Offer Walkthrough
Call with CFO Peterseim and IT lead Rex on the proposal submitted 2026-05-27 via Colayer.
Notes
- Effort estimate: Rex reviewed internally, thought ~4-5 days fewer if everything ran perfectly; both agreed that is theoretical and the buffer is comfortable.
- Tech stack: Snowflake confirmed. Peterseim asked about BigQuery/Google Cloud; Bijan explained the Snowflake preference (critical of Looker/Luca; BigQuery fine but Snowflake recommended). Both deferred to Bijan.
- Hosting: Snowflake in the Frankfurt (EU) region; a 30-day trial (no data egress, not training data) to be created only when needed. Snowflake’s embedded Postgres is enough for metadata/Airflow, so no separate AWS.
- Full vs. filtered dump: full historical dump chosen; the cost delta is marginal (~3-5% storage) and filtering is not worth the engineering effort, since Claude queries the semantic layer, not raw data. Delta load deferred to Phase 2.
- Schemas: Rex flagged BC vs. Sage schema inconsistencies as a risk; a staging normalization step is already planned.
- Freshness/monitoring: dbt freshness tests on key tables; Airflow for scheduling and alerting (Teams or email).
- Semantic layer / KPI definitions: Bijan flagged this as the most important factor for success (revenue is defined differently at every company). Plan: iterate from raw data and existing Excel reports, then align definitions; keep it budget-time. A centralized semantic layer keeps definitions consistent and avoids per-user context-file sprawl.
- Multi-source: CRM + ERP unified by ID mapping in the semantic layer. Calendar data is straightforward; email classification needs an LLM layer. Rex proposed shared mailboxes as a DSGVO-clean path for team-level email analysis.
- Excel: budget/forecast integration is a typical case; needs consistent column structure.
- AVV: Gemma’s AVV to be used; ROTOP also needs Snowflake’s (self-service).
The transcript cut off before an explicit next-steps recap; a revised offer was expected.