Custom business KPIs without breaking standard numbers
Define your own KPIs without skewing shared numbers: naming, versions, checking numbers before activation, who approves, with a profit-after-packaging example.
By The DataxLand team· · 5 min read
A standard metric set answers most questions: net revenue, contribution margin, profit after fees, return rate. But every business has a few numbers of its own. One shop wants to subtract gift-box costs from profit. A chain of stores counts revenue without free gifts. An operations team wants the failed-delivery rate by province. These are legitimate needs — and also the fastest road to "everyone has a different number".
This post shows how to add custom KPI definitions without breaking the standard numbers: how to name them, how to version them, how to check the numbers before switching them on, and who gets to decide.
Why custom metrics so often break shared numbers
Three familiar scenarios:
- Overwriting the standard number. Someone changes the "profit" formula in the shared file to subtract packaging costs too. Last month's report and this month's can no longer be compared, and nobody remembers when it changed.
- Two names for one thing, one name for two things. Marketing calls GMV "revenue", the accountant calls net revenue "revenue". Both are right by their own definition, but the meeting goes wrong.
- A metric nobody checked. A hastily written formula divides by the wrong denominator, and the number is presented as fact for three months.
The common root of all three: the metric has no owner, no version and no check before it is used.
Five principles for custom metrics
| Principle | How | What it prevents |
|---|---|---|
| Don't edit the standard number, add a new one | A custom metric has its own name and id, e.g. a ws. prefix | Losing comparability with past periods |
| Build from existing blocks | Combine metrics and columns already in the semantic model | Formulas rewritten from scratch that drift from the standard |
| Versioned | Changing the formula bumps the version, with a date | "How was last month's number calculated?" |
| Check the numbers before activating | Compute over a sample period, compare with a reference metric | A wrong formula nobody notices |
| Someone else approves | The creator is not the one who activates | One person both writing and certifying |
The second principle matters more than it looks. "Profit after fees and packaging" should be defined as standard profit after fees minus packaging cost, not as a new profit formula rewritten from revenue. When the standard definition is updated (for example a new marketplace fee type is added), the custom metric picks up the change automatically.
Example: profit after fees and packaging
Take the DataxLand sample data (a simulated scented-candle shop, 1 Aug – 29 Oct 2026) and one assumption: boxes, wrapping paper and a card cost 6,000 ₫ per valid order.
ws.profit_after_fees_and_packaging = profit after fees − 6,000 ₫ × valid orders
= 89,301,200 − 6,000 × 2,309
= 89,301,200 − 13,854,000
= 75,447,200 ₫The assumed packaging cost eats about 15.5% of profit after fees. It's the kind of cost that is small per order but large over a quarter, and a generic metric set doesn't know about it because it sits outside the marketplace data.
Two things to do when activating a metric like this:
- Check it on a sample period: compute it for one month and compare by hand (or with standard profit after fees minus the total packaging cost in your books). The two numbers must match.
- Write the assumption into the description: "6,000 ₫/order, per the box supplier's quote from August". When the box price changes, change the calculation and the version goes up.
The lifecycle of a custom metric
| Status | Meaning | Who |
|---|---|---|
| Draft | Being written or its calculation just changed; not yet used in dashboards, AI or reports | Analyst |
| Active | Checked against data and activated | An Admin or the Owner activates |
| Archived | No longer used, history kept | Admin or above |
Changing the calculation or the unit sends the metric back to draft: it has to be checked and activated again. Changing the display name or description doesn't, because the number doesn't change.
When to edit the description and when to create a new metric
- You only want a different name (for example "profit after fees" becomes "net sales profit" because that's what the team is used to): edit the display name, description and synonyms. The formula stays the same.
- You want to calculate it differently, even by one item: create a new metric. Don't edit the standard metric's formula.
- A question that keeps being asked has no metric yet: turn that answer into a defined metric, instead of letting every ask produce a different calculation.
How DataxLand does it
The industry × department packs are a starting point, not a limit. In Metric handbook → Workspace metrics, click Create metric, declare the calculation from the columns and metrics in the semantic model, the unit and the allowed dimensions, then Save draft. Click Check: DataxLand computes the metric over a Sample period, shows the Value and compares it with a reference metric, reporting Number check passed or Number check failed with the version. The Analyst role creates and checks metrics; an Admin or the Owner clicks Activate. Details are in Workspace metrics.
A custom metric has the id ws.<name> and runs on the same metric engine as the standard ones, so dashboards, Si, scheduled reports,
alerts and MCP all read the same definition. Every number comes with a confirmation level: Standard for pack metrics,
Approved for workspace metrics, Unconfirmed for Si's exploratory answers.
When you ask a question for which Si has to put a formula together itself, Si can propose it as a metric; the proposal appears under Drafts from Si with a Proposed by Si label, waiting for you to check it and then Approve or Discard. See also Metrics and the semantic layer and the semantic layer page.
FAQ
Does editing the description of a standard metric change the number?
No. In the metric library, Edit description only changes the display name, description, synonyms and the workspace note; an edited metric carries an Edited by workspace label and can be reset with Restore original. The formula stays the same.
Who can create and activate custom metrics?
The Analyst role can create, edit and check them. Activating and archiving need an Admin or the Owner.
What happens to old reports when the formula changes?
Changing the calculation bumps the version and sends the metric back to draft until it is checked and activated again. Reports and alerts that use the metric always read the version in use.
Should AI create metrics on its own?
AI can propose, but people approve. Si's exploratory answers carry the Unconfirmed label; only after being checked and activated do they become a shared Approved metric.
Read next: What is a semantic layer · AI answers with sources · P&L dashboard: 8 metrics