Almost every creator program looks successful in its first two weeks. Revenue spikes. Click-through rates are high. The launch thread on LinkedIn gets traction. The exec team sends a note to the CFO. Then the 90-day mark arrives, the cohort doesn't come back, and the program quietly rolls into maintenance mode while the team starts planning the next launch.
This is the shape of failure that nobody in the industry likes to talk about — because the launch numbers are real and the teams worked hard to get them. But a creator program that converts a customer once and loses them is not a program. It is a paid promotion. The distinction matters, and it matters specifically at the 90-day boundary where durability either shows up or it doesn't.
The Metric Most Programs Don't Watch
Take any creator program that has been running for six months and ask this: what is the 90-day repeat purchase rate for customers acquired through each creator? In most cases, the answer is "we don't measure that." In some cases, the team will produce a single blended number that treats every creator the same. In the rare case where the number is tracked per creator, a clear pattern almost always emerges — two or three creators produce cohorts that come back at 2–3x the site average, and the rest produce one-and-done customers.
This is the metric that separates a creator program from a creator promotion. If the 90-day repeat rate is below the site baseline, the program is acquiring customers who don't value the brand — they valued the creator's recommendation that one time. That is not a durable commerce relationship. It is a paid endorsement with a worse CAC than a Meta ad.
The 90-day test
A creator program passes the 90-day test if the customer cohorts acquired through top creators have a 90-day repeat purchase rate at or above the brand's site baseline. If they don't, the program is not compounding — it is extracting.
Why Launch Numbers Are Misleading
The launch-week numbers for a creator program are almost always inflated relative to the steady state. The creator has just mentioned the brand to their audience, the campaign has allocated budget to boost the initial surface, and the brand team is watching the campaign closely. Of course the first week looks great.
Three things compound to make launch numbers misleading. First, the audience that converts in the first 72 hours is the audience that trusts the creator most — the existing superfans. Second, the campaign budget is usually concentrated in the launch window, so the paid push is front-loaded. Third, the measurement window is short, which amplifies variance and makes any one data point look important.
By day 14, the superfans are already saturated. By day 30, the boost budget has decayed. By day 60, the campaign is competing with whatever the creator is posting about next. By day 90, the cohort has either returned or it hasn't — and that is the number that matters for whether the program is durable.
What Actually Predicts 90-Day Repeat Rate
When we look at creator programs across brands that measure cohorts carefully, three variables consistently predict whether a creator's cohort will pass the 90-day test.
Audience-product fit at the category level, not the creator level. A creator whose audience genuinely needs the category will produce a durable cohort. A creator whose audience can be persuaded once but doesn't structurally need the category will not. This is the single biggest predictor. It is also the one most brand teams get wrong, because the selection criteria tend to be follower count and vertical fit — not audience-category need.
Product selection depth. Creators who put meaningful thought into curating a small set of products (6–15 SKUs, clearly explained) produce cohorts that come back. Creators who link to the entire catalog or to a single hero SKU produce cohorts that don't. The curation is the trust signal. When the curation is weak, the attribution collapses to the creator's reputation, and reputation alone doesn't drive repeat behavior.
Continuity of creator presence on the surface the customer returns to. This is the one most teams miss entirely. If a customer was acquired through a creator and then the creator's presence disappears from the shopper's experience afterward — no mention in the email flow, no tag on the reorder page, no creator-specific re-engagement — the emotional thread that drove the acquisition is cut. The customer doesn't return, not because they didn't like the product, but because the brand relationship they formed was with the creator, and the creator is gone.
Why the Industry Avoids This Framing
If the 90-day repeat rate is the real test, why do so few teams use it? Three reasons.
Attribution doesn't support it. If creator identity lives only in a UTM, the data decays before the 90-day mark. If it lives only in an affiliate platform, the platform reports on conversions, not on customer LTV. The only way to run a 90-day cohort test on creator-acquired customers is to have creator identity written into the Shopify customer record — which requires an architectural decision most brands have not made. The full case for that architecture is in Why the Storefront Is Your Analytics Layer.
Organizational incentives are backward. The creator marketing team is usually measured on launch-week revenue or cost-per-order. Nobody is measured on the 90-day cohort. So nobody builds the dashboards, runs the queries, or designs the program to optimize for it. The metric lives in nobody's weekly review.
The honest answer is uncomfortable. Most creator programs, measured against the 90-day test, would show that the majority of creators are acquiring cohorts below the site baseline. That is not a comfortable finding for a team that has already spent money on creator contracts. So the metric is quietly avoided.
What a Program That Passes Looks Like
When a program passes the 90-day test, the data surface changes. Cohort retention curves start clustering by creator. Top creators produce cohorts that come back 30–50% above the site baseline. Bottom-tier creators produce cohorts that perform at baseline or below. The middle is a long tail that averages out.
Once this pattern is visible, the program's operating model changes. Budget concentrates on the top decile. Weaker creators are either given targeted catalog changes to improve fit or rolled off the program. The roster becomes smaller and more performant instead of larger and more diluted. This is what Cozy Earth's team built toward when they measured a 214% conversion rate lift and 67.37% higher AOV across 600+ storefronts — the aggregate number looked strong, but the internal view showed the distribution, and the operational decisions were made against the distribution, not the average.
Healf's architecture across 1,700+ creator storefronts, with 2,000+ collections and 1,200+ content assets, supports per-storefront measurement at the granularity needed to see the 90-day pattern. The ability to see the cohort is the precondition for operating against it.
How to Run the Test in Your Own Store
If you have creator attribution in your Shopify customer record, running the 90-day test takes an afternoon. The steps are concrete.
First, build a ShopifyQL query or a Shopify Segment that pulls all customers whose first order was tagged to a creator and whose first-order date falls in a 30-day window at least 120 days ago. This gives you a cohort of customers where the 90-day measurement window has completed.
Second, compute two metrics for each creator: the percentage of the cohort that placed a second order within 90 days of their first, and the percentage that placed a second order within 120 days. Compare each creator's percentages against the site baseline for the same period.
Third, sort by the 90-day repeat rate. The creators at the top of the list are the ones your program is actually built on. The creators at the bottom are either adjacent-category mismatches, curation-light links, or launched-and-left relationships.
The full workflow for building these queries is in How to Track Creator Attribution in Shopify Analytics, and the reference for how creator identity gets written into the Shopify order and customer record is in the Shopify order and customer tagging reference. You don't need a BI team to run it — the Shopify admin can produce the cohort numbers directly.
Why This Matters Beyond One Brand
The 90-day test is a forcing function for the industry. If brands started measuring creator programs against it consistently, a few things would shift.
Creator selection would move from follower count to audience-category fit. Curation would become a graded skill, not a checkbox. Programs would get smaller before they got bigger. Creator-side platforms would face pressure to expose cohort-level data, not just conversion-level. Brand-side teams would be evaluated on durability, not on launch volume. And the narrative around creator marketing — is it working, is it not, is it the future — would be anchored in a metric that actually measures whether the work compounds.
None of this requires a new technology. It requires an attribution architecture that can survive 90 days, a measurement discipline that asks the right question, and an organizational willingness to act on what the data shows. The first two exist. The third is the hard part.
The Counterargument
The strongest counterargument to the 90-day test is that some creator programs are intentionally short-cycle — product launches, limited drops, seasonal campaigns — where the 90-day metric is not the right measure. That is fair. The 90-day test is not universal. But the critique doesn't let the always-on programs off the hook. An always-on creator program that doesn't measure durability is making a structural bet without a mechanism to verify it.
The secondary counterargument is that some creators are genuinely top-of-funnel — their value is the awareness they generate, and the conversion happens later through other channels. This is also fair, and the right measurement for those creators is upstream (branded search lift, direct-traffic incrementality, assist-role attribution). But if you are going to carve out a creator from the 90-day test, you need an explicit alternative measurement, not a silent exclusion. The lack of a measurement is the failure, not the choice of which measurement to use.
What Changes When You Adopt the Test
The shift is more operational than analytical. The data has been there (or reachable) all along; what changes is how the team uses it.
Program reviews start with the cohort retention chart, not the launch revenue chart. Creator contracts are renewed or ended based on 90-day performance. Budget allocation follows the distribution. Creator-facing communication shifts from "here are your top products" to "here is how your cohort is performing." And the entire program starts to look more like a performance marketing channel with long feedback loops, and less like a PR campaign with a conversion tag.
Teams that make this shift tend to discover that their best-performing creators are not the biggest names. They are the creators whose audiences needed the category and who put real thought into the curation. These creators tend to be smaller, cheaper, and more willing to go deep on the catalog. The economics shift along with the metric.
Frequently Asked Questions
Is 90 days the right window for every brand?
For most DTC brands with a 45-to-90-day purchase cycle, 90 days is reasonable. For subscription or consumable brands with shorter repurchase cycles, 30–60 days may be more appropriate. For durables with 6–12 month repurchase cycles, the test should extend to 180–365 days. Pick the window that matches the natural repeat rhythm of your category.
What if a creator's cohort is too small to be statistically meaningful?
If the cohort is under 30 customers, treat the repeat rate as directional rather than conclusive. For smaller creators, roll up multiple campaigns or extend the measurement window until the sample is meaningful. Don't make final decisions on thin data.
Does the test apply to creator-gifted programs (no commission, just product seeding)?
Yes, with a modification. Gifted programs often don't have a tracked URL or coupon, so attribution is harder. If you can identify the cohort through a UTM campaign on the creator's post or a custom segment, the test still works. If you can't, the program is operating without measurement and the durability question is moot — fix the attribution first.
How do I separate creator-driven repeat behavior from natural repeat behavior?
Compare the creator cohort's 90-day repeat rate to the site baseline for the same period. If the creator's cohort is significantly above baseline, the creator is adding durability. If it is below, the creator is acquiring weaker customers than the brand's average channels. The comparison against the baseline is the signal, not the absolute number.
What about customers who didn't come back within 90 days but came back at 180?
These count as non-repeat for the 90-day test but may still indicate durability on a longer horizon. Track both 90-day and 180-day repeat rates in parallel to avoid misclassifying slow cohorts as dead cohorts.
If most of our program fails the 90-day test, what should we do first?
Don't end the program — reshape it. Identify the creators whose cohorts are above baseline and double the investment there. For the creators below baseline, figure out whether the problem is audience-category fit (re-evaluate selection), curation depth (work with the creator on catalog), or continuity (check the email flow and post-purchase experience). Most programs have 2–3 durable creators hiding inside a larger roster that doesn't perform; the goal is to surface them.
How does this relate to affiliate commission tracking?
Affiliate tracking answers "how much revenue did each creator drive" — which is a commission question. The 90-day test answers "how durable was the customer base each creator acquired" — which is a program-health question. Both are useful. Don't conflate them, and don't use commission totals as a proxy for durability.
Can this test be automated?
Yes. Once the creator attribution is in the Shopify customer record, the cohort query can run on a schedule (weekly or monthly) and the output can feed a dashboard. Most teams that operate creator programs at scale have this automation in place — the test becomes a standing program review rather than a one-off analysis.
What if the problem is that the creator was great but the product is bad?
The 90-day test will reveal that pattern — a creator with strong audience-category fit producing low-repeat cohorts is usually a signal that the product doesn't keep the promise the creator made. That is the brand's problem to fix, not the creator's. Use the test to diagnose product-fit issues that are hiding behind acquisition numbers.
Is this metric biased against new-customer-heavy creators?
Slightly, in the sense that creators whose cohorts are entirely new-to-brand customers have less reinforcement from prior loyalty. But the test is actually useful here too — a creator with a heavy new-customer skew whose 90-day repeat is at baseline is valuable, because they are building the brand. A creator with a heavy new-customer skew whose 90-day repeat is below baseline is expensive, because they are acquiring customers the brand can't retain.
%201.png)
%201.png)
