AI is only as good as the data it’s built on. Make sure your tech is invested in both

In 2024, every tech platform is touting robust AI capabilities—and every marketer is under pressure to figure out how to use them.

But selecting the right AI to help power your marketing efforts is much like the process of selecting a CDP, where Nick Kobayashi, group product manager at Klaviyo, says one of the biggest challenges is “seeing through the bullshit.”

Like people, AI models don’t just pop up fully formed and functioning. They need to be educated on how to make the decisions they do—and that means that in terms of earning your confidence as a marketer, what’s behind the scenes with AI matters. A lot.

The key factor in whether you can trust—and use—a platform’s AI has less to do with the shiny bells and whistles we typically associate with AI marketing, and more to do with a simple question: What’s the AI studying?

For many AI-powered marketing technology platforms, the answer is easy: AI majors in data. But dig beyond the surface, and things get a lot more complicated.

For example: How clean is the data that taught the AI? How comprehensive and historical is it? How up to date, expansive, and reliable is it?

Read on to learn more about what makes data AI’s most influential teacher—and what kinds of martech platforms are actually up to the challenge.

The behind-the-scenes data issues that are front and center with AI

The gap between collecting data and using it wisely pre-dates AI. For marketers like Richard Cowell, VP of strategy and operations at Citizen Watches, it’s a tale as old as time.

“It’s all good data we’ve harvested,” Cowell says. “But being able to do something with it has been difficult for us.”

He’s not alone. According to recent Klaviyo research, ecommerce marketers have plenty of data. But they’re still unable to use that data to answer even basic questions about their customers, let alone drive revenue through personalized marketing, because:

• The data isn’t clean: 71%
• The data isn’t stored in a consistent format: 57%
• The data isn’t easily accessible: 53%

When you consider the typical martech stack, it’s no wonder. The average small or midsize company uses 10 or more software applications to engage with customers through various touchpoints. Each carries out a separate function, such as marketing automation, customer feedback syndication, and data management—and each creates its own unique set of data.

“Integrating and maintaining these tools is a massive headache,” says Matt Preyss, lead product marketing manager at Klaviyo. “If you have to make one change, you have to make multiple development changes.”

“Expanding use cases requires more resources to actually achieve that use case, and adding more tools requires more developer resources,” Preyss explains. “It continuously adds up.”

The more unwieldy, redundant, and disjointed the martech stack, the more difficult and expensive it is to not only store and access but also leverage your data—and that’s reflected in the marketing experiences you can, and cannot, create.

Imagine, for example, that you send a promotional text, followed, minutes later, by an email with the exact same messaging. A subscriber who loves your brand enough to have opted in to both SMS and email is annoyed by the duplicate content—and promptly unsubscribes.

Or imagine one of your customers typically buys a birthday gift for a friend or family member around the same time each year. If your marketing platform doesn’t use historical data to predict future behavior, your brand might not be top of mind around the time they’re most likely to make a purchase. By the time the birthday rolls around, your customer hasn’t heard from you for a while, so they place an order with another brand.

If your tech stack has a data problem, your team isn’t the only party that pays the price. Your customers do, too.

Better data = better AI = a better customer experience

40% of businesses say that improving the customer experience is the No. 1 incentive for using AI. But because of the symbiotic relationship between data and AI, red flags like data silos and inconsistencies get even redder when marketers start using AI to build and maintain smarter digital relationships.

Remember the principle “garbage in, garbage out”? Since AI models are built on machine learning programs that rely on historical data, training AI is like that. A comprehensive, foundational data infrastructure is necessary to properly train AI models on things like accurately predicting a customer’s future behavior.

If your data is disjointed, inconsistent, unreliable, or all of the above, the consequences are as predictable as they are disastrous: The AI learns wrong. And when the AI learns wrong, it can’t do its job of improving the customer experience.

It’s nothing short of essential, then, to have a framework for collecting, storing, and accessing well-organized data over the lifetime of a customer across all major touchpoints, from real-time browsing behavior to transaction behavior from years ago.

The AI X factor: breaking down customer data silos

To crack the code on using AI to run successful marketing initiatives, companies first need to solve their customer data problem—and solving the customer data problem is exactly why Klaviyo built its solution from the bottom up, starting with data infrastructure and building the application interface on top of it, according to Preyss.

Here are the 5 key pillars of Klaviyo’s approach to data collection and storage that make it such a strong home base for AI ideation and execution:

1. Collecting raw event data and storing it unaggregated at the customer profile level

According to Preyss, Klaviyo processes over 2B shopper data points or events per day, “which we then collect and store within the shopper profile.”

Klaviyo customers, then, “have that activity data to understand their customers’ behaviors in addition to all of the historical data,” Preyss explains. “We store this data unaggregated for segmentation and personalization, meaning our customers can get very granular and precise with their engagement strategies.”

2. No archiving customer data

Most organizations archive data after a set time period—anywhere from 30-180 days, Preyss observes. And once that happens, “it takes more time to access that data and use it.”

Not so with Klaviyo. “We do it differently,” Preyss says. “We don’t archive data. It’s easily accessible in real time for our customers, enabling them to execute faster without any usage restrictions, like creating segments or reports that go beyond a certain length in time.”

3. Flexible data storage and ingestion limits

With more than 300 pre-built integrations “that plug into all these different data sources,” Preyss says, Klaviyo empowers businesses to unify their data into single customer views that allow them to “actually ‘see’ their whole customer.”

Effectively, this means marketers can “bring in all their data into Klaviyo and store it indefinitely in a way that’s adaptable, scalable, and flexible,” Preyss adds.

4. No pre-configuration requirements

Data teams spend countless hours cleaning, configuring, and mapping data to make it usable for things like machine learning. But Klaviyo ingests data in any format, Preyss points out—which means “no mapping is needed.”

That accelerates implementation and time to value, Preyss says—and it also gives marketing and technical teams “more time to work on higher-value activities.”

5. Flexible data partitioning for segmentation

The result of all of this, Preyss says, is that businesses can use Klaviyo to create “hyper-targeted segments” by combining any piece of profile or event data across their tech stack:

• Metadata
• Profile data
• Historical data
• Predictive analytics
• …and more

And, Preyss adds, it all happens “within an intuitive interface to ensure our customers can build and execute quickly.”

A platform built for the new era of machine learning

At a time when 75% of marketers are being asked to cut martech spend to deliver better ROI, according to Gartner research, “seeing through the bullshit” of AI product marketing is more important than ever.

If that’s you, think about it the same way you would if you were picking out a school for yourself or your kids: AI, as a student, is like all students. In order to have the best chance at doing its best work, it needs to learn from the best, too.

If there’s one thing AI needs in order to help you better understand your customers and use those insights to drive business results, it’s not just another ChatGPT integration. It’s a whip-smart, rock-solid data foundation—and Klaviyo, Preyss says, “is and always has been a data platform at its core.”

In other words, tenure goes a long way. Klaviyo integrates marketing automation with a centralized data repository, with AI and machine learning tools that improve the customer experience deeply embedded throughout the platform.

“We have a lifetime’s worth of clicks, visits, add-to-cart events, purchases, email/SMS/mobile push opens, and more powering our content generation, predictions, and recommendations,” Preyss explains.

Klaviyo not only includes all the predictive, generative, and autonomous AI features you’ve been reading about for a year. It also drives better results because data science is built in to the platform, not bolted on as an afterthought.

And that’s important. Because the most valuable AI marketing feature, in the end, isn’t a feature at all. It’s a philosophy.