Give me read access to a store and an hour, and there are about seven numbers I want before I have an opinion. None of them mean much alone, which is the whole point.

Store owners usually open with conversion rate. It is the number they have been told to care about, and it is the easiest one to feel bad about. On its own it tells you almost nothing, because a conversion rate can fall for reasons that are good for the business and rise for reasons that are terrible.

What I want is a set of numbers that constrain each other. When one moves, the others tell me whether the movement was real.

A frame to hang the numbers on

I sort everything into four stages: traffic, experience, conversion, retention. It is not a sophisticated model. Its job is to stop me from attributing a problem to the wrong stage, which is the most common analytical error I see.

Traffic covers who arrives and where from. Experience covers what happens between arrival and intent to buy. Conversion covers the mechanics of completing an order. Retention covers everything after the first purchase.

A drop in orders can originate in any of the four. The metrics below each belong to one stage, and reading them in order usually locates the problem faster than staring at the aggregate.

Conversion rate is a ratio, not a verdict

Conversion rate is orders divided by sessions. Both halves move independently, so the ratio can change without anything meaningful happening to the business.

A store runs a broad awareness campaign. Sessions climb, orders climb a little, conversion rate falls. Nothing broke. The traffic mix changed, and the new visitors are further from purchase.

The reverse is more dangerous. A store cuts its top-of-funnel spend, keeps only branded search and returning visitors, and conversion rate jumps. The dashboard looks like a success. Revenue is quietly shrinking.

So I never look at site-wide conversion rate without splitting it. New against returning. Device. Channel. Landing page group. The aggregate is an average of populations that behave nothing alike, and averages of unlike things are mostly noise.

Average order value hides its own distribution

Average order value is the metric most likely to be technically correct and practically misleading. A handful of large orders will pull the mean somewhere no actual customer lives.

I ask for the median alongside it, and ideally a histogram. A store with a $95 mean and a $42 median is two businesses sharing a domain, and the strategies those two businesses need are different.

AOV also reacts to things that are not merchandising decisions. Free shipping thresholds bunch orders just above the line. Bundles inflate it without improving margin. A discount code circulating on a deal forum can drag it down for a month while volume looks healthy.

When someone tells me they want to raise AOV, my first question is whether they want more revenue per order or more profit per order. Those point at different tactics, and they occasionally point in opposite directions.

Acquisition cost only means something next to margin

Customer acquisition cost is meaningless as an absolute. A $40 CAC is excellent for a business with a $300 first order and a high repeat rate, and fatal for one selling a $35 consumable once.

The comparison I want is CAC against contribution margin on the first order, then CAC against contribution over some defined window. Twelve months is common. The window matters more than people admit, because a business that only works over three years is a business that needs financing, not a marketing plan.

Blended CAC and channel CAC answer different questions too. Blended tells you whether the business is viable. Channel-level tells you where to move budget. Reporting one and deciding with the other is a recurring mistake.

I also want to know what counts as an acquisition cost in their calculation. Some teams include creative production, agency fees and platform costs. Others count media spend only. The second number is always prettier and always wrong.

Repeat purchase rate

For most stores this is the number that decides whether the business compounds or treadmills, and it is the one least likely to be on the dashboard.

Repeat rate needs a time frame to mean anything. “Thirty percent of customers order again” is not a fact until you say within what period. Sixty days and twelve months describe different businesses.

Cohorts are worth the effort here. Group customers by the month of their first order and watch each group’s behaviour over the following months. This separates a genuine improvement in retention from the arithmetic effect of a big acquisition month landing in the denominator.

A store with a weak repeat rate can still be healthy if the first order pays for itself with margin to spare. What does not work is planning as if retention will arrive later, without any mechanism that would cause it.

Checkout completion is its own question

I keep checkout completion separate from conversion rate on purpose. Conversion rate measures the whole journey. Checkout completion measures a short mechanical sequence performed by people who have already decided to buy.

When completion is low, the causes are usually concrete and fixable. A payment method the audience expects and does not find. A shipping estimate that appears too late. A validation error that rejects a legitimate address format. A third-party script that fails on one browser.

This is also the step where analytics most often lies, because payment redirects break session continuity and duplicate events inflate the numerator. Before treating a low completion rate as a UX problem, I check whether it is a measurement problem.

Channels disagree with each other

Every ad platform reports on its own behalf, with its own attribution window and its own definition of a conversion. Add the platform-reported numbers together and you will usually exceed the actual order count, sometimes by a lot.

The store’s own order data is the only figure that ties to money in the bank. I treat platform numbers as directional signals about relative performance inside that platform, and the backend as the source of truth for totals.

Two habits help. Compare each channel’s platform-reported orders to backend orders tagged to that channel, and track the ratio over time rather than chasing an exact match. And look at revenue trend against total spend at the account level, because that relationship survives attribution disputes.

Reading them together

Here is the pattern I run through when a store reports that revenue is down.

• Are sessions down? If yes, the problem is traffic and the site metrics are consequences, not causes.
• Are sessions flat and conversion down? Split by device and channel before anything else. A single broken template or a mix shift explains most of these.
• Are sessions and conversion flat but revenue down? Look at AOV and discounting. Something changed in what people buy, not whether they buy.
• Is everything flat except margin? The problem is upstream of marketing, in pricing, shipping cost or supplier terms.
• Do the numbers look fine and the bank account disagree? Check refunds, chargebacks and cancelled orders, which sit outside most marketing dashboards.

The ordering matters more than any individual metric. Conversion rate on its own has caused more misdirected work than almost any other number in e-commerce, mostly because it is the number that fits on a slide.

If I could keep only two, I would take orders and contribution margin, both split by new and returning. Everything else is a way of explaining why those two moved.