Skip to content

Guides · · 3 min read

One definition per metric: what you keep from a data project

Analytics and BI turn data into numbers your teams agree on: we give every metric one definition, an owner and tests that run with every refresh.

By Etheon

On this page

Analytics and BI turn data into numbers your teams agree on: we give every metric one definition, an owner and tests that run with every refresh, and build the reports and dashboards your teams read.

What makes a report one your teams can trust?

"Reports nobody trusts" is one of the needs on our Services page, and it leads to Data: one set of pipelines and definitions behind every report. One definition and one owner for every metric, and tests that run with every refresh, before anyone reads a number.

What does a metric come with?

Its definition, its owner, its source, its refresh and its tests. Here is the format, on an illustration (not a client's figures): weekly active customers, customers with at least one order in the week, each counted once.

Part

Example

Owner

Finance

Source

fct_orders

Refresh

Daily, before the working day

Tests

Not empty, one row per week

Who sees what?

Data governance decides who can see which data, where it is kept and for how long. Access is granted by role, kept to what each role needs, and every access is logged:

  • Analysts see orders, events and aggregates in full, with customer identity and ticket text masked.
  • Data engineers see every dataset to build and fix the pipelines, with customer identity masked.
  • Support agents see orders and tickets to answer a customer; names and emails are masked, and every lookup is logged.
  • ML models read orders and events to learn from, with ticket text masked and no customer identity.
  • Executive dashboards show aggregates only: totals and trends, with no single customer's record.

Personal data is masked where it is not needed, and retention and deletion run on a schedule.

Where is our data stored?

In the region you choose. Your data and its backups stay in that region.

What about the data a model learns from?

It is versioned with the model. ML data is the data your models learn from and are tested on. We label datasets for the model's task, redact personal data before labelling, and version each dataset with the model it trains, so any result can be traced back to its data.

What do you keep?

Every metric and every dataset comes with its definition. You keep:

  • The metric definitions, in the format above.
  • An access map by role.
  • Retention and deletion rules.
  • Access logs your auditors can read.
  • Labelled datasets, with their versions linked to each model.
  • A record of how each set was made.

Do you train models on our data?

We never train models on your data without your opt-in. Your data stays isolated from other clients. How we handle it is on Trust, and our data processing agreement is public.

What you keep is drawn on our Data page, with who sees what beside it. Tell us your use case and timeline. We reply within one business day.

Published
Category
Guides
Reading time
3 min read

More news. What else Etheon has shipped.

Get started with Etheon

Tell us your use case and timeline. We reply within one business day.

Data

Pipelines, platforms and BI

Explore Data

Research index

Every paper, milestone and release from our Lab.

Browse all research