There is enormous potential in new AI technologies. How can you get the most out of your AI investment? Build your programs on a solid foundation.
There is enormous hype these days about the promise of all things artificial intelligence (AI). As a marketer, a technologist, or both, you are no doubt under pressure to somehow tap into this new, magical capability and improve every aspect of the business. While AI and machine learning (ML) aren’t magic, as I’ve pointed out before, they do hold enormous potential if implemented on top of the right foundation, and you use them in the right ways. Here I’ll explore these foundational elements, and specifically what you need to have in your marketing technology stack, to ensure you’ve set yourself up to get the most out of these new advances in technology.
Before you commit to a marketing program dependent on AI investments, get a clear vision of your goals. I know that this sounds like obvious advice, but you’d be surprised how many times I’ve seen customer data initiatives with goals that stopped with collecting data into a “data lake” with no plan for how to access the data, or how to govern it over time.
This approach always leads to problems down the road, often because the organization didn’t consider details like the data structures, they didn’t account for different types of collected information, or they had the wrong granularity. The result is all of the data is, indeed, in one place—but you can’t do anything with it and it just sits there, becoming more obsolete by the minute.
Bottom line: Applications that are powered by AI and ML need data to operate; it’s the fuel they run on. It’s important to understand your end game—or at least your initial goals—so you can ensure you are collecting the right data in the right way and that you can deliver the right answers and outcomes.
Customer data is at the heart of modern marketing initiatives, and machine learning technology thrives on large datasets. The more “signals,” or data, available to ML algorithms the more accurate the predictions will be.
The nature of AI algorithms is that they look for combinations of factors that need to occur to produce the goal you desire, then find those combinations in the future to achieve more of the things you want to happen. Seeing data from many sources or channels gives it a wider set of signals or factors to operate on, and produce more of the situations where you can achieve your desired goals.
Bottom line: Start with a large dataset collected from diverse sources to ensure your ML systems give you the best results.
Once you’re pooling data from multiple tools and systems, you need to have an identity resolution strategy. That is, you need to have policies and approaches to how you are going to resolve an individual’s profile from data collected via multiple marketing tools and customer devices.
The best way to establish this strategy is to start small. Choose an achievable project with a rich dataset that will help you realize your desired outcomes. This way your organization can identify which behaviors and identifiers are important to your program and first focus on that data.
Some of this work is very straightforward, while some of it is less so. For example, when a user clicks on a link in an email you’ve sent them, it’s simple to connect their behavior on your website to that address. That level of identity resolution is standard practice for any customer data platform, and many marketing tools out there, for that matter.
Whose identity are you resolving?
Things get more complex when you have many devices associated with a single profile. For example, multiple devices, each of which is actually used by different people, can be attached to a what we call a “household account.” Consider this example: You are a parent, with two kids away at college. You share with the kids one loyalty account at a chain coffee shop because you want some kind of small reward for those late-night coffee runs you're paying for anyway.
Now, the coffee chain may want to connect email data to their loyalty program, but that’s only going to map to you, the account holder. And perhaps the brand wants to enrich your profile with geographic data. Should that geographic enrichment be applied to all three devices? What about the kids that are living 1000 miles away, in different directions? Whatever plans for content personalization that coffee chain offers likely won’t make sense to the college kids in locations different from yours.
Now also consider how the coffee chain brand will use your identity for the purpose of usage analytics. How would your profile, which tracks coffee purchases for three people, compare to a profile of a single individual? Any comparisons between these two profiles could be problematic because the data does not actually represent usage statistics of two individual customers.
Viewing and privacy considerations
Consider how widely you’ll share customer profiles, and who will have a centralized view. In large conglomerates, you might make available to marketers data collected by their own brand, and exclude data from other brands. You could further restrict their view of that data down to their specific geography, or share some aggregate signals across brand or geographies, but not detailed data.
You also want to have the ability to apply policies to this data to ensure you’re using only the data that you are legally allowed to use. In the EU, with the GDPR now in effect, that means you can only use the data that was collected under consent from the data subject. In the US, states are passing their own data privacy regulations and marketers need to make sure they're complying with these new rules.
Bottom line: Have policies and approaches to resolving individual profiles from data collected via multiple channels and devices. And make sure you have a documented plan to ensure you're using customer data in accordance with any privacy regulations that may apply to your organization.
Data is the fuel for machine learning and AI, and like your car or body, you need to feed it the cleanest, purest stuff you can. Garbage in, garbage out, as they say. To ensure you have clean data, you need to think through your data resolution and validation strategy. Will you throw out data with missing elements? Is there a threshold of minimum viable data? Can you fill in holes from other sources? If a particular dataset is particularly lossy or noisy, can you live without it and still reach your goals?
The dirtiest data is, not surprisingly, the least valuable in terms of signal strength and predicting outcomes. Much of the demographic data that brands feel compelled to use is the most problematic, if only because it is often outdated, offering the weakest signal of customer interest or intent.
Meanwhile, the often overlooked “digital exhaust” data that is the footprint of our first-party interactions with customers typically carries incredible signal strength. I’ve seen many of our customers see huge gains by using behaviorally-derived, first-party data instead of demographics-based data.
Bottom line: Think through how you’ll clean your data, what you can live without, and how much impact a piece of data really has for your goals and desired outcomes.
Customer interests shift and change over time and brands that can give people what they want in the moment will win. The last foundational consideration is the data pipeline velocity that will allow you to know what customer want in real-time. One of the most promising things about ML and AI technology is the ability to react at lightning speed, but the systems need fast and constant data inputs.
Speed matters in many cases, and depending on your goals, you need to consider how quickly each of your sources of data can inform your central view of the customer. You also need to know how quickly you can update your machine learning models and make automated decisions in the moment so you can execute the right experience through the right delivery channel to that customer.
Each of these decisions will impact your customer experience pipeline, and you need to evaluate which pieces are adding latency to the process. Based on your needs, you should consider which sources can meet your velocity needs, and make sure your data repository, decisioning, and delivery tools can respond at a similar pace.
Bottom line: In order to meet your customers’ real-time needs makes sure your data sources are rich and always available.
A solid foundation will maximize AI success
There are great potential and promise in the rapidly evolving space of AI-powered digital marketing, but you have to make sure your foundation is solid. These five core principles will make sure you’ve set yourself up for success and maximize the potential value you can derive from these new technological advances.