make their money on the coffee they sell, rather than the machine—but they have absolutely no idea what their consumers like to drink. Again, they sell coffee but don’t really have a complete picture of who buys it or why. Same problem for the beer or soda company, where the sale (and customer data relationship) resides with the retailer. The default is to go to panel-based solutions that sample a tiny percentage of consumers for insights, or waiting for complicated and expensive media mix models to reveal what drove sales lift. But what if a company could partner with a retailer and a beacon company to understand how in-store visitation and even things like an offline visit to a store shelf compared with online media exposure? The marketer could use geofencing to understand where else consumers shopped, offer a mobile coupon so the user could authenticate upon redemption, get access to POS data from the retailer to confirm purchase and understand basket contents—and ultimately tie that data back to media exposure. That sounds a lot like closed-loop attribution to me.
Overcoming Walled Gardens
Why do specialty health sites charge so much for media? Like any other walled garden, they are taking advantage of a unique set of data—and their own data science capabilities—to better understand user intent. (There’s nothing wrong with that, by the way). If I’m a maker of allergy medicine, the most common trigger for purchase is probably the onset of an allergy attack, but how am I supposed to know when someone is about to sneeze? It’s an incredibly tough problem, but one that the large health site can solve, largely thanks to people who have searched for “hay fever” online. Combine that with a 7-day weather forecast, pollen indices, and past search intent behavior, and you have a pretty good model for finding allergy sufferers. However, almost all of that data—plus past purchase data—can be ingested and modeled inside a marketer DMP, enabling the allergy medicine manufacturer to segment those users in a similar way—and then use an overlap analysis to find them on sites with $5 CPMs, rather than $20. That’s the power of user modeling. Why don’t site like Facebook give marketers user-level media exposure data? The question answers itself.
Understanding the Full Journey
Why is “data science the new measurement?” Because, when a marketer has all of that data at their fingertips, something close to true attribution becomes possible. Now that marketers have the right tools to draw with, the winners are going to be the ones with the most artists (data scientists).
It’s a really interesting space to watch. More and more data is becoming available to marketers, who are increasingly owning the data and technology to manage it, and the models are growing more powerful and accurate with every byte of data that enters their systems.
It’s a great time to be a data-driven marketer!
[This post originally appeared in AdExchanger on 8/12/16]