Intelligence / Data Analytics
Turn what you already collect into decisions.
Pipelines, models and reporting that answer the questions the business is actually asking.
Start a projectTwo dashboards, two numbers
Nobody trusts the reporting because the same question returns different answers depending on where it is asked.
The data exists but cannot be reached
Everything is in the production database, and every analytical query threatens the application.
Reports describe the past
Plenty of history, no signal — nothing that changes a decision this week.
What we do
- 01Data pipelinesIngestion and transformation that is versioned, tested and rerunnable.
- 02Warehouse modellingOne definition per metric, written down, so the number stops depending on who asks.
- 03Reporting and dashboardsBuilt around decisions rather than around the available fields.
- 04Event tracking designDeciding what to capture before capturing everything and hoping.
- 05Data quality monitoringTests on the data itself, alerting when the shape changes.
- 06Migration from spreadsheetsGetting the business off the shared file without losing the logic buried in it.
How we work
- 01
Questions
1 week
The decisions that need support, before any pipeline is drawn.
- 02
Model
2–3 weeks
Metric definitions agreed and written down.
- 03
Build
4–10 weeks
Pipelines, warehouse and reporting, tested end to end.
- 04
Run
Ongoing
Quality monitored; definitions maintained as the business changes.
Marketplace operations reporting
- The challenge
- Vendor performance, refurbishment grading and dispute rates all tracked separately, with no shared definition of a healthy vendor.
- The approach
- One warehouse model with a single definition per metric, fed by tested pipelines, surfaced in the admin tools the operations team already used.
- The result
- Vendor decisions made against one number instead of three, inside the tool where the decision happens.
Questions we get asked
Do we need a warehouse?
Not always. Below a certain volume a read replica and disciplined modelling is the right answer, and we will say so.
Which tools do you use?
Postgres or a managed warehouse, dbt-style transformation, and whichever BI layer your team will actually open.
Can you work with our analysts?
Yes — the goal is that they own the model afterwards. Building something only we can maintain would be a failure.
What about data protection?
Minimisation and access control are part of the design, not a later review. This overlaps with the security work and we treat it that way.
How long until we see something?
A first useful report usually lands weeks before the full model does, and we sequence deliberately for that.