Ecommerce analytics: how to count people instead of clicks

Most shop analytics count events: page views, sessions, clicks. Those are easy to collect and they answer questions nobody asked. The questions a shop actually has are about people — how many came, how many bought, how many came back — and people and sessions are not interchangeable.

The difference is not academic. Conversion rate computed per session is systematically lower than per person, because the people who convert tend to visit several times first.

Choose the unit deliberately

Three units, three different meanings, and reports routinely mix them:

• Page views tell you about content and almost nothing about customers.
• Sessions are an artefact of a timeout — usually 30 minutes of inactivity. A person who browses at lunch and again in the evening is two sessions and one person. The number is real but it is not a count of anybody.
• People are what a shop reasons about. How many distinct humans visited, of whom how many bought.

Pick people as the default unit for anything reported to a business, and keep sessions for diagnosing behaviour on the site.

The four numbers worth a dashboard

1. People who visited, this period against the last.
2. People who bought, and therefore conversion as a share of people.
3. Revenue per visitor — the number that moves when either of the first two does, and the one that catches a traffic increase that brought nobody useful.
4. Repeat purchase rate, as a share of buyers. The slowest-moving and most important of the four.

Everything else is a diagnostic you open when one of these moves.

Read events, not just orders

An order tells you somebody bought. Only the events tell you how many did not, and the gap between the two is where the decisions are.

• Product views without a cart — interest that did not convert to intent. Usually a page problem.
• Carts without checkout — intent that did not convert. Usually a shipping-cost or payment problem, and the cheapest funnel to fix.
• Checkout without order — the most expensive leak and the most specific. Watch it per payment method.

Reported per person rather than per session, those three ratios point at a cause. Reported per session, they mostly measure how many times people reloaded.

Living with the traffic you cannot see

A share of visitors decline tracking, and that share must be respected rather than worked around. That means your analytics measure a subset, and the honest response is to say so.

1. Know your consent rate and report it beside the numbers. "Conversion rate 2.1 % across 64 % of sessions measured" is honest; a bare 2.1 % is not.
2. Compare like with like. Trends within the measured subset remain valid even when absolute counts are incomplete.
3. Use order data as the anchor. Orders are complete regardless of consent, because they are transactions rather than tracking, so any revenue figure should be computed from orders rather than from analytics.
4. Do not model the gap away and then report the model as measurement.

One person, several devices

Counting people requires deciding who is one person, and the honest answer is that a phone and a laptop are two until one of them identifies — by ordering, by subscribing, by signing in.

When that happens, the two trails should merge, and everything the person did beforehand should attach to the same record. A system that only creates a person at checkout throws away the browsing that led there, which is exactly the part that explains the purchase.

Expect your visitor count to be slightly high and your repeat rate slightly low as a result. Both improve as more visitors identify, which is itself an argument for getting the email address early.