Key takeaways from Debug'Em All Meetup

Yesterday I attended the Debug’Em All Meetup – Zurich 2026 (Analytics Debugger x Digitec Galaxus) at Pause im Foifi. Sharing a short recap of the most interesting takeaways from the talks (useful for anyone working on tracking, data quality, analytics engineering, or experimentation).

1) Building a robust data collection setup (Digitec Galaxus – David Hermann)

The core message: treat tracking like a production system — with clear quality principles, ownership, and maintainability as first-class requirements.

The “7 pillars” they use to evaluate data collection quality:

Relevance (only collect what is needed)
Richness (enough context/parameters to answer business questions)
Coverage (consistent tracking across surfaces and journeys)
Accuracy (correctness of event values and semantics)
Data latency (freshness / speed to availability)
Ease of consumption (analysts can use it without friction)
Traceability / lineage (know where it came from and how it changed)

Interesting (and controversial) design choice: They said they use a fully hard-coded implementation for Snowplow, i.e. no Tag Management System (no GTM/Tealium) — the tradeoff is less flexibility but more control, versioning, testing, and engineering discipline.

2) Introduction to Malloy (Marcus Stade)

Malloy was presented as an open-source modern analytics language that sits between raw SQL and BI tools.

Why it’s interesting (especially for web/app analytics):

helps create a semantic layer: turning technical tables/fields into business concepts (e.g., dimensions/measures with definitions)
promotes reusability (avoid repeating SQL logic / KPI definitions everywhere)
reduces analysis mistakes by keeping consistent naming + metric logic
positions itself as a way to improve “data democratization”: analysts/managers can consume metrics without needing to understand physical schemas

They showed examples in Google Cloud Shell / notebooks, defining GA4-style models where you standardize common concepts like sessions, pageviews, unique users, etc.

3) Automating A/B Tests at scale (Digitec Galaxus – Clara Goebel)

They built an internal end-to-end experimentation framework to scale experimentation to 100+ A/B tests/year with less manual work and more standardization.

Their framework includes:

Implementation
Data collection pipeline
Monitoring dashboard
Evaluation script
Documentation

Main outcomes they reported:

Speed to insights: evaluation time reduced ~2x
Higher volume of tests
Stronger statistical consistency (less “handwavy” analysis)
Better presentation + documentation
Analysts spend time on tests that are actually worth it

What they want to improve next:

move from fixed A/B tests to adaptive experimentation (e.g., Multi-Armed Bandits, sequential testing)
shift from session-based to visitor/user-based testing for better long-term interpretation
use LLMs to generate experiment summaries, and perform meta-analysis across experiments (patterns, learnings)

4) Open-source server-side automated event data validation (Defused Data – Maciek Stanasiuk)

This was on event QA / data quality assurance, specifically:

server-side
fully automated
validating incoming events in near real-time
positioned as an open-source solution you can deploy quickly

Core point: there are many ways to do data QA (manual testing, scenario tests, warehouse tests), but validating events as they arrive is a powerful layer for catching issues early.