Baachee platform · Data products
Marketing reports composed in conversation, from funnels to attribution
Ten composable skills run on the tracked BigQuery events, so a question like “compare the demo funnel for organic and paid traffic” becomes one saved report.
- Client
- Reporting
- Industry
- Marketing analytics
- Region
- Netherlands
- Period
- Deployed 2026
- composable reporting skills
- 10
- time to insight (preliminary)
- hours → s
Challenge
- Dashboards answer the questions someone thought of last quarter.
- Each new cut of the data waits for an analyst, and results end up scattered across tools.
Approach
- 01Reporting capabilities ship as independent skills: funnel analysis, attribution, cohort tracking, event exploration, anomaly detection, natural-language BI, report generation, traffic source intelligence, user journey mapping, and memory and learning.
- 02Skills compose in conversation; the agent calls several tools, joins the results and presents one comparison.
- 03Reports are saved with the SQL and the result behind them.
- 04Deployment needs only the standard BigQuery event schema, a GTM server container and credentials. No model fine-tuning and no custom ML pipeline.
Architecture
01 · Question
- Plain language
02 · Skills
- Funnel
- Attribution
- Cohort
- Journey
- Anomaly
03 · Data
- BigQuery events
- MCP tools
04 · Report
- Table and chart
- Saved with SQL
In production
- Data source
- server-side tracked events in BigQuery
- Query safety
- SELECT-only
- Model setup
- no fine-tuning; schema discovered at runtime
- Output
- saved reports with SQL and results
Results
| Metric | Before | After |
|---|---|---|
| Time to insight(preliminary, working paper v0.2) | hours | seconds |
Stack
- BigQuery
- Google Tag Manager (server-side)
- Gemini
- Model Context Protocol
Source: Baachee working paper “Context Engineering for Marketing Intelligence” (v0.2, May 2026), section “The Skill Library Pattern”.
Next step
Want this running on your data?