Which Numbers to Watch to Tell If Your E-commerce Is Working
Assess commercial performance without stopping at visits and revenue.
SqualiOnline editorial team · 2026-09-07
An online store produces a lot of numbers and very few decisions. Revenue and visits are the two everyone looks at, and the two that, on their own, don't let you choose what to do next week: they tell you whether it went well or badly, not where to act. This guide is meant to build a small dashboard of well-defined numbers and use them to generate checks instead of opinions.
Define first, count later
Almost every discussion about an e-commerce store's data is a discussion about definitions that were never agreed on: two people look at the same screen and mean different things. Fix them once and write them next to the number.
| Number | Definition to fix | Trap |
|---|---|---|
| Sessions | Visits in the period, excluding internal access and automated systems | Counting your own access and that of monitoring services |
| Carts | Sessions in which at least one product was added | Confusing adding to cart with intent to buy |
| Valid orders | Accepted orders, net of cancelled, unpaid, test, and internal ones | Using the total of orders registered by the store |
| Net revenue | Value of valid orders excluding taxes, shipping collected, and discounts | Comparing gross revenue against net costs |
| Returns and refunds | Attributed to the order's period, not the refund's period | Seeing a great month that turns mediocre two months later |
| Customers | Distinguished between first purchase and repeat purchase | Treating a returning customer as a new one |
The rule that brings the most order: every number has one owner and one source. If revenue is read from the store and costs from accounting, the two worlds need to be reconciled at least once a month, otherwise meetings turn into arguments over which screen is right.
Revenue isn't the result
An increase in sales can go hand in hand with a worse bottom line. You need two more measures, both reconstructible from data the company already has.
- Margin per order: selling price minus the product's cost, minus the shipping you pay, minus packaging, minus payment fees. Shipping offered for free is a cost, not the absence of one.
- The observable acquisition cost: what you spend on advertising and tools, divided by the valid orders that came from there. Observable means you don't include estimates of brand awareness or word of mouth.
- The cost of returns: refund, return shipping, the time of whoever inspects and restocks, and goods that can no longer be sold at full price.
- Support time per order: if a certain category generates two questions per order, it costs more than the margin suggests.
With these measures the question changes: not are we selling more, but which products and which channels bring in orders that leave something behind.
A sample dashboard
Illustrative, and deliberately short: a dashboard no one reads is worse than no dashboard. Every tile carries the definition and the limit of the data next to it.
- Valid orders and net revenue for the period, compared against the same period last year and not just last month.
- Estimated overall margin, with a note on which costs are included and which aren't.
- Orders by source channel, with a note that a portion isn't attributable.
- The path in four numbers: sessions, carts, checkouts started, orders. It's there to show where things narrow.
- Returns for the period, broken down by product category.
- A technical control tile: orders recorded by the store versus orders recorded by the analytics. If they diverge, everything else needs to be read with caution.
Comparisons that hold up
Comparison is where wrong conclusions get manufactured, almost always in good faith.
- Small samples: with few orders a month, a percentage change means nothing. Look at absolute numbers and longer periods.
- Composition: if you've sold a lot of a low-margin product, revenue rises and margin falls. It's not that the store is working worse, it's that the mix has changed.
- Devices: on phones, people browse more and often buy less. Comparing conversion across devices without accounting for this leads you astray.
- Channels: whoever arrives searching for your store's name already knows you. Mixing them with the rest makes every average useless.
- Promotions and holidays: they need to be marked on the data calendar, otherwise they become the following month's normal baseline.
From anomaly to check
A number that moves isn't a conclusion, it's the start of a check. The right order is to look for the boring explanations first, which are also the most frequent ones.
- Describe the anomaly in one sentence, with a period and a segment: since Monday, orders from phones are half what they were.
- Check that it isn't a measurement problem: tracking changed, consent settings changed, a new filter.
- Check the external, mundane causes: a holiday, a promotion that ended, a campaign turned off, a product out of stock, a carrier running late.
- Reproduce the path by hand, on the device and channel involved.
- Only at that point form a hypothesis and decide on one intervention, just one, with a date to review the data again.
What this guide doesn't cover
This guide covers monitoring a store that's already running. If the store doesn't exist yet, the prior question is whether the organization is ready to sell online — that is, warehouse, shipping, returns, and support — and that has its own dedicated guide. The obstacles that keep an order from being completed are covered in the guide on checkout.
Frequently asked questions
How often should you look at these numbers?
Once a month for the full picture, with reconciliation between the store and accounting. The technical check — whether recorded orders match — is worth doing more often: it's the one that breaks silently and invalidates everything else.
How do I attribute a sale if the customer passed through several channels?
Carefully, and by stating the method you use. No system reconstructs the entire path, and forcing attribution produces false rankings. It's more useful to watch the overall trend when you turn a channel on or off than to assign every order to a single culprit.
Do I need paid tools to do this?
Not necessarily. A spreadsheet updated once a month, with written definitions and sources noted, is already enough for most stores. Tools become worth it when volumes make the manual work too slow, not before.
Let's define the data that's useful for evaluating your online store.
If you’d like to talk it through, the service that handles this is E-commerce.