Most brands claim real-time engagement but deliver fragmented journeys. Here's why disparate data kills CEP ambitions — and how to fix it.
Your customer opens the Grab app, taps through to your brand’s in-app storefront, abandons a cart, then walks into your physical outlet an hour later — and your staff has no idea any of that happened. This is not a technology gap. It is a data architecture gap dressed up as a customer experience problem.
The uncomfortable truth is that most brands investing in customer engagement platforms are running real-time interfaces on top of batch-era data logic. The signals are arriving in real time; the decisions are still being made in silos.
The Four-Touchpoint Illusion
Tealium’s Gary Albertson frames the core challenge precisely: a banking customer who moves from app research to website browsing to a call centre query to a branch visit is experiencing one continuous relationship. Inside the institution, those four moments are often owned by four separate systems, teams, and data schemas. The customer sees coherence; the organisation sees four disconnected events.
This is not a banking-specific pathology. Any brand operating across Shopee storefronts, owned web properties, LINE OA broadcasts, and physical retail faces the same structural problem. The touchpoints multiplied faster than the data plumbing connecting them. And when engagement logic is built on top of fragmented identifiers — device IDs that don’t resolve to user profiles, session cookies that expire before the next visit, loyalty IDs that never get joined to CRM records — even the most sophisticated personalisation engine is making educated guesses.
The fix is not another tool. It is agreeing on a canonical customer identity layer before anything else gets built.
AI Orchestration Without Clean Data Is Expensive Noise
Alchemer’s newly launched Iris platform is a useful case study in where the market is heading: an AI-native CX layer that ingests feedback signals, surfaces recommended actions, and pushes work between systems without requiring human routing at every step. The ambition is right. Feedback collected at a touchpoint should trigger a response within that same journey — not land in a dashboard that someone checks on Thursday.
But Iris, like every AI orchestration layer, is only as coherent as the data it reasons over. Feed it unified, timestamped, identity-resolved signals and it can genuinely close the loop between insight and action. Feed it the typical enterprise data swamp — partial CRM exports, delayed event logs, channel-specific customer IDs that never reconcile — and it surfaces confident-sounding recommendations built on a shaky premise.
The fine-tuning versus training debate in ML (Monte Carlo AI frames this clearly: fine-tuning adapts an existing model cheaply; training from scratch is expensive but precise) maps neatly onto CEP strategy. Most brands should be fine-tuning engagement logic on top of robust foundational data — not rebuilding the engagement layer every time a new channel appears. But fine-tuning a model trained on bad data still gives you bad outputs, faster.
What a Functional CEP Architecture Actually Requires
Three components that most roadmaps underinvest in, relative to the glamour spending on AI and creative:
Identity resolution at ingestion, not at query time. Stitching customer IDs together when you are about to send a message is already too late. Grab, Sea Group, and the regional super-apps resolve identity continuously at the infrastructure layer — so by the time a trigger fires, the customer profile is current. For brands not building their own data infrastructure, a CDP with real-time identity stitching (Tealium, Segment, mParticle) deployed before the engagement layer is non-negotiable.
Event schemas that travel across channels. A product view event on your mobile app and a product view event on your desktop site should share a canonical schema — same event name, same property structure, same timestamp format. This sounds obvious. In practice, mobile and web are often instrumented by different teams using different taxonomies, and the reconciliation debt compounds for years.
Suppression and fatigue logic that operates across channels, not within them. This is where most batch-era thinking survives longest. A customer who converted via Shopee notification should be suppressed from the LINE broadcast going out four hours later. That suppression requires cross-channel state — something most campaign tools do not maintain natively. Build this into your CEP orchestration layer explicitly, or you will train your best customers to ignore everything.
The Southeast Asia Complexity Multiplier
The regional context makes all of this harder and more urgent simultaneously. Mobile-first usage in markets like Indonesia, Thailand, and Vietnam means the primary engagement surface is an app — which has stricter data-sharing constraints post-iOS privacy changes and platform-specific notification mechanics. LINE dominates in Thailand; Zalo matters in Vietnam; TikTok Shop is eating into Shopee’s discovery role across multiple markets. Each platform has its own customer identifier, its own event taxonomy, and its own latency profile.
Multilingual audiences add another layer: a suppression rule built on Thai-language keyword triggers will not fire correctly on the same customer’s English-language web session. Journey logic needs to be language-agnostic at the data layer, even if the content served is localised.
Brands that get this right — Grab being the clearest regional example — do so because they treat data infrastructure as a product, with dedicated ownership and ongoing investment, not as a one-time implementation project handed to an SI and forgotten.
Key Takeaways
- Unify customer identity at the infrastructure layer before building any real-time engagement logic on top — otherwise AI orchestration amplifies your data fragmentation, not your personalisation.
- Canonical event schemas shared across mobile, web, and in-store touchpoints are the single highest-leverage technical investment most brands are still not making.
- Cross-channel suppression logic is not a nice-to-have — it is the difference between a CEP that builds relationship equity and one that trains customers to tune you out.
The platforms are not the bottleneck anymore. GPT-6 Astra can reason across complex customer signals in ways that would have required a data science team eighteen months ago. The constraint is the quality and coherence of what you feed it. The question worth sitting with: if you resolved your identity layer tomorrow, how much of your current engagement logic would still hold up — and how much was always just compensating for bad data?
At grzzly, we spend a lot of time with growth and CRM teams who are ready to activate — they have the platform, the creative, the budget — but the data foundation underneath is quietly undermining every campaign. If you are building a CEP or CDPstrategy for Southeast Asian markets and want a second opinion on whether your architecture will actually scale, we are happy to pressure-test it with you. Let’s talk
Sources
Written by
Brooding GrizzlyDesigning CEP frameworks that move beyond batch-and-blast into real-time, context-aware engagement — across channels, devices, and the messiness of actual human behaviour.