Analytics Engineering Levels

Summary

Zach Wilson lays out five levels of analytics engineering maturity in a small, crunchy tweet. The mental model: analytics engineering as a maturity ladder, not a job description — each level represents a fundamentally different relationship between a team and its data.

The five levels:

• L1: Writing queries. SQL, BI tools, light transforms. This is where most teams actually live.
• L2: Modeling analytics data. Dimensional modeling, dbt fundamentals, SCDs, orchestration. The jump from “writing queries” to “building models” is the first real inflection point.
• L3: Engineering the analytics layer. Staging-to-marts layer architecture, CI for data, data quality testing, reverse ETL. This is where AE becomes a proper engineering discipline.
• L4: Operating at scale. Semantic layers, performance optimization, observability, data contracts, compliance. The concerns shift from “does it work” to “does it work reliably at scale with governance.”
• L5: Analytics as platform. Self-serve analytics, metric APIs, embedded analytics. Data is no longer a service team — it is a platform other teams build on.

The punchline: “Most teams think they are at Level 3. Most are actually stuck between Level 1 and Level 2.”

This maps directly to 06-reference/2026-04-03-data-maturity-processes-tools — Wilson’s levels are a more granular view of the analytics engineering dimension within broader data maturity frameworks. L1-L2 roughly corresponds to “Data Informed,” L3-L4 to “Data Driven,” and L5 to “Data Led” in the framework from 06-reference/2026-04-03-scaling-data-informed-driven-led.

For 01-projects/phdata/index, this is a diagnostic tool for client engagements. Most enterprise clients self-assess at L3 but are actually L1-L2 — the consulting value is in honestly assessing where they are and building the roadmap to the next level. This connects to 01-projects/phdata/career-transition as a framing for the kind of work that differentiates senior data consulting from staff augmentation.

The “self-deception gap” (thinking you are L3 when you are L1-L2) echoes 06-reference/2026-04-03-uber-data-culture-first-principles — Uber’s data culture worked because they were honest about foundations before chasing sophistication.

See also 06-reference/concepts/analytics-as-craft — Wilson’s ladder implies that analytics engineering is a craft with clear progression stages, not just a set of tools to learn.

Open Questions

• What is the minimum team size to meaningfully operate at each level? Can a solo analytics engineer reach L3, or does L3 inherently require a team?
• Where does the AI coding assistant (Claude, Cursor, etc.) intervene on this ladder? Does it compress L1-L2 into a single step, or does it create a false sense of being at L3?
• Is L5 actually achievable, or is it an asymptote? The “analytics as platform” vision has been promised for a decade — what is different now?
• How does this framework apply to Ray Data Co’s own data stack? Where are we on the ladder for our internal analytics?