AI is raising expectations for customer experience, pushing it toward a model where decisions unfold in real time and journeys shift moment by moment. But behind the scenes, many enterprises are still relying on fragmented systems that struggle to unify customer identities and maintain a reliable view of each interaction.
Decisions that appear instantaneous are often built on stale or incomplete data.
The modern tech stack is a patchwork of customer data platforms (CDPs), customer relationship management systems (CRMs), messaging orchestration tools, contact center platforms and data warehouses, each introduced at different points in time, each with its own latency profile. When a customer moves from app to website to chat agent to phone call, the journey crosses five or six systems before anyone makes a “real-time” decision. At every handoff, data ages, and the decisions downstream age with it.
And yet, budgets continue to flow into decisioning engines, AI layers, and orchestration platforms, frequently without a hard look at whether the underlying data systems can support true real-time execution.
“The biggest challenge is not the data volume, it’s the data readiness,” Hardik Parikh, Chief Revenue Officer and Co-Founder of Shaip, told CX Today in an interview.
“Most customer interaction data in today’s world is fragmented across systems, unstructured, in silos, [which is] very difficult to operationalize for AI in real time.”
The real gap, he argues, is in “turning raw interaction data into structured, compliant, AI-ready data sets that can actually support faster decision-making.”
How Real-Time Is Your CX Data, Really?
One of the core issues is the definition. “Real-time” has become a catch-all term that spans different technical realities.
At one end there’s event streaming, data captured and processed as it happens, often within milliseconds. This is what powers use cases like fraud detection or dynamic pricing in high-frequency environments.
Micro-batch processing, which runs every few seconds or minutes, is more common. Many customer data platforms (CDPs) and analytics platforms operate here, even if they’re marketed as real-time.
And then there’s near-real-time dashboards or systems that update every few minutes, sometimes longer. That’s often sufficient for reporting and monitoring, but it can introduce meaningful lag to decision making.
These approaches are often conflated. A vendor might stream events into a platform in real time, but if identity resolution runs every five minutes and activation pipelines update every 10 minutes, the end-to-end system is anything but instantaneous.
Where Latency Creeps In
Even in modern tech stacks, latency isn’t caused by a single bottleneck but builds across multiple layers, each introducing its own delays.
It often starts with data collection. Web and app tagging relies on software development kits (SDKs) that buffer events, retry failed sends, and sometimes drop data altogether. Network conditions, device constraints, and ad blockers all play a role.
Enterprise analytics systems typically capture roughly 85–95% of expected events, implying that 5–15% of interactions may be lost, particularly under complex conditions or peak traffic. That data is simply gone and no downstream processing can recover it.
From there, data moves into pipelines that transform and load it into warehouses or lakes. Scheduled batch jobs might run continuously, or on schedules measured in minutes. Either way, they introduce a gap between event capture and availability.
Identity stitching adds another layer. Matching anonymous and known users, resolving duplicates, and linking devices requires computation and, often, batch processing. Even advanced systems rarely perform this perfectly in real time. Match rates for cross-device identity resolution range from around 40% to 70%, depending on the richness of consented identifiers available, data shows.
Then pushing enriched data back out to engagement platforms, whether email, push notifications, contact center routing, or ad systems, adds further lag as each has its own ingestion limits and refresh cycles.
By the time a “real-time” decision is executed, the underlying data may already be several minutes old, at least.
What’s Broken in Customer Identity Resolution?
Identity is the other, arguably more complex, side of the coin. Modern customer experience depends on stitching together interactions across devices, channels, and contexts. A single customer might browse anonymously on mobile, log in on desktop, call a contact center, and respond to a marketing message, all within a short window.
Connecting those touchpoints into a unified profile is hard enough. Doing it in real time is harder still.
Industry benchmarks put cross-device identity match rates at 40–70%, and they fall further in environments where consent signals are fragmented and third-party cookies are gone.
Parikh added a dimension that enterprises consistently underestimate:
“It’s not just about having a large data set. It’s about ensuring the model is exposed to various scenarios, many different customer situations, edge cases, interactions.”
Models trained on a narrow slice of customer behavior perform confidently, and incorrectly, when they encounter customers who don’t match their training distribution. As Parikh said:
“In many cases, having a broader range of scenarios is even more valuable than simply increasing the volume of the data.”
The contact center adds another complication. Identity matching often depends on IVR inputs, agent workflows, or CRM lookups, which can lag behind digital interactions. A customer who just completed a transaction online may still appear as “unknown” when they call.
All of this means that the “single customer view” is often fragmented in practice. These gaps have real consequences for customer experience and operations.
Mis-personalization is one of the most visible. A customer abandons a cart, completes the purchase minutes later, and still receives a reminder email. Or they’re shown an offer for a product they’ve already bought. These moments erode trust, even if they seem minor.
Consent drift is more serious. Regulations require that preferences be respected across channels, but those preferences are often stored and updated in separate systems. If consent signals don’t propagate quickly enough, brands risk contacting customers who have opted out, or failing to interact using their preferred channels. In regulated industries, that can carry legal and financial risk.
Operational misrouting is another issue. In contact centers, decisions about routing or prioritization often depend on customer value, recent activity, or intent signals. If those inputs are stale or incomplete, high-value customers may be misclassified, and service levels can suffer.
When multiple users share an account, identity resolution and personalization break down further. The emergence of AI agents acting on behalf of customers is pushing this problem to a new level.
Mary Ann Miller, VP Evangelist & Fraud Executive Advisor at Prove, argues that trust frameworks built around human-to-human or human-to-system interactions are no longer fit for purpose.
“Our trust frameworks really need to look at things from a different perspective — with new fresh eyes, new governance and new data,” she told CX Today.
“New ways that we determine who’s at the other end of an interaction. It’s understanding if it’s a human or not, understanding if it’s an agent and if that agent is authorized.”
How can enterprises approach these challenges effectively?
How Do Leading Organizations Deliver Real-Time CX?
Some teams are starting to close the gap by addressing the foundations rather than chasing faster decisioning layers. Parikh said:
“In the organizations that are successful versus not, we have seen one common trait: they treat data as infrastructure, as a core part of their AI deployment strategy.”
It begins with data discipline. Teams establish formal data contracts that define structure, quality, and timeliness upfront. Expectations around schema, freshness service-level agreements (SLAs), and acceptable error rates are documented and enforced, rather than assumed. This creates a shared understanding between systems and reduces the silent failures that can undermine downstream decisions.
From there, visibility and precision are key. Mature teams invest in observability, continuously tracking data freshness, completeness, and identity match rates, to detect anomalies before they propagate into bad decisions rather than after they’ve affected customers. They also get disciplined about latency by use case. A fraud check may demand sub-second responsiveness, whereas a marketing journey can tolerate delay. Aligning infrastructure to those realities avoids both over-engineering and underperformance.
Identity strategies are evolving in parallel. Instead of chasing a perfect, always-on unified profile, teams prioritize context using the best available signals at the moment of interaction and improving resolution over time. This reduces the risk of decisions being made on mismatched or incomplete profiles, which will only increase with agentic AI adoption. As Miller put it:
“It’s great that it’s going to be a convenience for the customer. But it’s only going to be a convenience for the customer if it’s safe for the customer.”
Finally, there’s a shift toward accountability. Decision auditability becomes a core capability: teams log not just what action was taken, but what data informed it, how fresh that data was, and what logic was applied. When something goes wrong, they can trace it, fix it, and prevent it from repeating.
The investment required to build this kind of foundation is significant. But so is the cost of ignoring it—misrouted interactions, failed personalization, compliance exposure, and systems that act quickly but incorrectly at scale.
For buyers, that means asking harder questions about where latency exists, how identity is resolved, and whether the data that feeds into decision systems is fit for purpose. For vendors, it requires greater precision in how capabilities are described, and more transparency around the trade-offs behind “real-time” claims. And for CX leaders, it reinforces that even the most advanced decision-making systems are only as reliable as the data beneath them.
