Analytics Careers in the AI Era: What Actually Survives

The analytics role that data practitioners signed up for five years ago is largely gone. Not disrupted — gone. The question worth sitting with isn’t whether AI ate your job description, but whether what replaced it is actually more valuable.

Towards Data Science contributor Rashi Desai puts it plainly: the transformation isn’t a threat to navigate around, it’s a restructuring to lean into. For marketing data teams across Southeast Asia — many of whom are already stretched thin across Shopee feeds, LINE CRM workflows, and multi-market attribution models — that framing matters.

The Query Is Cheap. The Question Is Not.

For most of the last decade, analytics value was measured in delivery speed: how fast can you pull the cohort, run the funnel, slice the CAC by channel? AI has commoditised that layer almost completely. A well-prompted LLM can write the SQL. A capable agent can schedule the report.

What it cannot do — at least not reliably yet — is determine whether you’re asking the right question in the first place. Desai’s argument is essentially this: the analyst who survives is the one who owns the hypothesis, not just the output. In CDP terms, that means shifting from running segments to designing the segmentation logic that reflects actual customer behaviour — understanding why a Grab-loyal customer in Jakarta behaves differently post-Ramadan than a Lazada-first buyer in Manila, and encoding that distinction into your activation framework.

This is where senior data practitioners genuinely earn their seat at the strategy table, not in the BI tool.

AI Agents Are Running in Production. Who’s Watching Them?

The second thread worth pulling is less about career positioning and more about infrastructure integrity. Databricks’ Agent Bricks writes trace data directly to Delta tables in an open, portable format — which sounds like a clean engineering win until you ask what happens when an agent misfires at production scale.

Monte Carlo’s integration with Agent Bricks addresses exactly this gap: monitoring agent telemetry in the same observability layer you’d apply to any critical data pipeline. For teams building AI-powered activation on top of a CDP — personalisation agents, next-best-action models, churn propensity triggers — this matters enormously. An unmonitored agent running bad logic against your unified customer profile doesn’t just produce wrong outputs. It erodes the trust your business stakeholders placed in the platform when they approved the licence fee.

The practical implication: before you scale any agentic workflow into your customer data stack, build the observability layer first. Traces in Delta tables are only useful if someone — or something — is reading them with intent.

---

The Southeast Asia Complication

Neither of these dynamics plays out in a vacuum, and in Southeast Asia the stakes are higher than in more consolidated markets. A typical mid-market brand operating across three SEA countries is managing data from at least four platform ecosystems (Shopee, Lazada, Grab, LINE), two or three languages, and regulatory environments that treat data residency very differently — Singapore’s PDPA versus Thailand’s PDPA versus Indonesia’s PDP Law are not interchangeable.

For analytics teams in this context, the shift toward question-ownership rather than query-execution is urgent. Automated agents can pull cross-platform behavioural data. They cannot yet interpret why conversion on a Thai LINE OA campaign drops 23% the week before Songkran, or how to weight that signal against a user’s Shopee purchase history when building a suppression list. That contextual, culturally-grounded reasoning is the durable human layer.

The brands getting this right — Sea Group and Grab both come to mind in how they’ve structured their data science functions — are the ones who’ve separated execution infrastructure from strategic interpretation. The agents run the pipelines. Humans define the rules and audit the outputs.

What This Means for How You Build Your Data Team

If you’re a marketing director or growth lead reviewing your data team’s mandate for the next 12 months, two things follow from all of this.

First, redefine what seniority means in analytics. The traditional ladder — junior analyst to senior analyst to analytics manager — was built around query complexity and dashboard ownership. That ladder needs rebuilding around hypothesis quality, stakeholder communication, and the ability to translate ambiguous business questions into rigorous data frameworks. These are harder skills to hire for and harder to train, which is exactly why they’re durable.

Second, treat agent observability as a data governance requirement, not an engineering nice-to-have. If your CDP activation layer includes any AI-driven components — and by 2026 it almost certainly does — your data quality monitoring needs to extend to those agents’ decision traces, not just the underlying tables they read from. Monte Carlo’s approach of keeping telemetry within the Databricks ecosystem rather than forwarding it to external tools is worth studying as a model: it keeps your observability portable and auditable.

The unified customer profile is only as trustworthy as the systems interpreting it.

Key Takeaways

• Redefine analytics seniority around hypothesis design and contextual judgment, not query speed — that’s the layer AI hasn’t commoditised.
• Build agent observability into your CDP activation stack before you scale; unmonitored AI logic against customer data erodes platform trust faster than any technical failure.
• In SEA’s multi-platform, multilingual market reality, cultural and contextual reasoning remains a hard human advantage — encode it into your segmentation frameworks, not just your creative briefs.

The deeper provocation here isn’t whether AI will replace analysts — it’s whether most analytics functions have ever been doing the work that actually creates strategic value, or whether they’ve been doing the work that was simply hard to automate at the time. Now that the automation bar has risen, the answer to that question is becoming uncomfortably visible.

---

At grzzly, we work with growth and data teams across Southeast Asia to design CDP architectures that connect behavioural, transactional, and declared data into activation-ready customer profiles — and build the governance layers that keep them trustworthy as AI becomes part of the stack. If your unified customer profile is earning its licence fee or you’re not sure whether it is, we’d like to hear about it. Let’s talk.