Platform
One platform for every company in the group
Data flows one way: from source systems to a warehouse with a space per company, through one metric layer, out to dashboards, models and agents. Governance covers every layer.
Five layers
Ingestion
Automated loads from ERP and accounting, CRM, bookings or admissions, payments, the website and app, and key spreadsheets. Change data capture reads only the rows that changed.
Airbyte or Fivetran connectors, CDC, GA4 or app events
Warehouse
One warehouse with a clean data model and full history. Raw data lands untouched first, then is cleaned and modelled.
BigQuery, Snowflake, Microsoft Fabric or Databricks; PostgreSQL for small clients
Transformation and metrics
Data is transformed inside the warehouse in layers: staging, intermediate, business tables. Each key metric is defined once, as code, with an owner.
dbt models, metrics as code, tests
Orchestration and quality
Jobs run on schedule in the right order. Tests check nulls, duplicates, accepted values, freshness and volume. A failed load alerts the team before anyone opens a dashboard.
Airflow, Dagster or Prefect; dbt tests; anomaly alerts
Dashboards and agents
Dashboards and AI agents read the same certified metrics, so they cannot disagree. Row-level security shows each person only what they may see.
Power BI by default; an LLM with tools, text-to-SQL and evaluation sets
Source systems
- ERP
- 1C
- CRM
- Bookings
- Payments
- App events
- Excel
Own space
Company A
Own space
Company B
Own space
Company C
Shared layer by agreement
A + B
One metric layer
Dashboards, models, agents
Governance: owners, access by role, masking, audit log
What changes for a group
A space per company, a shared layer by agreement
Each company keeps its data in its own space. The shared layer holds only datasets that two or more companies have agreed to share.
Shared definitions make companies comparable
To compare or combine companies, they share the same definitions of customer, date, location and product category. This is what makes synergies visible.
Linking people needs consent
A person is linked across companies only with consent and a hashed key, never with raw phone numbers in a shared table.
What you get
- Automated loads from every agreed source
- One warehouse with a clean model and history
- A metric layer: each key metric defined once, with an owner
- Nightly data quality checks, alerts and a runbook
- Access by role and an audit log
- Architecture notes and documentation as code
Questions about the platform
Which warehouse and BI tools do you use?
The design is tool-agnostic. The warehouse is chosen after the audit, based on your existing stack and where data may be hosted. Power BI is the default for dashboards, and metrics are defined as code in the warehouse.
Batch or real time?
Daily or hourly loads cover almost all management needs. Real time is for operations such as fraud or stock control, and it costs more.
What happens when a source system changes?
We agree a data contract with the source owner, test the schema on every load and alert on changes before they reach a report.
Start with one question
A pilot takes four weeks, covers one question your leader cares about and has a written measure of success.