Most brands that sell on Amazon know the dependency number by heart. It is the share of net revenue that arrives through one channel, and it tends to sit higher than anyone planned. This case study follows a US pet supplements brand that reached roughly 80 percent Amazon revenue and then spent seven quarters deliberately pulling that number down.
The interesting part is not the destination. Plenty of brands say they want a bigger direct business. The interesting part is the middle: two consecutive quarters of falling total revenue, a paid acquisition budget that produced worse returns than the marketplace it replaced, and a fulfillment migration that briefly broke the delivery promise the brand had been selling on.
The brand asked not to be named, so figures here are reconstructed from its channel reporting and rounded. Treat the numbers as representative of the shape of the transition rather than as an audited financial record. The sequence and the decision points are what transfer to other categories, and they line up with the broader patterns covered in the modern brand playbook for retail and e-commerce.
In short
- Starting point: about $18m net revenue, roughly 80 percent through Amazon, with direct-to-consumer sales at 11 percent and independent retail at 9 percent.
- Trigger: the move began while the Amazon business was still growing, not after a suspension or a fee shock, which is what made the financing possible.
- The unlock was the offer, not the ads: a subscription program on consumable supplements pushed direct lifetime value high enough to justify paid acquisition that looked unprofitable on a first order.
- The cost was real: two quarters of total revenue decline, funded by a deliberately shrunk Amazon advertising budget and a working capital facility rather than by new equity.
- Where it landed: roughly 52 percent Amazon, 38 percent direct and 10 percent retail, with blended contribution margin up about 6 points and the direct channel still smaller than the plan assumed.
The starting position: channel mix, margin and dependency risk
The brand sold dog supplements and functional treats, a category where the product is consumed and repurchased on a predictable cycle. It had launched on Amazon in the late 2010s, scaled through review velocity and sponsored products, and treated its own site mostly as a brochure with a checkout attached. By the start of the period covered here it was doing about $18m in net revenue.
What the mix actually looked like
Channel share alone understates the problem. The more useful view is contribution margin by channel, because Amazon revenue arrives after referral fees, fulfillment fees, advertising and returns, and each of those is charged in a different place. When the brand rebuilt its reporting to put all four on the same line, the marketplace channel looked considerably thinner than the top line suggested.
| Channel | Share of net revenue | Approx. contribution margin | Customer data owned |
|---|---|---|---|
| Amazon (FBA) | 80% | 21% | No |
| Own site (one-time orders) | 11% | 34% | Yes |
| Independent pet retail | 9% | 26% | No |
Two things stand out. The direct channel was already the highest-margin business the brand had, and it was the smallest. That gap is the entire commercial argument for the migration, and it is the reason the finance team supported a plan that would reduce revenue in the short term.
Dependency risk in plain terms
Dependency risk is usually described in dramatic terms: account suspension, a listing hijack, a policy change that removes a claim from a label. Those happen, but the slower risk matters more. When one channel owns the customer relationship, it also owns pricing power over the brand, and it can raise the cost of reaching the brand’s own repeat buyers year after year.
The pet category makes this concrete. It is one of the most contested consumables categories in US e-commerce, with a large specialist player, the general marketplace and a growing set of brand-owned subscriptions competing for the same repeat purchase. Recent trading updates in the sector, including the pressure visible around Chewy’s quarterly results against a cooling pet market, show how quickly repeat-purchase demand can flatten while acquisition costs keep climbing.
Why the brand moved before Amazon forced it to
The decision that separates this case from the failures is timing. The brand started the migration during a quarter when Amazon revenue grew about 9 percent year over year. Nothing was broken. That is precisely why it worked.
The three signals that triggered the plan
The leadership team pointed to three measurements rather than a single event. First, advertising cost of sales on Amazon had risen from roughly 9 percent of channel revenue to about 15 percent across two years, without a matching lift in organic rank. Second, the brand’s own repeat-purchase rate on Amazon was invisible to it, so it could not tell whether it was buying new customers or re-buying old ones.
Third, and most decisive, the brand ran a small holdout test. It cut sponsored product spend on its three best-selling ASINs for 21 days and watched the units. Sales fell about 34 percent, which told the team that a large part of what looked like brand demand was in fact rented traffic. That number, more than any strategy deck, unlocked the budget.
Why a crisis is the worst time to start
Brands that begin this transition after a suspension or a sudden fee increase have to build a direct channel while cash is contracting. The build takes two to four quarters before it contributes, so the crisis-driven version is usually financed by cutting the marketing that would have made it work. Starting from strength meant the brand could treat the first two quarters of direct spend as an investment rather than a rescue.
This is a pattern worth checking against other published examples before assuming it generalizes. A useful discipline is to read any case study for the counterfactual rather than the narrative, which is the standard set out in our guide to what a retail case study should actually contain to be useful.
Subscription as the offer that made direct viable
The brand’s first attempt at growing direct was a straight price play: a 10 percent discount for buying on its own site. It failed. Customers who already trusted the marketplace for delivery and returns were not moved by a small discount, and the discount ate the margin advantage that made the channel worth having in the first place.
The second attempt changed the product rather than the price. Supplements are consumed on a schedule, so the brand built a subscription program around a 30-day and 60-day cadence, with a dosing calculator that set the interval by dog weight. The subscription was not offered on Amazon at parity, which gave the direct channel something the marketplace listing did not have.
What the subscription changed in the economics
| Metric | One-time direct order | Subscription (first 12 months) |
|---|---|---|
| Average order value | $34 | $29 per shipment |
| Orders per customer, year one | 1.4 | 5.1 |
| Gross revenue per customer, year one | $48 | $148 |
| Contribution after COGS, shipping and payment fees | $16 | $54 |
| Allowable acquisition cost at target payback | $14 | $46 |
The last row is the whole strategy in one number. A brand that can only afford $14 to acquire a customer has almost no paid channels available to it in a competitive consumables category. A brand that can afford $46 has several.
The retention detail that decided it
Subscription programs are often reported at signup rather than at survival, which flatters them. The brand tracked shipment-level retention instead, and found the drop-off concentrated between shipment two and shipment three, where the customer had enough product to feel oversupplied. Moving the default interval from 30 days to 45 days on the largest weight band cut that churn point materially and raised year-one shipments per customer.
Two smaller changes helped. A one-click interval delay inside the account area reduced outright cancellations, because customers who wanted to pause were previously choosing the only visible option. Adding a low-cost treat product as a subscriber-only add-on raised shipment value without a discount.
Paid acquisition math when Amazon no longer subsidizes discovery
On Amazon, discovery is bundled into the fee structure. Shoppers arrive with intent, the listing does the convincing, and the brand pays a referral fee on the outcome. Off Amazon, none of that is included. The brand had to buy attention, build the trust the marketplace normally supplies, and absorb the cost of everyone who did not convert.
The first two quarters looked like a failure
Blended customer acquisition cost in quarter one of the direct push came in around $61 against an allowable of $46. Meta prospecting carried most of the spend and most of the loss. The team’s initial instinct was to cut, which would have ended the program, so instead they broke the number apart by intent.
Branded search converted at a fraction of that cost but had almost no volume, since it only captured demand Amazon had already created. Non-branded search on category terms was expensive and slow. Paid social worked only where the creative led with the dosing problem rather than the product, which is a pattern that also shows up in how a small skincare brand scaled to nine figures, where problem-first creative outperformed product-first creative by a wide margin.
Where the cost curve finally bent
Three levers moved acquisition cost from about $61 to roughly $44 over two quarters. None of them were media buying tricks.
- Offer structure: a first-shipment trial size at $9 with the subscription attached converted far better than a discounted full-size unit, because it lowered the risk of a product the dog might refuse.
- Landing experience: the dosing calculator became the landing page rather than a feature buried in the product description, which qualified traffic before it reached a price.
- Email and SMS on the back end: recovering non-converting traffic and lapsed subscribers reduced the share of growth that paid media had to carry.
Attribution stopped being reliable
One practical problem deserves its own note. On Amazon, the brand could attribute a sale to a placement with reasonable confidence. Off it, the same customer might see a paid social ad, search the brand name a week later, click an email, and buy on a phone that never carried the original cookie.
The team stopped trying to reconcile platform-reported conversions and moved to a blended view: total new customers acquired in a period against total acquisition spend in that period, checked against a monthly incrementality holdout in two geographic markets. The blended number was worse than what the ad platforms claimed and better than what last-click reporting showed. Running the business off the platform numbers would have led the team to over-invest in retargeting, which the holdouts suggested was largely harvesting demand that would have converted anyway.
It is worth being blunt about what did not work. Influencer seeding produced content the brand could use but very little measurable direct revenue. Affiliate placements drove volume that overlapped almost entirely with branded search. Retail media networks were not a fit for a brand trying to leave a marketplace, although they remain a rational choice for brands staying on one.
Fulfillment: leaving FBA without wrecking delivery promises
Fulfilled by Amazon is not just a warehouse. It is a delivery speed, a returns process and a customer service layer that shoppers have learned to expect. The brand’s direct channel had been quietly free-riding on that expectation, and the migration exposed it.
The first cut was too fast
The team moved direct fulfillment from a small in-house operation to a third-party logistics provider over six weeks, ahead of a promotional push. Average delivery time went from about 2.6 days to 4.9 days, and the subscription cancellation rate in the following month rose by roughly a third. Recovering that took longer than the migration itself.
The second attempt split inventory across two nodes, one on each coast, which brought most of the customer base inside a two-day ground zone. Delivery time settled near 2.9 days at a fulfillment cost roughly 12 percent above the previous single-node setup. The brand treated that difference as a retention cost rather than a logistics cost, which changed who owned the decision internally.
FBA against a dedicated 3PL for a direct channel
| Dimension | FBA | Two-node 3PL |
|---|---|---|
| Typical delivery time to US customer | 1–2 days | 2–3 days |
| Subscription batching and interval control | Limited | Full |
| Custom packaging and inserts | Restricted | Unrestricted |
| Cost predictability | Published schedule, changes annually | Contracted, negotiable at volume |
| Customer data captured | None | Complete |
| Peak season surcharges | Applied by Amazon | Applied by carrier and 3PL |
Fee levels on both sides of that table move. Amazon publishes its referral and fulfillment fee schedules and revises them periodically, and carrier general rate increases reset annually, so any brand modeling this should pull current figures from Amazon’s published selling fee pages and its own carrier contracts rather than relying on figures quoted in an article. Peak surcharges in particular have been rising faster than headline rate increases, a dynamic covered in our analysis of why peak parcel costs outrun the published rate increase.
What the brand kept on FBA deliberately
The brand did not leave Amazon. It kept its three highest-velocity SKUs on FBA at full inventory and stopped defending the long tail there. That preserved the marketplace as a discovery and trial channel while removing the operational cost of stocking twenty variants in someone else’s warehouse. Several products were made direct-exclusive, including the largest subscription size, which gave repeat buyers a reason to move.
Two quarters of falling revenue and how it was financed
This is the part most case studies omit. Total net revenue fell in two consecutive quarters, by about 4 percent and then about 6 percent year over year, before returning to growth in the third. The decline was not a surprise, but a plan on a slide is easier to hold than a number in a board pack.
Where the money came from
The brand financed the transition from three sources, in order of size. The largest was a deliberate reduction in Amazon advertising spend on defensive branded terms, which released roughly $1.4m annually and cost less marketplace revenue than the team feared. The second was a working capital facility against inventory, which covered the double stock position during the fulfillment migration.
The third was a temporary reduction in new product development. Two planned launches were deferred by three quarters, which is a real cost that does not appear in any channel report. No new equity was raised, and the founders were explicit that a raise at that point would have been priced off a declining revenue line.
The inventory problem nobody plans for
Splitting fulfillment across FBA and a two-node 3PL meant holding the same SKUs in three places. Working capital tied up in inventory rose roughly 40 percent at the peak of the migration, and the brand briefly ran out of its best-selling size in the western node while sitting on surplus in Amazon’s network.
The fix was mechanical rather than clever: a weekly reallocation review, a minimum cover level set per node rather than in aggregate, and a decision to accept slightly higher stockholding cost in exchange for fewer stockouts during the period when the direct channel was still proving itself. Stockouts during a subscription launch are more expensive than they look, because a missed shipment is a cancellation trigger rather than a delayed order.
The metrics the board tracked instead of revenue
Holding a plan through a revenue decline requires a different scoreboard. The brand reported on four numbers monthly: direct subscription active count, blended contribution margin, allowable acquisition cost against actual, and Amazon revenue excluding advertising-driven units. Total revenue stayed in the pack but was explicitly de-prioritized for three quarters.
That last metric mattered most. By stripping out units driven by sponsored placements, the team could see whether underlying marketplace demand was stable while spend came down. It was, roughly, which validated the holdout test that started the whole program.
Where the channel mix landed and what would be done differently
Seven quarters in, the mix had moved substantially, though not as far as the original plan projected. The direct channel reached 38 percent rather than the 50 percent target, and the team now describes that target as having been set by ambition rather than by category math.
| Measure | Start | After seven quarters |
|---|---|---|
| Amazon share of net revenue | 80% | 52% |
| Direct share of net revenue | 11% | 38% |
| Retail share of net revenue | 9% | 10% |
| Blended contribution margin | 23% | 29% |
| Owned email and SMS list | ~31,000 | ~186,000 |
| Active subscriptions | ~2,400 | ~41,000 |
The three things the team would change
Asked what they would redo, the operators named sequencing rather than strategy. They would build the subscription program and the fulfillment network before spending anything on acquisition, because the first quarter of paid spend pushed traffic into an experience that was not ready to retain it.
They would also migrate fulfillment in a slower, region-by-region rollout instead of a single cutover, and they would keep a larger share of Amazon advertising running on non-branded terms during the transition. Cutting defensive branded spend was correct; cutting category discovery spend at the same time briefly reduced the top-of-funnel that the direct channel was still borrowing from.
What transfers to other categories
The mechanism here is repeat purchase on a predictable cycle, which is what made subscription economics carry the acquisition cost. Categories without that cycle need a different bridge, usually assortment breadth, service, or a local advantage. The regional pattern in a regional grocer that beat Amazon Fresh in five years shows one version of that alternative, where density and delivery windows did the work that subscription did here.
Fulfillment sophistication is also becoming easier to buy than to build, with specialist DTC providers competing for exactly this kind of migration, a trend visible in coverage of the next wave of DTC fulfillment mandates. Broader context on how brands sequence channel, offer and operations decisions sits in the modern brand playbook.
One closing note on scale. US e-commerce remains a growing share of total retail sales according to the quarterly series published by the US Census Bureau, which means channel migration decisions are being made against a moving base rather than a fixed pie. That growth cushioned this brand’s transition, and a flat or declining category would have made the two down quarters considerably harder to hold.
This article is general business information based on a single anonymized example, not financial, legal or tax advice. Fee schedules, carrier rates and marketplace policies change frequently, so verify current figures with the relevant provider, and consult a qualified advisor before making financing or channel decisions for your own business.
FAQ on moving off Amazon
How long does it take to meaningfully reduce Amazon dependency?
In this case it took seven quarters to move from 80 percent to 52 percent of net revenue, with the first two quarters showing a revenue decline. Brands with a natural repeat-purchase cycle tend to move faster than those selling one-time or considered purchases, because subscription economics fund the acquisition cost sooner.
Should a brand leave Amazon entirely?
This brand did not, and most in its position do not. Amazon remained about half of revenue and continued to function as a discovery and trial channel while the direct business carried the higher-margin repeat purchase. Full exits usually make sense only when marketplace economics turn negative or when a policy conflict makes the channel unworkable.
What is the single biggest mistake in this transition?
Spending on acquisition before the offer and the fulfillment experience can retain the customer. The first quarter of paid media in this case pushed traffic into a site with no subscription program and a slower delivery promise, which produced expensive customers who did not come back.
Does a discount work as a reason to buy direct?
It did not here. A 10 percent price cut failed to move shoppers who trusted the marketplace for delivery and returns, and it removed the margin advantage that justified the channel. Offers that are structurally unavailable on the marketplace, such as subscription cadence, custom sizes or bundles, tend to work better than price.
How much should a brand budget for the revenue dip?
This brand planned for two quarters of decline and got roughly 4 percent and 6 percent year-over-year drops. The safer planning assumption is two to three quarters, funded before the program starts, since a mid-transition funding gap usually forces cuts to the marketing that makes the direct channel work.
Is it better to keep FBA for the direct channel?
Multi-channel fulfillment through Amazon is a common bridge, and it preserves delivery speed while the brand builds elsewhere. The trade-offs are limited control over subscription batching, restrictions on custom packaging and inserts, and a cost structure the brand does not negotiate. Many brands use it during the transition and move off it once subscription volume justifies a dedicated provider.
What metrics replace revenue during the transition?
This brand reported active subscription count, blended contribution margin, allowable acquisition cost against actual, and marketplace revenue excluding advertising-driven units. That last figure is the one that shows whether underlying demand is stable while advertising spend comes down.
How do you know how much of your marketplace revenue is rented?
A holdout test is the cheapest answer. Cutting sponsored spend on a small set of top listings for a few weeks and measuring the unit decline gives a rough share of sales that depends on paid placement. In this case a 34 percent drop over 21 days was the number that justified the whole program.
Does this work for non-consumable categories?
The specific mechanism does not transfer, since it relies on a predictable repurchase cycle. Brands in durable or considered-purchase categories generally need a different bridge, such as assortment depth the marketplace cannot match, configuration and fitting services, or a regional delivery advantage.