Headline statistics
Time span: January–December 2024. Synthetic, privacy-preserving overview for communication and demos.
Private clients
10,000
≈ 90.5% of base
SME clients
1,000
≈ 9.05%
Corporate clients
50
≈ 0.45%
What’s shown here?
A compact description of tables, a few randomly generated rows per table (not our data), and visual relationships across entities.
- No raw data exposed. Samples below are synthetic placeholders.
- Joins between tables is done by (cust_id).
Data model & synthetic process (short)
Tables
- customers — master data (type, group, language, KYC dates).
- transactions — dated ledger-like events; optional MCC/merchant country for card purchases.
- life_insurance — insured amount, scheduled payment, status codes.
- property_insurance — type E/P, premium, repurchase value, status.
- leasing_agreements — amounts and contract lifecycle status.
- margin_assets / margin_liabilities — monthly balances & interest bases.
- pension — contributions by fund and status.
How the synthetic noise works
- Per-customer noise: multiplicative factor (Laplace, clipped ±5%), consistent time/day offsets per customer.
- Temporal coarsening: timestamps rounded to 15 minutes; in-customer shuffling of minute/second fields.
- Splits: 20% of transactions split into 2–3 parts with extra ±30 min / ±1 day jitter.
- Merges: 20% chance to merge 2–3 consecutive transactions (same counterparty, within 7 days).
Sample records (5 per table)
Generated on the fly for illustration only. Joins are visualized via the highlighted cust_id headers. Hover the i icons for column help.
Data dictionary
Key columns and codes, summarized. The same content powers the header tooltips.