True 1-1 marketing isn’t segment or persona-based - Conversant’s AI personalization story

Profile picture for user barb.mosher By Barb Mosher Zinck February 19, 2018
Summary:
True 1-1 personalization is still a rare achievement. Conversant says their AI and machine learning approach is a way forward. But how is privacy protected? Can marketers learn to trust a "black box"?

personalization-model
Four to five years ago, no one cared about cross-device marketing. Now it’s one of the biggest problems for both marketers and martech vendors.

If you can figure out how to understand who a customer is across channels and devices, then you can also do one-to-one marketing. This, says Dave Scrim, VP, Product at Conversant, is something his company started looking to solve eight years - and there are three key challenges.

The first is identity

As Scrim points out, consumer identities are fragmented across channels. Each device a person owns or uses has its own view or profile of that person. Some of these profile elements are similar and don’t change often, but others - like cookies - fluctuate. To understand who you are talking to and what to say or show requires the ability to stay connected to people regardless of what device they are using.

You have to build identity around real people said Scrim. There’s cookie identification, email address, even device id identification, but only a few companies are building identity around real name and address identification. These companies anonymize information to a number, but you know it’s connected to a real person. Google, Facebook, even Amazon are doing this type of identification, but they are walled gardens – they use the identity for their environment alone. Conversant also builds identity around a real person, but it is not exclusive to running in any single environment.

Why can’t (and shouldn’t) every company do it? Scrim said that companies can’t do it at the scale required to make it meaningful. They need a network effort of many companies doing it and sharing their data. Conversant does this by connecting the data of over 4,000 retailers and their customer transactions, matching the customer in the background but not sharing specific customer details. Instead, they use the information as connection points to know where to deliver a message and what message to deliver.

Then you need to develop a 360-degree profile

You need a profile that not only connects the person to their online activity but also their offline behavior so that you have a true 360-degree view of that person. For Conversant this is done through its relationship with Epsilon (its parent company). Conversant has a large database of anonymous non-PII data, and it combines that data with Epsilon’s large database of offline PII data to provide both views of the consumer.

Now, what do you say to that person?

Once you have that 360-degree profile, with a clear identity across devices, you now must decide what you are going to say and when you are going to say it. Scrim provided the example of having 10 million customers and a billion permutations of what you could say, between the channel, the product, and the offer. So how do you decide? This is far from the mapping you can do when you define personas, create segments and map offers to those segments.

And this is where AI comes into play.

Machine learning enables 1-1 marketing

Most marketing automation systems work like decision-trees. One step executes and depending on the result; another step executes and so on. That’s a form of one-to-one marketing, but it’s not true one-to-one.

Scrim described how Conversant works, explaining that everyone in its database is scored for a given offer and that score is constantly changing based on what a person is doing online, and offline, and other business rules identified. Offers are then sent out using that score. But they aren’t sent out by marketers directly; the software does it with the help of AI.

This is the value that machine learning brings to personalized marketing. The ability to analyze millions or more transactional and behavioral data, it learns what works for a given person and optimizes the offers and messages for that person and each program. No marketer could do this.

To give you some context: Conversant brings in 120 million transactions per day, see 200 billion website activities on individuals per day and performs one billion machine learning decisions every five minutes.

The network supplies the transactions and gives the results of marketing efforts. The machine learns what works and what doesn’t and it’s constantly learning. Scrim referred to it as a closed feedback loop. But also key to Conversant’s ability to do this one-to-one marketing successfully is the media component of its business. He said they are buying media at scale all the time and they see every person on average 180 times per day. These data points, even though anonymous, help the machine understand what to do for each person.

Content in a 1-1 message

You now know who the customer is and have figured out the right offers to present, but doesn’t this require a lot of content - individually created? Most companies have creative teams that can only push out a certain amount of content every day or week. How do they feed this machine?

There’s a paradigm shift that involves technology, Scrim said. “It’s the balance between a creative team that building the baseline concepts and automation that closes out the last mile.” He talked about treatments - the format a message is delivered (video, large-form display, etc.) and the device, being separated from the message. When a person goes to a website, depending on the device they are using, a treatment is loaded, which then connects to the database to determine the message to load into the treatment for that person.

Let’s talk about video for a minute here, because you can’t do 1-1 video marketing, can you? You can. I’ve seen small-scale examples of this from video marketing providers, but Conversant goes a bit further. Scrim said that portions of the video are the same, but the offer or the product changes depending on the profile. You build in sections for the offer or the product and then place dynamic text that is specific to the customer.

Sounds like a black box system

All this machine learning and automation - it’s taking control away from marketers, right? They have to trust the black box system. But most marketers aren’t quite ready to give up that kind of control. They want to know what’s in the black box and they want the ability to insert control if they need to.

Scrim said Conversant understands this need and they are working to resolve this challenge. What is needed is a tool that gives marketers transparency into what’s happening. They can see the profile, data, decision rules, and they can override the system, creating their own models and business rules and push that model out. It’s probably not something they would want to do all the time, but circumstances might arise where they need to do something special.

The trick, however, is that Conversant works on a performance-based model. In cases where customer infuse their models, they must take responsibility for the performance of those campaigns.

My take

Conversant knows it’s not the only company leveraging AI to improve marketing. But their approach involves building a network of connections that provides so much data that improves the decisions the machine makes in a way that not every marketing technology vendor can do.

I wrote about another company - Adgorithms - that also leverages machine learning to send the best messages to customers, but it doesn’t have that larger view of the customer across multiple retailers. IBM Watson is another AI-based solution that provides insights based on thousands, if not millions of data points.

These are three examples of the power of AI and machine learning.

But Scrim also said something that makes a lot of sense. In the rush to perform highly personalized one-to-one marketing, there is still a need to do broader marketing campaigns and activities that reach out to many people. I think these non-personalized campaigns support brand awareness and brand messaging that will resonate with people and help promote the company overall.

Just because a company is great at providing me with products and offers personalized to my specific needs, doesn’t mean I will think it can do the same for my best friend, or a work associate. It might help to see things from the company not through the microscope that is my personal need, but through a wider lens that would make me want to advocate for it. Just something to think about.

In the meantime, brands can feel free to personalize my offers; I certainly won’t complain about relevant ads and marketing.