Turning “What happened?” into “Why did it happen?” with Metabase AI

While experimenting with Metabase’s Metabot for the #MetabaseAI Hackathon, we noticed something interesting:

Metabot is great at answering: “What happened?”

But like most BI tools, it stops short of: “Why did this happen?”

So we tried something different. Instead of expecting Metabot to magically infer the answer, we designed a structured interaction workflow that guides it step-by-step toward root cause analysis.

What this looks like in practice:

Ask in Slack: “Why did orders drop this week?”

Metabot initially responds with trends, charts, possible hypotheses

Then we guide it:

2. Identify which segment/category changed most

3. Compare week-over-week and which regions do the orders correspond to

4. Summarize the key driver

5. Generate a decision-ready summary

What changed?

We didn’t build a new model.

We leveraged the Metabase semantic layer, used Metabot for data access and added structured prompts to guide reasoning.

Architecture

Slack (Metabot)

Key takeaway

AI doesn’t need to “know everything”. Sometimes the real value comes from designing how humans + AI collaborate to reach answers

Instead of: “Here’s a chart”

We get: “Here’s what changed, where it changed, and what to check next”