E-commerce conversion tracking: define the funnel before judging results

Conversion tracking is useful only when everyone understands what the numbers represent. A visit, product view, checkout attempt and completed order are different events; combining them can make a change look more successful than it is.

Define the question and measurement period
Begin with a practical question, such as whether visitors who open a product page can complete a purchase. State the date range, included traffic and counting method. Distinguish people from sessions, and orders from customers, before comparing reports.

Map a short funnel

• Relevant visits or sessions, using a defined basis.
  
• Product views.
  
• Checkout starts.
  
• Completed orders.
  
• Fulfilled orders, cancellations and refunds.
  

For a hypothetical period with 1,000 sessions and 20 completed orders, orders divided by sessions gives 2%. If five orders are later refunded, the original purchase rate stays 2% under that definition; a separate retained-order measure would show 15 divided by 1,000, or 1.5%. State which measure you mean rather than silently replacing one with the other.

Investigate before making a claim
Check duplicate events, missing transactions, internal visits and differences between systems. A sharp change may reflect broken measurement or a different audience. Small samples and short periods can also be noisy.

Use the contribution calculation in our product validation guide alongside conversion data. More orders do not necessarily mean more money remains after costs. Which step in your purchase path needs a clearer definition before its performance can be judged?

BacklinkTalk editorial discussion. Examples are hypothetical, not personal earnings claims. Members are welcome to contribute their own experience and evidence.

Editorial follow-up: keep the denominator visible
Suppose one report shows ten orders from 200 product-page sessions, while another shows ten orders from 500 site sessions. The rates are 5% and 2% because they answer different questions.

Neither figure is automatically wrong, but comparing them as if they were the same measure is misleading. Put the event definition, denominator, date range and exclusions beside every rate you use in a decision or forum discussion.

Editorial follow-up: reconcile orders before analysing behaviour
Compare completed orders in the store's records with purchase events in the reporting system for the same period and time zone. Look for duplicates, missing events and differences in how cancelled orders are treated.

Use order identifiers privately where needed to investigate, and keep customer information out of public screenshots. Establishing that the basic counts agree is more valuable than optimising a funnel whose final event is being counted incorrectly.

Editorial follow-up: avoid drawing conclusions from a tiny test
If one page version receives two orders and another receives three, the extra order does not establish a dependable improvement. Traffic source, timing and chance may explain the difference.

Define the comparison and decision rule before looking at results, and record uncertainty instead of forcing a winner. Where traffic is limited, watching a few suitable users attempt a purchase may identify clear usability problems without pretending to prove a precise conversion lift.

Editorial follow-up: connect a rate to the economics
A hypothetical page could increase orders by promoting a cheaper item, while reducing the amount left after fulfilment and support. Review contribution per order and total contribution alongside the purchase rate.

Also track whether refunds or service requests change. The most useful measurement reflects the business question being asked. A higher percentage is not a complete success criterion when the costs or quality of the resulting orders have changed.