What item-level data adds
Session and page reports describe visits. Merchandising decisions are about products and the places they are shown: whether the homepage carousel earns its space, whether a category page sorts the right products to the top, whether a promotion banner sells anything. Those questions need data at item scope, and the events and parameters guide explains why: item-scoped parameters live inside the items array of the e-commerce events and are reported through the item dimensions, so a product-level question has to be asked of an e-commerce event, not of a page view.
The Google developer guide for e-commerce documents the events and their item fields, and the e-commerce
tutorial lists them in a reference table. For merchandising, three groups matter. The list events,
view_item_list and select_item, carry the list's identifier and name, with each item's position in the
index field. The promotion events, view_promotion and select_promotion, carry the promotion and the
creative and slot that showed it. The journey events, from view_item through add_to_cart and
remove_from_cart to purchase, carry the products with their category, brand, variant, price, quantity,
discount and coupon. Together they let each placement and each product be judged on what it led to rather
than on clicks alone.
Item-level reading has two prerequisites. The item identifier must be the same on every event, as the tutorial's items contract insists, or one product splits into several rows. And the categories and brand must be spelled one way, because the item reports group on the values exactly as they arrive.
Item lists and placements
A list is any set of products shown together: a category page, search results, a homepage carousel, a
"customers also bought" strip. The e-commerce tutorial notes that view_item_list and select_item power the
item list reports, which show which placements lead to views and purchases. For each list, four counts
matter: how often it was shown, how often a product in it was selected, how often a product selected from it
went into the basket, and how often it was bought.
| Measure | Events | What it tells merchandising |
|---|---|---|
| List views | view_item_list | Exposure: whether shoppers reach the placement at all |
| Selection rate | select_item over list views | Whether the products shown look worth clicking |
| Basket rate from the list | add_to_cart for items selected from the list | Whether the click matched what the shopper wanted |
| Purchases from the list | purchase items credited to the list | Whether the placement sells, not just attracts |
| Selections by position | select_item by index | How much of the list shoppers actually look at |
Two readings come up again and again. A list with a high selection rate and low purchases is attracting
curiosity rather than intent: novelty products, eye-catching images or prices that look different on the
product page. A list with steep decay by position shows that shoppers rarely look past the first row, so the
sort order is the merchandising decision, not the list's contents. Before trusting purchases by list, place a
test order through a known list and check in DebugView which list the purchase is credited to. Where the
list does not reach the purchase, the item fields allow the site to carry item_list_id and
item_list_name on the items in the basket and purchase events, which is the dependable fix.
When the list readings point at the category page itself, the guide to category and listing pages covers the fixes on the page, from what each product card shows to filters, sorting and the default order.
Internal promotions
Internal promotions are the banners, tiles and messages that advertise the shop's own offers. The developer
guide documents view_promotion for when a promotion is shown and select_promotion for when it is clicked,
with promotion_id, promotion_name, creative_name and creative_slot, and optionally the items being
promoted. The e-commerce tutorial lists them in its reference table and leaves them out of its steps, so a
short note on adding them belongs here.
Push view_promotion when the promotion is actually displayed, not when the page loads with the banner far
below the fold; the site can push when the element enters view, or the container's element visibility trigger
can stand in where the site cannot. Push select_promotion on the click, with the same four fields. Clear the
ecommerce object before each push, as the tutorial does for every e-commerce event, and add both names to
the trigger that feeds the e-commerce tag. Name promotions after the offer and creatives after the design, so
that two banners for one offer can be compared.
Read promotions by creative and slot: views, selections, and purchases that followed a selection. A slot is a position on the site, and the comparison that matters is usually two creatives in the same slot or one creative moved between slots. A promotion with many clicks and few orders usually sends people to a page that does not deliver what the banner promised, which is a landing page problem rather than a creative one.
Product funnels: view, basket, purchase
For each product and category, the journey events give three rates: views to basket additions (the
add-to-cart rateShare of sessions in which a product was added to the cart. at item level), basket additions to purchases, and views
to purchases. The ecommerce purchases report shows the counts by item; a funnel exploration with steps on
view_item, add_to_cart and purchase, each narrowed by item category as the
explorations tutorial describes, shows the same journey for a category with a
device breakdown. The conversion rate calculator turns the counts into
rates when you need them outside the interface.
Each pattern points somewhere different.
- Many views, few basket additions. The product page is not persuading: price, images, delivery or reviews. This is the territory of the product page optimisation guide.
- Many basket additions, few purchases. The product is being saved rather than bought, or something at checkout (delivery cost, stock, a size choice) stops it; a high cart abandonment rateShare of shopping carts created that did not become an order. concentrated on a few products is a product problem, not a checkout one.
- Few views, good rates. The product converts when found and is under-exposed; move it up lists or into promotions.
- High removals.
remove_from_cartby item shows products shoppers change their mind about, often a variant or size question the page did not answer.
Read categories over time as well as products. A category whose views hold steady while purchases fall has a range, price or stock problem; one whose views fall has an exposure problem, in navigation, search or promotion. When the product losing views or basket additions is a best-seller that is still in stock, the diagnosis for a best-selling product's sales falling splits the fall by source and checks each layer of its route in turn.
Baskets and products bought together
Basket questions sit between products and orders. Items per order, removals before purchase, and the share of orders containing a second category describe how shoppers build a basket, and they are the inputs to the levers the average order value guide covers. Average order valueAverage revenue per order. itself comes from the purchase event's value; the product mix behind it comes from the items.
Products bought together need the export. The BigQuery export basics guide describes the items array as a repeated record on each purchase row, next to the e-commerce record holding the transaction identifier, so pairs within an order are one self-join away:
WITH lines AS (
SELECT ecommerce.transaction_id AS order_id, i.item_id
FROM `project.analytics_property.events_*`, UNNEST(items) AS i
WHERE _TABLE_SUFFIX BETWEEN @start_date AND @end_date
AND event_name = 'purchase'
GROUP BY order_id, i.item_id
)
SELECT a.item_id AS item_a, b.item_id AS item_b, COUNT(*) AS orders_with_both
FROM lines a JOIN lines b ON a.order_id = b.order_id AND a.item_id < b.item_id
GROUP BY item_a, item_b
ORDER BY orders_with_both DESC;
Read the pairs against each product's own order count. Two best-sellers appear together often simply because both sell well; a pair that appears together far more often than their separate popularity suggests is a candidate for cross-sellingOffering a customer a related product alongside the one they are buying; upselling offers a higher-value version instead. placement, a bundle or a "goes with" list.
Price and discount fields deserve the same care. The item fields include price, discount and coupon, so
purchases can be split by discount depth or by coupon. Before comparing, confirm the convention the
developers used, because the e-commerce tutorial's convention is item price times quantity, with tax and
shipping sent separately; a site that mixes conventions across templates cannot compare prices across them.
From findings to merchandising tests
A merchandising finding is a hypothesis, not a result. A list whose purchases lag its selections suggests that different products, or a different order, would sell more; only a test shows whether they do. Turn each finding into a change with a predicted effect on a stated measure: purchases per list view for a carousel, revenue per visitorRevenue divided by the number of visitors in the same period. for a promotion that may pull sales from full-price products, orders per view for a re-sorted category.
Choose the measure closest to revenue that the traffic can detect. Selection rate moves quickly and is easy to win; purchases from the list are what the business wants. Run the change as an A/B test where the site's traffic allows it, following the A/B testing guide, and read the arms in analytics with the exposure event the A/B test analysis tutorial builds. Where traffic is too thin to test, make the change, note the date, and compare the list's measures before and after with a defined period on each side, accepting that the comparison is weaker.
Conversion rateShare of visits or clicks that convert. for the whole shop is the last place a merchandising change shows, and the noisiest. Judge each change on the placement it touched, then confirm that the shop-level measures did not fall.
A merchandising review table
A monthly review keeps merchandising analysis from being a one-off. The table below is the checklist; the example after it shows one review's findings.
| Question | Where to read it | Signal worth acting on | Typical action |
|---|---|---|---|
| Which lists earn their space? | Item list report; list events by list name | High selection rate with low purchases, or the reverse | Change products or sort order; test it |
| How far down a list do shoppers look? | select_item by index | Selections concentrated in the first positions | Put the best converters first |
| Which promotions sell? | Promotions report by creative and slot | Many selections with few orders | Fix the landing page or the offer |
| Which products are under-exposed? | Item views against item purchase rates | Good view-to-purchase rate on few views | Raise them in lists and promotions |
| Which products lose shoppers? | Item funnel; removals by item | High basket rate with low purchases, high removals | Review page, stock, variants, delivery |
| What goes together? | Export pairs query | Pairs bought together beyond their own popularity | Bundles, "goes with" lists |
| How are discounts used? | Items split by discount and coupon | Discounted sales replacing full-price ones | Narrow the offer; test the depth |
Product funnels and pair data are also evidence for a bigger decision than placement: the guide to choosing which products to put marketing behind uses repeat and cross-sell potential, alongside margin, demand and stock, to decide which lines lead the paid plan.
Marbletwig Toys reviews its lists and banners
Marbletwig Toys, an invented online toy shop, runs its first monthly review. Two lists stand out.
| List | List views | Selections | Selection rate | Purchases from the list | Purchases per selection |
|---|---|---|---|---|---|
| Homepage picks | 48,000 | 3,840 | 8.0% | 140 | 3.6% |
| Category: building sets | 31,500 | 1,890 | 6.0% | 265 | 14.0% |
The homepage picks win the clicks and lose the sales: the buyer had filled the carousel with new arrivals priced well above the shop's usual range, and shoppers clicked to look rather than to buy. The building sets category, sorted by best-sellers, produces nearly twice as many purchases from half as many selections. Selections by position show that 71% of homepage selections come from the first four tiles.
The promotions report tells a similar story. The "Rainy day bundle" creative in the hero slot drew 2,600 selections from 52,000 views but only 38 orders, while the smaller "Free gift wrap" creative in the second slot drew 900 selections and 96 orders. The bundle banner linked to a search results page with forty products; the gift wrap banner linked to a curated list. The team makes two changes: the homepage picks become a mix of best-sellers and new arrivals, tested against the current carousel, and the bundle banner links to a single bundle product page. Both are logged with their dates so next month's review can read the effect.
Common mistakes
- Reading selections as success. A list or banner that wins clicks and loses purchases is costing the shop the space it occupies.
- List names built from page addresses or free text. The list dimension fills with unique values and collapses into "(other)"The number of distinct values a dimension or field holds; high-cardinality dimensions, such as page paths with IDs, can exceed report row limits..
- Promotion views on page load. A banner nobody scrolled to is counted as seen, and every promotion's rate looks worse than it is.
- Inconsistent item identifiers. The same product as a name on one template and a SKU on another becomes two products in every report.
- Trusting purchases by list without a test order. If the list never reaches the purchase, the report credits nothing and nobody notices.
- Mistaking popularity for affinity. Two best-sellers bought together are not a bundle until the pair beats their separate popularity.
- Judging a merchandising change on shop-wide conversion. The effect is diluted by every visit the change did not touch; read the placement first.
- Comparing prices across templates that send different conventions. Check how price and discount are sent before splitting sales by them.
The GA4 hub lists the other GA4 guides, and the reports and dimensions guide explains the item scope, cardinality and "(other)" behaviour that shape every report used here.
