“Custom themes, apps and Functions, A/B tests that move conversion”
–converted
– of – visitors
templates/page.experiments.json
Experiments.
Testing is how a store stops arguing about opinions. This page runs a real A/B test on its own home page, shows the playbook behind every test I ship, and hands you the calculators I plan them with.
1A/B test live on this site right now
9/16builds here run tests or behavior analytics
4tools across them: Intelligems, Clarity, Convert, Shoplift
experiments/hero-lede-v1
Running on this site.
Not a mockup. First visits to the home page are bucketed by a Cloudflare Worker and these numbers come straight from D1.
Live on the home page
Does leading with outcomes get more people to reach out?
Conversion = copy the email, open Contact or send the form
“tests that prove it, and automations that keep it running”
–converted
– of – visitors
Chance B beats A–
Lift of B–
Visitors so far–
Loading live results…
How it's measured
- 01 · BucketThe Worker assigns A or B and sets a cookie
- 02 · RenderThe variant is in the HTML: no flicker
- 03 · ExposureCounted once the hero is on screen
- 04 · ConversionEmail copied, Contact opened or form sent
- 05 · CountOne row per visitor in D1
snippets/test-playbook.liquid
How I run a test.
Six steps, every time. Most of a test's value is decided before launch and after the end.
- 1
Research
Recordings, heatmaps, reviews and drop-off point to one problem worth solving.
Problem statement - 2
Hypothesis
Because we saw X, changing Y will move metric Z. No sentence, no test.
One-line hypothesis - 3
Sizing
Traffic and the smallest lift worth finding set the runtime before anything is built.
Sample size + end date - 4
Build & QA
Flicker-free variants, checked on every device, currency and tracking event.
QA matrix - 5
Run
Full weeks only, no peeking, and a sample-ratio check on day two.
SRM check - 6
Read & ship
Primary metric, then guardrails. Winners become theme code, losers get written up.
Write-up + cleanup PR
assets/stats.js
Plan it before you build it.
The three checks I run on every test: how long it needs, whether the result is real, and whether the split itself is healthy.
config/test-areas.json
Where tests pay off.
The levers I reach for first on a Shopify store, page by page. Patterns, not promises: which one wins depends on the store.
Product page
- Subscribe & Save by default
- Price per serving by the CTA
- Bundles and quantity offers
Moves: Conversion, AOV
Cart drawer
- Free-shipping and gift goals
- Gap-closer upsells
- One-click subscription upgrade
Moves: AOV, checkout rate
Offers & pricing
- Price points and anchors
- Percent vs. dollar savings
- Shipping thresholds
Moves: Revenue per visitor
Landing pages
- Advertorial vs. direct PDP
- Long vs. short form
- Hero by ad angle
Moves: Revenue per visitor
Navigation
- Shop by goal vs. by product
- Quick add on collections
- Mobile menu order
Moves: Product views
Checkout
- Checkout extension upsells
- Delivery estimates
- Post-purchase offers
Moves: Completion, AOV
Client numbers stay with the client. In an interview I'm happy to walk through specific tests, results included, under NDA.
Talk about a test