Welcome everyone. Today we want to take a step back, look at where we are with the data stack and especially the marketing attribution topic, and map out a clear path forward - one that not only solves the immediate Klar challenge but sets Juit up for AI-powered analytics.

Quick overview of what we'll cover. We'll start with an honest assessment of where things stand, then walk through the options, share our recommendation, and end with concrete next steps.

Let's start with the current state - the foundation is solid, but we have one piece that's not working.

The foundation is genuinely solid. BigQuery, dbt, Airflow - this is a modern, well-architected stack. The issue isn't the architecture; it's the marketing attribution layer, specifically Klar.

I want to be direct here. We've spent a year trying to make Klar work, and the honest assessment is: the problems are structural. Klar is a great tool for Shopify-native businesses, but Juit's architecture just doesn't fit. The numbers don't align with Looker, the support is limited, and we can't see inside the black box. Spending more time won't change any of these fundamentals.

So what do we do? Let me lay out the three options as we discussed in the last review, along with a clear evaluation.

These are the three options we identified in the last review. Option 1 means accepting Klar as-is and living with two disconnected worlds. Option 2 is more of the same. Option 3 is a clean break - build it ourselves on the foundation that already exists.

When you put the options side by side, the picture is clear. Option 1 is easy but you lose alignment and AI readiness. Option 2 is more of what hasn't worked. Option 3 requires some effort, but you already have 80% of the attribution models built in BigQuery - and it's the only option that gives you full data ownership and a foundation for AI.

Our clear recommendation is Option 3. Let me show you why it's not just the best option for marketing attribution - it's the path to a modern, AI-ready analytics architecture.

This is the core message. You've already invested in the right foundation. BigQuery, dbt, Airflow - that's a strong stack. And you're 80% there on the attribution models. The remaining 20% plus a proper semantic layer gives you something Klar never could: full ownership, full transparency, and a platform ready for AI.

Let me be concrete. You're not starting from zero - those attribution models are 80% there. Full ownership means no more black box, no more misaligned numbers. And here's the strategic angle: centrally structured, well-documented data in BigQuery is exactly the prerequisite for any AI use case - conversational analytics, marketing mix modeling, you name it. Klar locks your data away; in-house opens it up.

Here's where we're headed. Same solid ingestion layer. BigQuery stays the warehouse. The big change is the dbt semantic layer - this becomes the single definition of every metric, every dimension, every attribution rule. Then three consumption layers: Lightdash for dashboards, a conversational AI layer for self-service, and eventually marketing mix modeling. Everything reads from the same semantic layer - one truth, many interfaces.

Let me walk you through how we get there in three phases. Each phase delivers standalone value - you don't need to commit to all three upfront.

Three phases, each self-contained. Phase 1 is the critical one - attribution models and semantic layer. Phase 2 replaces Looker with Lightdash. Phase 3 adds the AI layer. And in the future, once you have clean attribution data, marketing mix modeling becomes feasible. The point is: each phase delivers value on its own. You can pause after any phase and still be better off.

Phase 1 is the foundation. We finish what's already 80% done, consolidate all marketing data sources, and build a proper semantic layer in dbt. This gives you one place where every metric is defined, documented, and tested. Once the numbers are validated, we turn off Klar.

Phase 2 is about the BI layer. Looker is powerful but expensive, and it runs a separate semantic model - LookML - that duplicates what dbt already does. Lightdash is built specifically for dbt. It reads your metrics and dimensions directly, so there's no drift between your transformation layer and your dashboards. Self-hosted means zero license cost and full data sovereignty.

Phase 3 is where it gets exciting. Once you have a clean semantic layer, conversational analytics becomes straightforward - either through Lightdash Cloud's built-in AI features or a custom solution. But the bigger picture is AI readiness. Marketing mix modeling needs clean, structured attribution data. Any LLM-based analytics tool needs well-documented metrics and dimensions. The work in Phases 1 and 2 directly enables everything here. You're not doing extra work for AI - you're doing the analytics work properly, and AI readiness comes as a natural consequence.

This is where we switch to a live demo of Lightdash. Show how a chart is created from dbt metrics, how the code-first approach works, and how fast it is to go from question to visualization. The key message: this is dramatically faster than Looker for teams that already use dbt.

Let's talk about what needs to happen next.

Four concrete next steps. The most important one is step 1 - do we agree to go with Option 3? If yes, we can start scoping immediately. The Lightdash evaluation can run in parallel and doesn't block the attribution work.

Over to you. What questions do you have? The main decision we need today is: do we go with Option 3? Everything else follows from that.