The 2026 entry product benchmark
- Entry product predicts repeat, well beyond chance. Within-brand variation runs at a median 4.0 times the binomial noise floor, and 5 of 7 brands exceed three times. If the first product carried no information this number would sit near 1.
- It holds out of sample. Ranking products on one set of months and measuring them on months they never influenced, the best third still returns 115.7% of brand-average retention against 69.1% for the worst, with a rank correlation of 0.54. That is the test that rules out the ranking being circular.
- Acquisition is already concentrated into a few front doors. The median brand takes 17.6% of its customers through one product and 45% through five, whether or not anyone chose them deliberately.
- The conventional way of measuring this overstates it. Attributing a multi-product first order to every product in it counts one customer several times, inflating the median brand's repeat rate by 1.4 points and one brand by 12.3.
- It is not a claim about causation. This shows which doors better customers walk through, not that moving a customer to a different first product would change them.
This is first-party data. Every figure was computed from customer-level order history across 17 managed DTC brands, linking each customer's first order to the products it contained and to whether they returned within 180 days.
Nothing here is estimated, modelled, surveyed, or drawn from another publisher. The analysis required a purpose-built pull, because the aggregated tables Interconnections runs day to day record what was bought and what was earned but not the link between them.
Why we ran this
The previous benchmark established that a DTC customer's first order accounts for the large majority of what they will spend in their first six months. If that is true, then which product somebody enters on stops being a merchandising detail and becomes one of the highest-leverage decisions a brand makes, because it is effectively choosing the customer.
We could not answer it from the tables we already had. They record how many first orders contained a product, and separately how much revenue a product earned, but nothing joins a customer's entry product to their later behaviour. So we built a pull that does, across 17 brands and 8,403 products.
The answer is that entry product matters a lot, that most brands have very few front doors, and that the way this gets measured in practice systematically flatters it.
the product they enter on predicts whether they come back
Take every product with at least 100 customers who entered on it, in a brand with at least 3 such products. That is 68 products across 7 brands and 29,400 customers.
There is an obvious trap in measuring this, and it is worth naming before the numbers rather than after. If you rank products by their observed repeat rate and then report the repeat rate of the top and bottom of that ranking, you will get a spread even when every product is identical, because ranking on an outcome and then reporting that same outcome is partly circular. Simulating this study with the effect switched off entirely, and every product given its brand's average rate plus ordinary sampling error, still produces a best-to-worst ratio of 1.6 times and a gap of 5.5 percentage points. Any honest version of this finding has to clear that bar.
So the headline here is the out-of-sample test instead. Products are ranked using one set of entry months, then measured on the months they had no part in ranking. The ranking and the measurement come from different customers, which removes the circularity completely.
The separation holds. The best third of a brand's entry products returns 115.7% of that brand's average retention and the worst third 69.1%, on months that played no part in sorting them. The rank correlation between a product's repeat rate in one period and the next is 0.54.
Measured in-sample the effect looks larger: a 2.4 times ratio between a brand's best and worst entry product, and 10.4 percentage points between them. Both sit above the 1.6 times and 5.5 points that noise alone would produce, so they are not artefacts, but they are inflated by selection and the out-of-sample figures are the ones to quote.
The variance test points the same way. If entry product carried no information, each product would sit at its brand's base rate plus sampling error and the observed variance would match the binomial expectation. It does not: the ratio is a median 4.0 times the noise floor, with 5 of 7 brands above three times.
What this does not establish is causation, and the distinction matters for what you do next. We cannot separate a product from the kind of person who buys it, so this is evidence about which doors better customers walk through rather than proof that steering someone to a different first product would change their behaviour. For a budget or merchandising decision those amount to the same lever, because choosing what to put in front of people chooses who arrives. For a claim about causation they do not, and this data cannot make one.
most brands have very few front doors
Before asking which entry products retain best, it is worth seeing how few of them there are. Across 16 brands, the median takes 17.6% of all its new customers through a single product and 45% through five.
Catalogues in this sample run from a handful of products to several thousand, and it makes little difference: acquisition collapses onto a few items regardless. That concentration is rarely a decision. It is the residue of which products happened to get ad budget, rank in search, or sit on the homepage.
Put beside Finding 1, that is the practical problem. A brand has roughly five doors, they differ by a factor of 2.4 in the customers they produce, and almost nobody has checked which is which.
the usual way of measuring this overstates it
A methodological finding, and the reason every number above uses a narrower definition than you might expect. When a first order contains three products, the conventional approach credits all three with acquiring that customer. One customer, counted three times.
That is not neutral, because multi-product first orders are placed by different people than single-product ones and they come back more often. Measured both ways across 17 brands on the mature window:
Pooled, the conventional method reports 17.6% against 15.6% on the unambiguous subset. The per-brand picture is the one that matters, because the pooled gap is set by whichever brand has the widest baskets: the median brand is overstated by 1.4 percentage points, the middle half of brands by 0.3 to 4.1, and the worst-affected brand by 12.3.
33.4% of first orders in this sample contain a single product, so the clean subset is not a rump.
What you have been told, and what the data shows
| The common claim | What this data shows |
|---|---|
| Push whichever product acquires cheapest | Entry products differ by 2.4 times in retention. Cheapest to acquire and best to acquire are not the same question. |
| Product mix is a merchandising concern, not a growth one | The median brand takes 45% of its customers through five products, so mix is the acquisition strategy whether or not it is managed as one. |
| Our entry product analysis says our hero product retains fine | Crediting every product in a multi-product first order overstates the median brand by 1.4 points and the worst by 12.3. |
| Retention is fixed by the category you are in | These gaps are inside a single brand and a single category, between products on the same store. |
What is actually happening underneath these numbers
A product is not only an item, it is a filter on who arrives. A discounted starter bundle, a gift-oriented item and a considered flagship purchase attract three different kinds of buyer, and the differences in what those buyers do next are what this study measures.
That framing explains why the effect is so large inside a single brand. Marketing, service, delivery and product quality are broadly constant across a store, so most of what could vary between two customers has been held still. What remains is which door they came through, and it moves retention by a factor of 2.4.
It also explains why the concentration in Finding 2 is expensive. If five products bring in 45% of customers, and entry products vary this much, then a brand's entire retention profile is being set by a handful of merchandising and budget decisions that were mostly made for other reasons.
What to watch, and what it actually means
What we would actually do with this
Report cost per acquisition and 180-day repeat rate together, by entry product. Neither number decides anything alone, and almost every reporting setup keeps them in separate places owned by separate people.
Audit the five products that bring in most of your customers. That is where the leverage is concentrated, and in this sample the ordering of those five by retention rarely matched their ordering by volume.
Re-run any entry product analysis you already have on single-product first orders. If it credited every item in the basket, it was overstating by roughly 2.1 points.
Treat a change in entry mix as a change in forecast. Shifting budget between products moves the retention profile of everyone acquired afterwards, on a lag long enough that nobody usually connects the two.
Frequently asked questions
Does the first product a customer buys affect whether they come back?
Substantially, and by more than chance explains. Across 7 brands and 68 products with at least 100 customers each, Interconnections measured within-brand variation in repeat rate at a median 4.0 times the binomial noise floor. If entry product carried no information that figure would sit near 1. It also survives the harder test: ranking products on one set of months and reading them on months they never influenced, the best third returns 115.7% of brand-average retention against 69.1% for the worst.
Which product should we push in acquisition campaigns?
Not necessarily the one that converts cheapest. Interconnections found the median brand acquires 17.6% of its customers through a single product and 45% through five, so the entry mix is already concentrated whether or not anyone chose it. The question worth answering is whether those few front doors are the ones that retain, which this data says is often not the case.
How much does entry product change repeat rate in practice?
Out of sample, which is the number Interconnections would defend, the best third of a brand's entry products returns 115.7% of its average retention and the worst third 69.1%. Measured in-sample the gap looks larger, a 2.4 times ratio between best and worst product, but pure sampling noise alone would produce 1.6 times, so the in-sample figure flatters the effect and the out-of-sample one does not.
Is this just that some products are bought by better customers?
That is exactly what it is, and it is the point rather than a confound. Interconnections cannot separate the product from the customer it attracts, and for a merchandising or budget decision the two are the same lever: choosing what to put in front of people chooses who arrives. What this study does not claim is that changing a customer's first product would change their behaviour, which would need an experiment.
Why measure only first orders that contained one product?
Because attributing a multi-product first order to every product in it counts one customer several times and inflates the result. Interconnections measured that inflation directly: it inflates the median brand's repeat rate by 1.4 points, and the worst-affected brand by 12.3. Every headline figure here uses the single-product subset.
How long do you wait before judging whether a customer came back?
180 days. Cohorts acquired after 2026-02 are excluded entirely, because they have not had 180 days to return and including them would drag every repeat rate down. That leaves 49,987 customers across 17 brands.
What we would measure next
The honest gap is causation. Everything here is observational, and the experiment that would settle it is straightforward: for customers arriving on the same ad, vary the product they land on and follow both groups. That is a test a brand can run, and Interconnections would rather run it than argue about the observational version.
Beyond that, two extensions. Whether the effect survives holding discount depth constant, since entry products differ in how heavily they are promoted. And whether second-order product choice carries the same signal, which would say whether this is about the customer or about the catalogue.
What this study cannot tell you
It is not causal. The product and the person who buys it cannot be separated in observational data. This shows which doors better customers walk through. It does not show that moving somebody to a different door changes them.
The strong test rests on 7 brands. Products need at least 100 customers and brands at least 3 such products before the variance test means anything, which is a high bar for a small catalogue. Findings 2 and 3 use the full 17 and 16 brand samples.
Two brand-months are excluded as platform-migration imports. When a store migrates onto Shopify the import stamps every historical order with the import date, and this pull buckets on that date, so an unscreened run would treat a decade of trading as one acquisition month. A screen removes any brand-month whose order count dwarfs that brand's typical month or whose repeats are already over by day 30. The proper fix is to re-pull on the field that preserves the true order date, which is not yet done, so those months are dropped rather than recovered.
In-sample spreads are inflated by selection. Ranking products on an outcome and reporting that outcome produces a gap even with no real effect. The null figures are printed beside the observed ones throughout, and the headline uses the out-of-sample test instead.
Nothing beyond 180 days. Cohorts after 2026-02 are excluded so that every customer counted has had the full window.
Guest checkouts are invisible. Following a customer requires a customer record. Brands with heavy guest checkout have real repeat behaviour this cannot see.
Discounting is not controlled for. Entry products are promoted at different depths and this study does not separate the product from the offer attached to it.
Method
A purpose-built pull reads every order for 17 managed DTC brands, groups them by customer, and links each customer's first order to the products it contained and to whether they ordered again within 180 days. A customer counts as acquired in-window only if their customer record was created inside it, the same test the cohort study applies, so a long-standing customer's first order in this window is never mistaken for an acquisition.
Every headline figure uses first orders that contained exactly one product, which is 33.4% of them. Finding 3 measures what the alternative costs.
Censoring is controlled. Cohorts acquired after 2026-02 have not had 180 days to return and are excluded from every repeat figure, leaving 49,987 customers across 17 brands at a pooled repeat rate of 15.6%.
Products are indexed to their own brand's average before pooling, so the comparison is between entry products rather than between brands. No brand is named, no per-brand row is published, and every published figure aggregates at least 5 brands. Product names are never published: they would identify an advertiser immediately.
- The variance test compares observed spread in repeat rate against the binomial expectation for the same customer counts. A ratio near 1 would mean entry product carries no information.
- Repeat means at least one further order within 180 days of the first, at any value.
- Deciles are computed on products, not customers, so a high-volume product counts once.
This study follows directly from the DTC repeat purchase benchmark, which found that a customer's first order decides most of their early value. The work of choosing and testing what goes in front of people is creative strategy and production and paid media management, the post-purchase side is email and retention marketing, and the creative companion study is the Meta creative fatigue benchmark. A growth diagnostic is how an engagement usually starts.