When AI Writes the Queries, Who Owns the Insight?

The analytics role that existed in 2021 — the one where you wrote SQL, built dashboards, and handed findings to a marketing director who nodded and moved on — is functionally obsolete. Not because the work disappeared, but because AI can now do the mechanical parts faster and cheaper than any analyst hired at market rate.

The question worth sitting with isn’t will AI replace analytics? It’s: what does the analytics function need to become before the answer stops being interesting?

The Shift From Query-Writer to Context-Setter

Writing in Towards Data Science, analyst Rashi Desai puts it plainly: the analytics career she signed up for five years ago no longer exists — and she’s made peace with that. The framing that resonates isn’t survival anxiety; it’s deliberate repositioning. The analysts who are pulling ahead aren’t defending their SQL turf. They’re becoming the people who decide what questions matter, which data is trustworthy enough to act on, and what the output should actually drive.

In customer engagement terms, this maps directly to CEP work. A journey architect who can only read dashboards is a passenger. The ones who shape the logic — what triggers matter, which signals indicate intent versus noise, how context changes the right action — those roles don’t compress easily into a prompt. That judgment layer is where the defensible value sits.

For teams running platforms like Braze, Insider, or MoEngage across Southeast Asian markets, the implication is concrete: your analytics function should be spending less time producing reports and more time defining the rules by which automated systems interpret customer behaviour.

Trustworthy AI Output Isn’t a Given — It’s an Engineering Discipline

There’s a parallel challenge emerging on the infrastructure side that engagement teams need to understand, even if they don’t own it. As brands increasingly deploy AI-generated content, recommendations, and responses — whether through chatbots on LINE, personalised push logic, or AI-assisted search on Shopee storefronts — the reliability of those outputs is not guaranteed by default.

Priyansh Bhardwaj’s practical breakdown in Towards Data Science on production RAG (Retrieval-Augmented Generation) systems makes a point that applies well beyond engineering: hallucinations and retrieval failures in AI systems don’t announce themselves. They degrade quietly, eroding trust with customers before any internal team notices the pattern. The fix isn’t a one-time audit — it’s continuous evaluation baked into the production workflow.

For engagement teams, this translates to a governance requirement: if AI is personalising your comms or powering your support flows, you need defined checkpoints for output quality, not just input quality. The brands in Southeast Asia that will get hurt are the ones treating AI deployment as a launch event rather than an ongoing operational discipline.

---

The Open Infrastructure Bet Worth Watching

On a longer horizon, the tooling landscape itself is shifting in ways that will reshape how data teams build and share analytics infrastructure. The dbt team recently announced that OSI — their open semantic interface — is now moving through Apache Software Foundation incubation as Apache Ossie. It’s early-stage, but the directional signal matters: the industry is moving toward open, interoperable standards for how data assets are defined, shared, and governed across organisations.

For Southeast Asian brands operating across multiple markets, multiple data environments, and multiple agency partners simultaneously, open standards for semantic layers could meaningfully reduce the integration tax. Right now, a brand running campaigns across Vietnam, Thailand, and the Philippines is often managing three partially-overlapping data definitions of what a “loyal customer” means. Federated semantic layers — if the Apache Ossie model gains traction — could make that reconciliation less painful and more portable.

It’s not a 2026 problem to solve. But data architects designing CEP frameworks today should be watching where the standardisation momentum is going, because the platforms you lock into now will either play well with that future or create expensive migration headaches.

What the Analytics Function Actually Needs to Own

Pull these threads together and a pattern emerges. The analytics and engagement functions that will remain strategically valuable — rather than becoming prompt-curation layers on top of AI tools — share three characteristics:

First, they’ve moved upstream. They’re involved in defining what data gets collected and why, not just what gets reported. In a CEP context, that means owning the event taxonomy, the identity resolution logic, and the criteria that determine when a signal is reliable enough to trigger an automated action.

Second, they treat AI output as a hypothesis, not a conclusion. The brands running production AI at scale without continuous evaluation loops are accumulating trust debt with their customers. A single hallucinated product recommendation on a high-intent journey can cost more in churn than the efficiency gain is worth.

Third, they’re building transferable infrastructure. Proprietary data models that only work inside one platform are a liability as the tooling landscape consolidates and open standards mature. The teams investing in portable semantic definitions now will have more options later.

The analytics career isn’t getting eaten. It’s getting harder to coast in — which, for the people who wanted the intellectual challenge in the first place, is probably fine.

---

Key Takeaways

• Shift your analytics function from query execution to context definition — that’s the layer AI can’t easily compress.
• Treat AI-generated engagement outputs as requiring continuous evaluation, not one-time QA; quiet degradation is the real risk.
• Watch the Apache Ossie incubation as an early signal of where open semantic layer standards are heading — and design your data architecture with portability in mind.

The harder question sitting underneath all of this: if AI can now generate the insight, who in your organisation is responsible for deciding whether the insight is worth acting on — and do they currently have enough context to answer that well?

---

At grzzly, we work with marketing and data teams across Southeast Asia to design CEP frameworks where the human judgment layer is deliberately positioned — not accidentally inherited. If your analytics function is at an inflection point, or your AI-powered engagement is running without proper evaluation guardrails, we’d enjoy the conversation. Let’s talk