If you open the statistics only at the end of the month, it is hard to remember what things were like before the promotion. Between the launch and the report, new posts may have come out, the price may have changed, or an external recommendation may have appeared. That is why part of the work with the result needs to be done before the first placement.
Before the start: leave a short record
Name the campaign, specify the dates, links, spend and the main action: an inquiry, a registration or a confirmed purchase. Save the metrics for the preceding comparable period. Separately note the circumstances that hinder comparison: a premiere, a holiday, a sale, a rescheduled event.
Check the path on the website: the page opens, the form submits, the test event is visible in the right place. After checking, exclude your own tests from the working report. Do not wait until the end of the month to discover that requests were not being recorded at all.
Record the ordered activity separately from advertising, real visits and customer actions. Otherwise, a counter that increased as part of an order is easily mistaken for independent audience interest.
Identify the group you will observe
A cohort is a group with a common characteristic. For example, new visitors who came via links from a specific campaign during the first week. This approach lets you see what happens to one group over time. In GA4 cohort analysis, the group is built from device data; User-ID is not taken into account when it is created. Google Analytics help.
A hypothetical example: this group included 100 new visitors. It is these one hundred that remain the basis of the calculation. New people from the following week should not be quietly added to the old group, otherwise the final shares will answer a different question.
This does not mean that analytics infallibly recognizes a person across all devices. Record the method of accounting used and its limitations. If visits cannot be reliably combined, do not pretend that you have obtained a complete history of each customer.
After seven days: look for breakdowns and the first signs
Check the actual inquiries and the path to them. Read the reasons for abandonment without including personal data in the general report. In our hypothetical example, out of one hundred visitors, eight left substantive requests, and three people paid. These are 8% and 3% of the original group, not market benchmarks.
With such small numbers, the conclusion must be cautious: the path works, the first actions have appeared, but the exact effectiveness has not yet been established. If the form broke for two days, fix it first and note the period of the failure. Do not call a source bad when it brought people to a page that was not working.
After twenty-eight days: return to the same people
Suppose that by the end of the observation period the original group had six unique buyers, and two of them made a repeat purchase. The buyer share is 6%. Repeat buyers do not turn six into eight; the number of transactions is counted separately.
Reconcile payments and refunds with confirmed records. Look at which costs relate to the campaign and to fulfillment. If the money cannot yet be verified, state plainly that the result is preliminary. A demonstration order placement is not revenue.
To compare sources, keep a single attribution model: changing it can redistribute conversions between channels. Metrica rules.
Choose the next check
The seven- and twenty-eight-day timeframes here are a convenient working example, not a promise that the entire effect will show up within that time. For a long customer decision, the observation will have to be continued.
If the plan includes post views, keep those parameters separate from the customer cohort. The form below helps you view them. If it operates in demonstration mode, completing it is not considered a payment.
✦ Choose a quantity and place an order →
Close the review with a single action: repeat the placement, change the offer or stop the spend. Next to it, note which data confirmed the decision and what is still missing.