Real-time is the most overused promise in the customer data platform market. Nearly every vendor claims it. Far fewer deliver it in a way that holds up at scale, across channels, or with data you actually trust. When evaluating the best CDPs for real-time use cases, the conversation should start not with speed alone, but with what the platform does to ensure the data moving quickly is also complete, accurate, and actionable.
This post breaks down what real-time actually means in a CDP context, where common architectures struggle, and what the strongest platforms do differently.
What "Real-Time" Actually Requires From a CDP
Real-time in marketing typically means one of three things: streaming event ingestion, sub-second profile updates, or triggered campaign activation within seconds of a behavioral signal. These are related but distinct requirements, and most CDP architectures handle them with varying degrees of success.
Streaming event ingestion—capturing a page view, an add-to-cart, or a support ticket the moment it happens—is table stakes for any modern CDP. The harder problem is what happens next. Does that event update a unified customer profile? Does that profile update trigger a downstream audience or journey without a batch delay? Can the system make a real-time decision about which message or offer is appropriate, drawing on the full history of that customer?
The answer is often: sometimes, partially, or only if you configure it just right. That gap is where real-time CDP promises break down in practice.
It's also worth distinguishing between latency and freshness. Latency describes the delay between an event and a system response. Freshness describes how current the underlying profile data is when that response fires. A CDP can have low latency on trigger logic but stale profile data if its identity resolution or data sync runs on a nightly batch. For real-time use cases, both need to be tight.
The Architecture Problem Most Vendors Don't Discuss
Many traditional CDPs—Segment, mParticle, and earlier versions of Salesforce Data Cloud among them—were designed to ingest and store customer data inside their own proprietary systems. This made sense when data infrastructure was fragmented, but it creates meaningful problems today.
First, there's the data copy problem. When a CDP pulls data from your warehouse, CRM, and event streams into its own store, you now have multiple versions of truth. Reconciling those versions is time-consuming and error-prone, particularly when the CDP's profile doesn't reflect updates that happened in your source systems since the last sync.
Second, there's the compute cost of real-time at scale. Streaming pipelines are expensive. When a vendor runs those pipelines inside their own infrastructure and charges per event or per profile, costs can grow faster than the business value.
Third—and most critically for real-time use cases—there's the question of data completeness. If your CDP only knows what it has ingested, its real-time decisions are only as good as its ingestion coverage. A customer's purchase history sitting in your data warehouse, or a support interaction stored in Zendesk, may never surface in the CDP's profile in time to influence the next interaction.
The best CDPs for real-time use cases solve this by keeping customer data in the warehouse and computing profiles there, rather than requiring a separate proprietary store.
Key Capabilities to Evaluate
Streaming Ingestion and Profile Latency
Look for CDPs that support native streaming ingestion via Kafka, Kinesis, or Pub/Sub integrations, and that update unified profiles in near-real-time rather than on a scheduled batch. The profile update latency should be measured in seconds to low minutes for behavioral events, not hours.
Ask vendors specifically: how long after a behavioral event fires does that event influence downstream audience membership? If the answer involves a nightly batch or a manual refresh, the real-time claim needs scrutiny.
Identity Resolution at Event Speed
Identity resolution is the process of stitching together anonymous and known identifiers into a single customer profile. For real-time use cases, this has to happen fast. If a user completes a form on your website and then immediately receives a push notification, the CDP must have already resolved that session to their profile and updated downstream systems.Some platforms resolve identity in batch, meaning a recently identified user won't be recognized until the next resolution run. This can introduce delays of hours or more, which makes real-time personalization during an active session impossible.
Trigger-Based Activation
Real-time use cases are mostly trigger-based: a user abandons a cart, completes an onboarding step, crosses a spending threshold, or exhibits a churn signal. The CDP should be able to define these conditions, evaluate them continuously, and activate downstream actions—emails, push, paid audience updates, CRM writes—without manual intervention.
This is where many CDPs require a separate orchestration tool or marketing automation layer, which adds latency and operational complexity. The strongest platforms handle trigger evaluation and multi-channel activation natively.
Decision Logic That Uses Full Profile Context
Low-latency triggers are only valuable if the action they fire is the right one. A trigger that fires an abandoned cart email regardless of whether the customer already purchased through another channel, or is currently in the middle of a support escalation, creates bad experiences.
The best CDPs for real-time use cases evaluate trigger conditions against full customer profiles, including suppression logic, channel history, and current journey state. This requires that the profile be complete and current—which brings the architecture question back to the center.
Where Composable CDPs Outperform Packaged Alternatives
A composable CDP approaches the architecture problem differently. Rather than ingesting customer data into a proprietary store, it works directly against the data already in your warehouse—whether that's Snowflake, BigQuery, Databricks, or Redshift. Profiles are computed where the data lives. Activation pulls from those warehouse-native profiles.
This architecture has a specific advantage for real-time use cases: completeness. Because the composable CDP reads from the same warehouse that receives your transactional, behavioral, and third-party data, profiles reflect the full picture, not just what's been forwarded to a separate system. When your data warehouse is the system of record, there's no lag between a data update and its availability for real-time decisions.
It also means streaming pipelines write directly to the warehouse, and the CDP evaluates conditions against warehouse tables that are updated continuously. The result is trigger-based activation that draws on current, complete data—not a subset of it.
For companies already running strong data infrastructure, this is a meaningful architectural advantage over platforms that require a parallel data store.
What to Look for When Comparing CDPs for Real-Time Work
When you're comparing vendors specifically for real-time use cases, here are the questions worth asking:
Profile update frequency: Is identity resolution and profile unification run in real-time, near-real-time, or batch? What's the SLA? Trigger evaluation latency: How quickly after a qualifying event does a downstream action fire? Is this configurable? Data completeness: Does the platform require you to forward all relevant data to a proprietary store, or does it work against your existing warehouse? What data is excluded by default? Suppression and conflict logic: Can the platform suppress real-time triggers based on recent channel interactions, active journeys, or support status? How is this configured? Cost at scale: How does pricing scale with event volume? Are there per-event charges that make high-frequency streaming cost-prohibitive? AI-assisted decisions: Does the platform support automated decisioning for choosing the best action in real-time, or does it only support rule-based triggers?These questions surface the difference between platforms that market real-time capability and platforms that have actually built their architecture around it.
One Approach Worth Examining
Hightouch's Composable CDP is designed specifically around the architectural principles that real-time use cases require. It operates zero-copy against the customer's existing warehouse, which means profile data is always current and always complete—there's no secondary store to sync.Identity resolution runs as part of the Composable CDP layer, stitching known and anonymous identifiers into unified profiles that update as new data arrives. Downstream activation—to email platforms, push providers, ad networks, CRMs, and more—can fire based on streaming triggers evaluated against those warehouse-native profiles.
For teams managing complex real-time journeys, the Agentic Marketing Platform adds an orchestration and decisioning layer on top. The Lifecycle Marketing Studio within the AMP includes AI Decisioning, which evaluates customer context at the moment of trigger and selects the best next action across channels. This isn't static rules-based logic—it draws on the full profile to make dynamic decisions.
The practical difference shows up in scenarios like: a customer who abandons a cart but also has an open support ticket. A rules-based trigger would fire the cart recovery email regardless. AI Decisioning suppresses it and routes to a service-first response instead. This kind of contextual decision requires both low latency and complete profile data—exactly what the composable architecture provides.
Hightouch also supports marketer-facing tools like Customer Studio for audience building and Hightouch Lifecycle Marketing Studio for channel orchestration, so real-time capability doesn't require deep technical configuration for every campaign.
The Vendors Worth Knowing
Segment (now part of Twilio) remains widely used for event ingestion and basic profile building, but its real-time activation capabilities depend heavily on integrations with downstream tools. It works well for teams with strong engineering resources who want flexibility, but the platform itself doesn't natively solve real-time decisioning or suppress based on full profile context.
ActionIQ targets enterprise buyers and has invested in real-time trigger infrastructure, though it maintains a proprietary data store model that can introduce the freshness problem described earlier. It's a reasonable option for large organizations with established data governance workflows.
For teams that have invested in a modern data warehouse and want real-time activation without duplicating data, a composable approach—like Hightouch's—is structurally better suited to the use case.
The Real-Time Gap Is Mostly Architectural
Most CDP vendors have added real-time marketing language to their positioning over the past three years. But the architecture underneath hasn't changed as quickly as the positioning. Platforms that were designed for batch ETL workflows, or that store customer data in proprietary systems with periodic sync cycles, face inherent limits on how fast and how completely they can operate.
The best CDPs for real-time use cases are the ones that treat data completeness and profile freshness as architectural requirements, not optional features. That means either investing heavily in streaming infrastructure within a proprietary store—or rethinking the architecture entirely to work against the warehouse where data already lives.
For companies evaluating CDPs today, the architecture question is the most important one to ask early. Speed of activation is only valuable when the data driving that activation is accurate and complete. Getting that foundation right makes everything downstream—triggers, personalization, suppression, decisioning—more reliable and more effective.
Summary
Real-time CDP capability depends on three things: low profile update latency, complete data coverage, and trigger logic that draws on full customer context. Traditional CDP architectures often compromise one or more of these. Composable CDPs that operate zero-copy against the warehouse address the data completeness problem structurally. Platforms with integrated identity resolution and AI-assisted decisioning go further, enabling contextual real-time actions rather than simple rule-based triggers. When comparing options, ask about architecture before evaluating features—the answer will determine everything else.