A customer calls your support line. Your AI agent answers. It has no idea this customer just visited the pricing page, complained last week, or has been with you for three years. The agent is flying blind.
This is happening everywhere. Qualtrics found that AI-powered customer service fails at nearly four times the rate of other AI applications. The problem isn't the models. It's the data infrastructure feeding them.
We all made the same bet a few years ago. When regulators tightened privacy rules and the industry signaled the end of third-party cookies, we invested heavily in first-party data infrastructure. Customer data platforms, data clean rooms, consent management. The logic was sound: own the relationship with your customer, own the data that powers personalization.
That infrastructure is now being asked to do something it was never designed for.
The speed problem
88% of organizations now use AI in at least one business function. Nearly a quarter are scaling agentic AI systems. But those AI agents, along with live sales reps, support teams, and commerce experiences, need customer context delivered in milliseconds, not the hours or days our current data architecture was built to tolerate.
The infrastructure most of us built was optimized for campaigns. Batch-processed outreach. Audience segments refreshed overnight. Customer profiles updated every 24 hours. That cadence made sense when we were sending marketing emails, running sales sequences, and serving display ads.
Conversations operate on a different clock. When a customer is mid-interaction (whether with an AI agent, a support rep, or a checkout flow), you can't wait for an ETL pipeline to finish. You need to know, right now, that this customer just visited the pricing page. That they've been with you for three years. That their last interaction was a complaint.
The bar has been raised from static customer data to real-time customer context. Most first-party data strategies weren't built to clear that bar.
Campaigns vs. conversations
The way I've started explaining this to my peers is simple: first-party data now has two jobs, and most of us only built for one.
Job one is powering campaigns. Marketing automation, sales outreach, nurture flows, retargeting. These are scheduled, segment-based, and largely one-directional. CDPs were designed for this. They collect data from touchpoints, unify it into customer profiles, and activate it for outbound efforts. The latency built into that workflow is acceptable because campaigns operate on similar timescales.
Job two is powering conversations. Live support interactions, sales calls, commerce moments, AI agents. These are real-time, individual, and bi-directional. They require not just data, but context: conversation history across channels, predictions, sentiment signals. Is this customer frustrated? Are they about to churn? What did they just do on the website? A sales rep or AI agent needs to know this in the moment, not after a nightly batch job runs.
The infrastructure gap isn't about marketing versus everyone else. It's about scheduled versus live. Talking at customers versus talking with them.
The shortcut that's backfiring
In the race to deploy AI agents and modernize customer interactions, many organizations are skipping the first-party data integration entirely. They stand up new experiences quickly because integrating with data warehouses and CRMs is hard. The result? Conversations that know what's happening in the current moment but nothing about the customer's history.
The consequences show up fast. More than half of consumers report that AI rarely has context from their past interactions. Only 15% feel that human agents receive the full story after an AI handoff. Meanwhile, 82% of business leaders think they deeply understand their customers. The gap between perception and reality is widening.
It's no surprise that only about a third of organizations have genuinely scaled AI across functions. The rest are stuck in pilots. The data integration challenge is a bigger factor than most want to admit.
From customer data platform to contextual data platform
Think about how CDPs have evolved. The first generation focused on data routing: take data from your website, send it to your analytics tool. The second generation focused on campaign activation: unify profiles and power outbound marketing and sales.
What's emerging now is a third phase. Some in the industry are calling it the shift from "customer data platform" to "contextual data platform." The emphasis moves from knowing who your customer is to having real-time context about what they're doing right now, how they're feeling, and what they need next.
This isn't about abandoning what you've built. Your existing CDP still serves its purpose for campaigns. But you need a parallel capability: a real-time data layer that delivers customer context at the speed conversations require. That means streaming infrastructure. Native integration with the channels where interactions happen. Architecture that treats conversational context as a first-class data type, not an afterthought.
Building for both jobs
The organizations pulling ahead aren't the ones with the most sophisticated AI models. They're the ones that recognized the data infrastructure problem early. They understood that first-party data strategy was the right priority for 2022, but real-time context infrastructure is the priority for 2026.
The question for every enterprise now is whether your architecture can serve both jobs, or just the one you originally hired it for. If you're only built for campaigns, you'll keep deploying conversational experiences that feel impressive in demos but disappoint customers in production.
The good news? You don't have to start over. You have to build the real-time layer alongside what exists. The teams doing this well are treating it as an evolution, not a replacement. They're keeping what works for scheduled campaigns while adding what's needed for live conversations.
Your customers already expect you to know them. Soon they'll expect every interaction, human or AI, sales or support, to reflect that knowledge in real time. The infrastructure you build now determines whether you can deliver on that expectation or keep making excuses for why you can't.
