Product reports rank items by revenue, units or margin. They answer what sells. They do not answer a question that matters more for a store that depends on repeat orders: when a new customer’s first order contains this product, how likely are they to come back, and what are they worth afterwards?
Products that score well on that second question are often called gateway products. Eric Davis of Little Stream Software, whose Repeat Customer Insights app popularised the term, notes that gateway products are often solid sellers rather than bestsellers, which is why regular product reports miss them. The opposite exists too: popular products whose buyers almost never return. If you spend ad money on the second kind and starve the first, your acquisition numbers can look fine while your customer base stays shallow.
What the difference looks like
Two published examples show how large the gap can be, and why the answer depends on when you look. Stormly, a product analytics vendor, describes a skincare brand where buyers of a $29 sampler returned within 28 days at a 62 percent rate, against 31 percent for buyers of a $65 full-size serum. The hero products that received most of the ad budget were not the best first purchase.
Common Thread Collective’s analysis of the cosmetics brand Bambu Earth adds a twist. Sample Kit buyers repeated faster: 18.7 percent within 30 days, against 14.2 percent for the Full Size Kit. Over the full customer lifetime the order flipped, with 42.9 percent of Full Size Kit buyers repeating against 37.8 percent for the Sample Kit, and the curves crossing just before 75 days. Full kit buyers simply had more product to use up first.
Build the first-order table
You need one row per customer: what they bought first, when, how they arrived, what discount they got, and what they did in a fixed window afterwards. On WooCommerce, the analytics tables have everything except channel. wc_order_stats holds one row per order with a customer_id that covers guests as well as accounts, and wc_order_product_lookup holds the line items. The query below picks each customer’s first order and measures the next 180 days.
-- One row per customer: first order, its products, and 180-day follow-up
WITH orders AS (
SELECT order_id, customer_id, date_created, net_total
FROM wp_wc_order_stats
WHERE parent_id = 0
AND status IN ('wc-completed', 'wc-processing')
),
ranked AS (
SELECT o.*, ROW_NUMBER() OVER (PARTITION BY customer_id
ORDER BY date_created, order_id) AS n
FROM orders o
),
firsts AS (
SELECT customer_id, order_id AS first_order_id, date_created AS first_date
FROM ranked WHERE n = 1
AND date_created < NOW() - INTERVAL 180 DAY -- only closed windows
)
SELECT f.customer_id,
f.first_date,
GROUP_CONCAT(DISTINCT l.product_id) AS first_products,
SUM(CASE WHEN r.n > 1 THEN 1 ELSE 0 END) AS later_orders_180d,
COALESCE(SUM(CASE WHEN r.n > 1 THEN r.net_total END), 0) AS later_revenue_180d,
MIN(CASE WHEN r.n = 2 THEN DATEDIFF(r.date_created, f.first_date) END)
AS days_to_second
FROM firsts f
JOIN wp_wc_order_product_lookup l ON l.order_id = f.first_order_id
LEFT JOIN ranked r ON r.customer_id = f.customer_id
AND r.date_created <= f.first_date + INTERVAL 180 DAY
GROUP BY f.customer_id, f.first_date;
Two details matter. The 180 DAY filter on first orders keeps only customers who have had the full window to come back; including recent customers drags every repeat rate down, the same censoring problem we covered in our guide to time between orders. And the join to line items multiplies the later-order counts when a first order has several products, so in practice you either aggregate the follow-up in a separate subquery or assign each customer one first product (the highest-value line, or the product category). Magento has the same structure in sales_order and sales_order_item, with customer_email as the safer key for guests.
Add two more columns before you analyse anything: the acquisition channel of the first order (WooCommerce 8.5 and later stores it as order attribution meta) and the discount on the first order from wc_order_coupon_lookup. You will need both in a moment. Then roll the table up by first product or category.
| First product category | New customers | Repeat within 90 days | Repeat within 180 days | Median days to 2nd order | Revenue per customer, days 1 to 180 |
|---|---|---|---|---|---|
| Starter kits | 1,240 | 21% | 34% | 48 | $61 |
| Refill packs | 980 | 29% | 38% | 36 | $58 |
| Single accessories | 2,110 | 9% | 15% | 71 | $19 |
| Gift sets | 760 | 6% | 12% | 94 | $22 |
| Sale items | 1,430 | 8% | 13% | 63 | $17 |
Four traps that produce false gateway products
1. Small groups. A product that was the first purchase for 40 customers, 12 of whom came back, has a 30 percent repeat rate with a 95 percent interval of roughly 16 to 44 percent. That range covers most of your catalogue. Set a minimum group size before ranking (a few hundred customers per group is a reasonable floor) and roll smaller products up to their category.
2. The window. As the Bambu Earth curves show, a 30-day view and a lifetime view can rank products in opposite order. Pick the window that matches your payback target. If paid acquisition needs to pay back in 90 days, rank on 90 days, and keep the 180 or 365-day view next to it.
3. Who bought it, not what they bought. First products are not assigned at random. Some are pushed in paid social, some in email to people who already know you, some only appear in sales. Discounts matter too: Peel Insights summarises research showing that larger first-purchase discounts tend to raise order volume but lower average order value and lifetime value. When channel and discount differ between products, the product comparison can reverse once you split the data, a pattern known as Simpson’s paradox.
| Product A | Product B | |
|---|---|---|
| Email: customers, repeat rate | 400 at 40% | 100 at 45% |
| Paid social: customers, repeat rate | 100 at 20% | 400 at 22% |
| All customers | 500 at 36% | 500 at 27% |
4. Testing everything at once. With 60 products compared at the usual 5 percent significance level, three or so will look significantly better than average by chance alone. Any ranking of many products needs a correction for multiple comparisons, or you will promote noise.
Testing whether the differences are real
The customer-level table from the query above is a good fit for a standard set of tests. Repeated or not is a yes/no outcome, so first product against repeat is a chi-square test, with Cramér’s V to tell you whether a significant association is also a large one. Revenue after the first order is heavily skewed, with many zeros and a few large customers, so a rank-based test such as Kruskal-Wallis is safer than comparing means. Run the same tests with channel and discount band as the grouping column to see how much of the product effect they explain.
If you would rather not assemble this by hand, you can load the table into QuantumLayers, which runs chi-square with Cramér’s V, ANOVA with a Kruskal-Wallis fallback, and a Benjamini-Hochberg correction across every test, so trap four is handled for you. Its categorical tests take columns with 2 to 20 distinct values, which is one more reason to roll products up to categories first. Its guide to cohort retention tables covers the time dimension if you want to see the curves by first-order month as well.
Shopify stores get much of this from apps: Lifetimely, for example, segments cohorts by first item purchased, discount code, channel and several other dimensions. On WooCommerce and Magento you usually build it yourself, which has one advantage: you control the window, the margin definition and the minimum group size.
What to do with the answer
- Judge acquisition on later revenue, not first-order ROAS. Stormly’s skincare example ran a paid campaign for the sampler at close to break-even on the first purchase because its buyers came back. A product whose buyers return can carry a higher cost per first order.
- Feature gateway products to new visitors. Homepage slots, first-order offers and welcome emails can point at the products that create repeat customers, not just the ones with the best conversion rate.
- Fix or stop promoting the dead ends. A popular product with low repeat might have a quality issue, a missing refill, or simply attract deal hunters. Check return rates for it before cutting spend.
- Time follow-ups by product. Median days to second order differs by first product, so a single 30-day reminder fits some products and is far too early for others.
- Consider bundles as an entry point. NP Digital’s comparison across 148 brands reported that customers whose first purchase was a bundle had 27 percent higher lifetime value than the baseline. Test it on your own data before redesigning offers.
The analysis is observational, so the last step is an experiment. Pick the most promising gateway product, put it in front of a random half of new visitors for a few weeks, and compare the 90-day repeat rate of the two groups. That tells you whether the product makes customers come back or merely attracts the ones who would have anyway.
Sources
- Little Stream Software, Gateway products: products that heavily influence customer behavior
- Stormly, Customer Lifetime Value Analytics for eCommerce: The Product-Level View (May 2026)
- Common Thread Collective, First purchase, LTV and repeat rate
- Peel Insights, How First-Purchase Discounts Impact LTV
- NP Digital, Which offers lead to higher customer lifetime value?
- Eightx, Lifetimely for ecommerce: review
- QuantumLayers, Inside the QuantumLayers Statistical Pipeline
- WooCommerce, Installed Database Tables


