Dunwell Insights · Conversion Check · Sample
How far visits got, 25 November 2020 to 31 January 2021
Share of all visits reaching each step in order. The lighter bar adds visits that reached the step without the one before.
Reached in orderReached, skipping a step
Summary for the owner
What we found
What we'd do
Open the live dashboardExplore the funnel by day, compare raw and like-for-like rates for any visitor group, change the assumed lift on each test, and download the test plan and the group tables.1 · Tracking first
Add-to-cart and checkout events each day
Raw event counts from the tracking. Checkouts carried on while cart events stopped.
Add to cartBegin checkoutPurchase
Tracking checks
2 · The funnel
Each step, by visit and by visitor
Share of visits that bought, each day
Tracked period. Grey band: the 2020 holiday rush.
3 · Who buys
Purchase rate by traffic source
Share of visits that bought. Like for like holds the mix of new and returning visitors fixed.
As recordedLike for like
Purchase rate by first page
Groups with at least 1,000 visits. Same two measures.
As recordedLike for like
New and returning visitors
Share of visits reaching each point
Why January sold less than December
Change in the share of visits that bought, split into visitor mix and change within groups
4 · Predicting a purchase at the first product view
Every model on the test month
Predicted against actual
Test month, ten equal groups of visits by predicted chance. On the line means the chances can be taken at face value.
Chosen modelGroup rate
How many buyers the top of the list holds
Test month. Visits sorted by predicted chance, highest first.
Chosen modelGroup rateRandom order
What the model leans on
Share of the model's total gain from each input
Does it hold up for every group?
Test month, log loss (lower is better)
5 · What to test
Test backlog
January 2021 traffic. Extra orders assume a 5% relative lift (low case 2%, high case 10%). These lifts are assumptions, not findings.
6 · How we did this
The plan was locked before any group was compared
The models
Checks we ran
What the model can see
Data dictionary
Decisions log
What this data can't tell us
This analysis is based on the data described here and the methods shown. It supports decisions. It is not a promise of any result, and it is not financial, legal or tax advice.
Every job. Done well.