Why did my conversion rate go up but my profit go down?
Short answer: Usually because the change that lifted conversions also lowered what you keep per order — most often a discount, sometimes a cheaper product mix. Conversion rate only counts buyers; it ignores order value, discounts and product costs. Compare profit per session for each version before you ship the “winner”.
A simple example (illustrative numbers)
| Control | Variant with 20% off | |
|---|---|---|
| Sessions | 1,000 | 1,000 |
| Conversion rate | 2.0% | 2.6% |
| Orders | 20 | 26 |
| Average order (before discount) | $80 | $80 |
| Discounts given | $0 | $416 |
| Net sales | $1,600 | $1,664 |
| Product costs (40%) | $640 | $832 |
| Profit | $960 | $832 |
| Profit per session | $0.96 | $0.83 |
The variant “won” on conversion rate (+30%) and lost $128 in profit on the same traffic.
These numbers are made up to show the math.
Where it happens most
- Discount tests — the discount is the reason conversions rise, and it comes straight out of margin.
- Free-shipping thresholds — more orders, but more shipping absorbed.
- Theme or layout changes that push cheaper products — more buyers, smaller baskets.
What to check before you ship
- Net sales per variant, not just orders.
- Total discounts per variant.
- Profit per variant (uses your product costs — make sure costs are filled in).
- Profit or revenue per session — the fair comparison when traffic is split.
With Upshot
On the Pro plan, Upshot shows net sales, discounts, profit and revenue per session for every variant of your Rollouts experiments, so a “conversion win, profit loss” is visible before you apply the result.
About discount experiments: Shopify doesn't attribute revenue or orders to variants when a rollout includes a discount — it reports visits and conversion rate only. Upshot fills the gap with an estimate: orders that used the test discount count as the treatment, other Online Store orders as the control. It shows the estimate only when the discount is automatic (not a code), has no minimum or product conditions, and no other discount test runs at the same time; otherwise it tells you revenue by test group isn't available. Estimated numbers are always labeled, and the probability in the verdict comes from conversion rate, not from the estimate.