"AI can be extremely powerful, but it can also be problematic if you start with that technology," Nick Craig, head of go-to-market at Rokt mParticle, said on a recent episode of "Behind the Numbers." "Start with the use cases, start with the value, then let AI take hold as the enabler to reach it."

Downstream AI tools, from creative generation to campaign orchestration, are becoming commoditized and crowded as marketing platforms rush to add AI capabilities. The real competitive advantage lies not in having AI, but in the quality and comprehensiveness of the customer data that trains these AI agents.

Craig joined our analyst Yory Wurmser to discuss why upstream data infrastructure determines whether marketing AI succeeds or fails.

Where AI agents actually get their power

Most AI innovation in marketing is happening at the downstream level, in email service providers, demand-side platforms, and creative tools. But these agents face a fundamental limitation.

"Downstream agents, while extremely valuable, they tend to be limited by the quality of inputs that they receive," Craig said. "If you're in a downstream platform at this point, you're creating a segment, that AI layer within that segment can only be trained on that limited data set."

The solution is building AI capabilities upstream, where foundational customer data is captured. Craig's company has operated in this space for 13 years, focusing on real-time behavioral data and historical information that creates a complete customer picture.

"It's critically important that we're capturing this real-time information and we're able to create impactful campaigns downstream," Craig said. "Humans operate in real time. We don't operate in batches."

Marketers evaluating AI tools need to ask where their data lives and what data points are available. An AI agent can only make recommendations based on the data it can access. If you're sending limited data sets to individual platforms, that's all the AI has to work with.

Speed and automation unlock new marketing capabilities

AI agents are enabling marketers to work far faster, eliminating traditional bottlenecks that required data engineering and analytics teams for every campaign adjustment.

"No longer do you have to wait weeks to kick off a campaign to implement a new strategy. We're not even talking about days. We're at the point with the tool sets in place that in hours you can actually develop and execute on something," Craig said.

The technology enables natural language segment creation. Instead of manually building audiences, marketers can describe their target: "I want to target users who have visited my website but haven't signed up for a premium subscription." More sophisticated requests allow the AI to analyze all available data and suggest multiple segmentation strategies.

"You're a marketing VP trying to drive premium subscriptions. Please, as the agent, help me make money for my business," Craig said, describing how marketers are now using these tools. "This is finally when these agents are becoming true strategic partners of these marketers hand in hand."

Wurmser noted that this speed advantage relies on proper data foundation. "If you have that ability where everything is connected and everything is well defined on the base level, then you can move a lot quicker," he said.

The feedback loop tightens campaign performance

Real-time data ingestion creates faster optimization cycles than traditional campaign management allowed.

"Historically, as a marketer, you have a campaign, you set it up in a manual fashion, you execute it, you review the results, you go back to point one," Craig said. "We are in real time ingesting the behavioral information. We're starting to understand which advertisements are resonating with users with the data, and being able to action on that very quickly."

Batch processing gives way to continuous optimization, where AI agents can identify what's working and adjust campaigns without human intervention.

Wurmser emphasized thinking beyond efficiency gains.

"People think when they think about agents, they're thinking a lot about efficiency," he said. "But I think they need to spend more time thinking about what does that enable, and what can you do with that."

The recommendation: Start with business goals and use cases, not the technology.

"Understanding the business, understanding the potential goals that you have, and then AI really is just the enabler," Craig said.

Listen to the full episode.

This was originally featured in the EMARKETER Daily newsletter. For more marketing insights, statistics, and trends, subscribe here.