Conversion rate is a means, not the end
A business outcome is what the company actually needs: profit, qualified pipeline, retained customers. A conversion metric is a measurable step towards it: purchases per visitor, sign-ups, demo requests.
The trap: you can raise a conversion metric while hurting the outcome.
A worked example
| Conversion rate | Average order value | Revenue per visitor | |
|---|---|---|---|
| A (control) | 3.0% | €60 | €1.80 |
| B (20% discount banner) | 3.3% | €52 | €1.72 |
Illustrative numbers. B “wins” on conversion and loses on revenue per visitor, and the discount may cut margin further.
Primary metric plus guardrails
For every test, pick:
- One primary metric that tracks the outcome closely and that the change can plausibly move (for example, revenue per visitor or completed orders per visitor).
- Guardrails: metrics that must not get worse, such as average order value, refund rate, lead quality or page speed.
Experimentation research calls a well-chosen primary metric an overall evaluation criterion (Kohavi, Tang and Xu, 2020). The lesson is the same: decide what “better” means before you look at results.
Practical exercise
For a store you know, write down: the business outcome (for example profit), one primary test metric that tracks it, and two guardrails that must not get worse.