Why context-aware customer engagement beats personalisation at scale — and what AI architecture debates reveal about building smarter CEP frameworks in Southeast Asia.
There is a quiet crisis running through most customer engagement platforms: brands have more data than ever, and they are using almost none of it well. The problem is not volume. It is context — specifically, what signal is actually fed into a decision at the moment it matters.
The Duchamp Problem in Customer Data
Tealium’s Nick Albertini recently framed an uncomfortable truth through an unlikely lens. When Marcel Duchamp rotated a urinal ninety degrees and called it Fountain, the object did not change — its context did. The meaning was entirely constructed by placement, framing, and the moment of encounter. The parallel to customer data is sharper than it sounds.
A purchase event means something entirely different at 7am on a Monday than at 11pm during a Shopee 11.11 flash sale. A cart abandonment from a first-time visitor is not the same signal as one from a loyalty member who has bought from you eleven times. Same data type, completely different context — and most CEP configurations treat them identically, triggering the same re-engagement sequence regardless.
The fix is not a new data source. It is upstream architecture: ensuring that the context window fed into your engagement logic is rich, current, and shaped by the right signals — session depth, recency, channel of entry, device, and behavioural sequence — not just the transactional event alone.
RAG vs MCP Is Actually a CEP Problem in Disguise
The AI infrastructure debate currently consuming data teams — RAG versus MCP — is, at its core, a question that marketers have been wrestling with for years: how do you get the right information to a decision system at the right moment?
Monte Carlo Data’s Lior Gavish cuts through the noise cleanly: RAG (Retrieval-Augmented Generation) pulls relevant documents to an LLM at query time. MCP (Model Context Protocol) standardises how AI models connect to external tools and data sources. They are not competitors. They solve different layers of the same problem.
For CEP teams, this maps directly. Your segmentation logic is retrieval — pulling the right behavioural history to inform a decision. Your channel connectors (LINE OA, Grab notifications, Lazada messaging) are the protocol layer — the standardised interface between your engagement engine and the point of delivery. Conflating the two leads to the same architectural mistake in both worlds: patching a context problem with a connectivity solution, or vice versa. Getting clarity on which layer is broken is the first diagnostic step before any re-platforming conversation.
Why CEP Frameworks Break at the Strategy Layer
The Boathouse Group’s Fifth Annual CEO Study, covered by CustomerThink, found that CMOs score well on cross-functional alignment and financial literacy — but face more scrutiny on strategy and growth outcomes. That gap is telling. It is easy to align on channel KPIs and brand metrics. It is harder to demonstrate that your engagement architecture is actually driving revenue decisions, not just optimising open rates.
This is where most CEP frameworks quietly fail. They are operationally coherent — sequences fire, suppression lists work, reporting dashboards update — but strategically shallow. The decision logic is built around what is easy to measure, not what drives the behaviour that matters. In Southeast Asian markets especially, where a customer might move across Shopee, a brand’s own app, a LINE chatbot, and an in-store QR code within a single purchase journey, batch-based segmentation does not approximate real-time behaviour closely enough to drive meaningful lift.
The architecture fix requires honest answers to three questions: How stale is your segment membership when a message fires? How many contextual signals are actually available at decision time versus logged after the fact? And which channels in your stack can accept dynamic, real-time attributes versus requiring pre-computed segments?
Building for Rare Events, Not Just Common Ones
There is one more lens worth bringing to CEP design, borrowed from an unexpected domain. Research published by Towards Data Science on using transformer models to forecast rare solar flares highlights a persistent ML challenge: standard models trained on frequent events perform poorly when rare but high-impact events occur. The fix requires architectures that explicitly account for class imbalance and temporal sequences — not just pattern frequency.
The same principle applies to customer engagement. Most CEP logic is calibrated around the median customer behaviour — the repeat buyer, the average session length, the predictable churn signal. But the highest-value interventions often sit in the tail: the loyal customer showing the first signs of disengagement, the new user who skips onboarding but makes a large first purchase, the dormant segment that reactivates during a specific seasonal trigger.
Building engagement logic that can detect and respond to low-frequency, high-value signals requires deliberate instrumentation: explicit event schemas for edge cases, suppression overrides for high-intent moments, and holdout testing that measures lift on rare cohorts specifically — not just overall campaign performance.
Key Takeaways
- Audit what context is actually fed into your CEP decision engine at trigger time — session recency, device, channel entry point, and behavioural sequence should be available, not just transaction history.
- Separate your retrieval logic (segmentation and behavioural history) from your protocol layer (channel connectors and delivery APIs) — conflating them is the root cause of most re-platforming failures.
- Instrument your engagement architecture specifically for low-frequency, high-value customer moments — not just the median behaviour your reporting dashboards optimise toward.
The brands that will pull ahead in Southeast Asia’s next phase of digital growth are not the ones with the most data or the most channels — they are the ones that have genuinely solved the context problem: getting the right signal, correctly interpreted, to the right decision point in near real-time. The provocation worth sitting with: if your engagement platform went down tomorrow, would your customers notice the difference in experience — or just in your internal metrics?
At grzzly, we spend a lot of time inside exactly this challenge — helping brands across Southeast Asia move from channel-centric execution to context-aware engagement architecture. Whether you’re auditing an existing CEP setup or building the case for a more sophisticated approach, we’re happy to think it through 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.