We started with a fairly simple question: can a brand’s presence on Amazon improve the performance of its own e-commerce website… or vice versa?
We did quite a bit of research, mostly in the United States and to a lesser extent in France, and the answer is less comfortable than you might expect.
Yes, Amazon can reassure a shopper who discovers a brand on its own website. But traffic can move in the opposite direction too: a Meta campaign designed to send people to the brand’s website may ultimately lead them to buy on Amazon. Or from a retailer. Or on the marketplace of a store they already know.
The customer does not wonder which channel deserves credit in the marketing dashboard. They buy wherever feels most convenient, reassuring or advantageous.
For the brand, this is where things become genuinely complicated. A campaign can look mediocre in Meta while still influencing sales elsewhere. Conversely, an increase in Amazon sales during a campaign does not prove that the advertising caused it.
Between the two sits a concept that appears frequently in multichannel commerce analysis: the halo effect.
And I have to admit, the expression is appealing. Perhaps a little too appealing 😅
A Meta ad, an Amazon purchase
Consider an example.
Someone discovers a product in an Instagram ad. They click, browse the brand’s website and leave without buying.
In the website’s advertising data, that visit looks like a failure.
That evening, the same person searches for the product on Amazon. Their account, Prime delivery, reviews and a familiar checkout process are all there. They place the order.
The halo describes the gap between where demand is created and where the sale is completed.
For Meta, the ad generated no sale on the website. For Amazon, the order may appear as a branded search or an organic sale. Yet without the original ad, that person might never have heard of the product.
The halo describes this movement between the place where demand is created and the place where the sale is completed.
Several reports now attempt to measure this phenomenon. Their findings are interesting, but far less definitive than some headlines suggest.
An analysis published by Fospha in 2026 covers data from 172 brands in North America and Europe. For each advertising platform, its model attempts to estimate sales completed on Amazon after being influenced by a campaign, even when that campaign originally directed consumers to the brand’s website.
When Fospha includes those sales in its calculation, the ROAS attributed to TikTok increases by 50%, while Meta’s increases by 46%. These figures do not mean that TikTok and Meta generated 50% and 46% of Amazon sales, respectively. They mean that the ROAS calculated by Fospha rises when it is no longer limited to orders recorded on the DTC website.
A hypothetical example makes this easier to understand.
A brand spends $100,000 on TikTok. Sales completed on its website and attributed to the campaign total $200,000. The visible ROAS is therefore 2.
If the model estimates that TikTok also influenced $100,000 in sales completed on Amazon, the recalculated ROAS rises to 3. Moving from 2 to 3 is indeed a 50% increase, but it tells us nothing about what share of total Amazon sales came from TikTok.
$200,000 in attributed sales for $100,000 spent.
$300,000 in sales estimated as influenced for the same investment.
+50% calculated ROASThe chart illustrates the calculation in the article. It does not measure the share of Amazon sales attributable to TikTok.
The phenomenon itself is credible. The percentages deserve more caution.
Fospha sells the measurement solution used in the report. The company does not provide enough detail to determine how many brands were included in each platform-specific calculation, reproduce the allocation of Amazon sales or assess the margin of error. The report also refers to “influenced” revenue without demonstrating that all of those sales were genuinely incremental.
Our takeaway is therefore that an effect may exist beyond the website data. We would treat that conclusion with much greater confidence than the specific percentages reported.
An influenced sale is not yet an incremental sale
This distinction forced us to revisit a word we had been using a little too casually: incrementality.
An Amazon sale may have been influenced by TikTok without having been caused by TikTok.
Suppose a loyal customer already planned to reorder their usual product. They see a TikTok ad in the morning and place the order on Amazon that evening. The ad played a role in the journey, but the purchase might have happened anyway.
The sale was influenced. Whether it was incremental remains to be proven.
A sale is considered incremental when it would not have occurred without the action being studied. One way companies can estimate this is through geo testing.
Two groups of regions are selected whose Amazon sales had previously followed similar trends. The first group is exposed to the TikTok campaign, while the other serves as a control.
Suppose sales rise from $100,000 to $130,000 in the exposed regions. In the control regions, they also increase, reaching $110,000 because of seasonality, existing brand awareness or other commercial activity.
The control group helps estimate what would probably have happened without the campaign.
Without the campaign, the test regions would probably have followed a trend similar to the control group. The increase attributable to TikTok is therefore not $30,000, but approximately $20,000.
That $20,000 is the estimate of incremental Amazon sales generated by TikTok.
This calculation remains an estimate because two regions are never perfectly identical. Differences in inventory, competition, promotions or local behavior can affect the result. A well-designed test tries to reduce those gaps without pretending they can disappear entirely.
What the TikTok tests actually tell us
The report Deciphering TikTok’s True Incremental Impact, published by WorkMagic, draws on more than 100 lift tests conducted during the six months before publication, including 16 TikTok tests.
In 94% of those TikTok tests, last-click attribution underestimated the effect measured by the test. The report also estimates that, for the omnichannel brands studied, an average of 33.6% of TikTok’s impact materialized on Amazon.
That last figure, however, is based on only eight advertisers. Treating it as a market benchmark would therefore be unwise.
WorkMagic also sells an incrementality measurement solution. As with Fospha, that fact does not invalidate the work, but it does justify viewing the findings with some distance. The report does not publish all the data needed to independently verify each test, compare product categories or calculate the dispersion around the reported averages.
Another case study, focused on Nordic Naturals, was published by TikTok for Business. The brand used a geo test to measure the effect of its TikTok campaigns across all sales, including those completed on Amazon. The test estimated an incremental ROAS of 1.94 and identified part of the effect on Amazon even though the campaign directed users to the brand’s website.
This case shows that movement between TikTok and Amazon was observed for Nordic Naturals during that campaign. It does not predict what would happen for another company. It is also worth remembering that the study was published by the advertising platform involved.
After reviewing this work, our position is fairly simple: the halo between advertising, DTC and marketplaces is not merely an intuition. Several tests have observed it. Its size, however, varies by brand, and the most spectacular results often come from companies selling the tools or platforms being studied.
The phenomenon deserves to be measured. The percentages do not yet deserve to be treated as general benchmarks.
The halo does not move in only one direction
Much of the content we found presents the DTC website as the place where the brand creates demand and Amazon as the place where it captures it.
That interpretation works in some cases. It becomes fragile as soon as retailers enter the picture.
Imagine a sporting-goods brand sold through its own website, Amazon and specialty stores.
A customer may discover the product through a brand campaign and buy it from a nearby retailer. But they may also receive advice from a salesperson, try the product in a store and later order it on Amazon. Or they may see it on a retailer’s website, search for the brand on Google and complete the purchase on the DTC website.
The retail network is therefore not always a passive beneficiary of the brand’s marketing. It can create visibility, trust and demand itself.
Amazon also plays several roles. The platform can capture demand created elsewhere. Its presence in search results, reviews and recommendations can also contribute to discovery or reassure a shopper before an order on the brand’s website.
DTC website
Can create demandContent, campaigns and brand experience.
Can convertA direct relationship and control of the journey.
Amazon
Can build trustReviews, availability, delivery and buying habits.
Can also drive discoveryInternal search and recommendations.
Retailers
Can recommendAdvice, demonstrations and local presence.
Can convertIn stores or on their own marketplaces.
The same consumer may pass through several of these roles before completing a purchase.
We found no evidence that simply being present on Amazon automatically improves the ROAS of Meta, TikTok or Google campaigns that direct traffic to the DTC website.
It is a plausible hypothesis in some categories, particularly when Amazon reviews reinforce trust in a relatively unknown brand. But it must be tested, because Amazon may just as easily divert purchases that would have generated a higher margin on the brand’s website.
An incremental sale is not necessarily profitable
Incrementality answers one question: would this sale have existed without our action?
On its own, it says nothing about profitability.
A campaign can generate $20,000 in incremental sales and cost more than that once advertising spend, discounts, product costs, commissions and logistics are deducted. The sales are incremental, but the company loses money on the campaign.
That is when customer acquisition cost, or CAC, becomes important. It is the amount spent to acquire a new customer. Ideally, a brand should focus on incremental CAC, calculated from customers who truly would not have been acquired without the campaign, rather than relying only on the number of customers Meta, TikTok or Google claim in their dashboards.
The analysis cannot always stop at the first order. For a brand with repeat-purchase potential, an acquisition that loses money initially can become profitable if the customer buys again.
This is where lifetime value, or LTV, comes in: the value generated by a customer throughout their relationship with the brand. It should be calculated from the contribution actually retained, not merely from hoped-for future revenue.
A customer who costs $100 to acquire and generates only $70 in contribution on the first order initially loses the company $30. If later purchases generate another $150 in contribution, the acquisition becomes profitable over time.
That repeat behavior still needs to be observed. Otherwise, a theoretical LTV can be used to justify almost any acquisition cost.
The distinction between channels also matters. On its own website, the brand can identify the customer, track orders and develop retention. On Amazon or through a retailer, it controls less of the relationship and has less data. A first sale therefore does not necessarily have the same future value depending on where it occurs.
Halo
Where did demand move before the sale was completed?
Incrementality
Would the sale or the customer have existed without the marketing action?
Contribution
What does this acquisition actually contribute, now and over time?
Stopping at the halo can make a campaign look better than it is. Stopping at the first order, on the other hand, can lead a company to cut an investment that becomes profitable over time. Both matter, without inventing future loyalty to make the numbers work.
That is the challenge with the halo: it describes a journey, not its economic value.
Margin brings some clarity
Comparing DTC, Amazon and wholesale revenue is not enough, because one dollar of sales does not have the same value in every channel.
On its own website, the brand retains the retail price but bears the cost of acquisition, payment processing, order preparation, shipping, returns and customer service.
On Amazon, it pays commissions, advertising, logistics or storage fees and, depending on its model, various costs related to FBA or its Vendor relationship.
In a retail network, the invoiced price is often far below the price paid by the consumer. On the other hand, the sale may require less direct advertising spend and transfer part of the logistics, sales advice or service burden to the distributor and retailer.
This is fundamentally a management accounting exercise. For each channel, start with the revenue actually retained, then subtract the costs directly caused by those sales:
- discounts and refunds;
- cost of goods sold;
- commissions and payment fees;
- shipping, storage and fulfillment;
- returns;
- attributable marketing expenses;
- channel-specific sales costs.
The amount shows how much the channel contributes. The rate shows what share of revenue the brand retains.
The result is a contribution margin. It allows the company to compare what each channel contributes before shared corporate overhead.
This financial calculation remains incomplete if it ignores the burden placed on the organization.
A channel can produce a reasonable contribution while consuming enormous amounts of time: channel-specific content, catalog management, negotiations, inventory forecasts, Amazon cases, customer service or adapting to partner promotions… the list is endless.
Conversely, a retail network may show a lower unit margin while providing local presence, demonstrations, recommendations and credibility that the brand would struggle to fund on its own.
Ideally, two elements should therefore be tracked separately. First, each channel’s contribution margin: net revenue minus product costs, discounts, returns, commissions, logistics and directly attributable marketing expenses. Both the amount generated and the percentage retained matter.
Then add a straightforward assessment of the effort required: team time, complexity, inventory commitment and reliance on partners.
The goal is not to produce one perfectly precise number, but to understand what each channel contributes, what it requires and where the next dollar invested will create the most value.
Practical examples
But all of this must be adapted to each company’s business model 😅
To see why, consider two companies in very different situations.
DTC is already the largest channel
Priority question: where will the next dollar or hour of work produce the greatest marginal contribution?
DTC remains underdeveloped
Priority question: does that 10% reveal weak demand, or simply a lack of investment in the channel?
The first brand already has its entire infrastructure in place. Its DTC website accounts for 60% of revenue, its Amazon operation with FBA is running, its logistics provider can handle wholesale orders and it has a retail network. It knows the contribution margin of each channel.
It might be tempted to choose the channel with the highest margin today and concentrate its investment there. Right?
No.
Its main challenge is to understand what the next dollar invested in each channel would contribute. The channel that has historically been the most profitable is not necessarily the one with the greatest remaining upside.
A DTC investment can improve customer knowledge and support Amazon or retailers. An Amazon campaign can increase brand visibility, but it can also capture sales that would have generated more margin on the website. An initiative for retailers can support regions where the brand could not acquire customers directly at a reasonable cost.
The company must therefore compare marginal contribution: what happens if the brand adds $10,000 in budget, one month of work or additional inventory to this channel?
The second company grew primarily through a B2B model and generates approximately 65% of its revenue from distributors and retailers, 25% on Amazon and only 10% through DTC.
At first glance, this mix seems to show that the direct website matters very little. Except that the company is doing almost nothing to develop it, even though the website and logistics are already in place.
That 10% does not necessarily measure market appetite for DTC. It also reflects how little effort the company has invested in the channel so far.
For this company, launching a limited DTC test would make more sense than a sweeping strategic shift. It could select a product line, region or audience, set a budget and observe what happens without destabilizing prices or existing partners.
Of course, the test could fail. Acquisition costs might be too high, or customers might prefer to buy from retailers they already know.
But it could also reveal previously untapped direct demand, with a stronger margin and new opportunities for retention.
Either way, the company would finally have an observation instead of drawing conclusions from a channel it never truly tried to develop.
Sell-through: what can you measure when retailers do not report their sales?
To calculate sell-through, the brand needs to know the retailer’s sales to end customers.
If a brand ships 100 products to a retailer and the retailer sells 60 during the period, sell-through is 60%.
known sell-in
reported sell-through
Without point-of-sale data and inventory levels, the brand knows only its sell-in.
The brand generally knows its sell-in: the 100 products it sold to the retailer. It only knows sell-through if the partner reports its sales and inventory levels.
When a distributor sits between the brand and retailers, visibility decreases further. The brand knows what it sold to the distributor. The distributor knows what it shipped to stores. But no one necessarily knows how many products consumers ultimately purchased.
It would therefore be misleading to place DTC customer sales, Amazon orders and invoices to distributors in the same table as though they were perfectly comparable data.
That does not make management impossible. It simply requires naming each metric correctly.
Each week or month, a brand can consolidate:
- DTC sales, net of discounts, refunds and returns;
- sales on Amazon and other marketplaces;
- sales to retailers and distributors, clearly identified as sell-in;
- contribution margin by channel;
- marketing spend, promotions and stockouts;
- sell-through data provided by selected partners, without automatically extrapolating it to the entire network.
This dashboard will not prove every halo effect. It will nonetheless give the company a consolidated view of the business and prevent Amazon, DTC and wholesale teams from each working with their own version of reality.
Where does MMM fit in?
Marketing Mix Modeling, or MMM, uses a company’s historical data to estimate the share of sales variation associated with its different marketing investments.
For example, the model can compare weekly sales across all channels with spending on Meta, TikTok or Google. It also accounts for other factors that may influence demand, including promotions, pricing, seasonality, distribution coverage and stockouts.
For a multichannel company, its value lies precisely in not being limited to conversions recorded on the website. It can help detect that a campaign intended for DTC also appears alongside increased sales on Amazon or through retailers.
But MMM produces an estimate, not proof.
If advertising investment and Amazon sales rise at the same time before Christmas, advertising does not necessarily explain the entire increase. Seasonality, a promotion, better product availability or a retailer campaign may each have played a role.
Geo tests or holdout periods can show whether the model’s estimates hold up under observation. Without that calibration, an MMM can produce very precise-looking results while assigning too much weight to advertising.
The tool becomes most relevant when the company has sufficient history, budget variation and reliable data across channels. For a small or midsize business launching its first campaigns, a few well-designed tests will often be more informative than a sophisticated model fed with too little data.
So, should a brand sell everywhere?
A brand can spread its inventory, budget and team across five channels and manage none of them properly.
Yet choosing a single channel simply because it shows the best observable ROAS would be just as risky.
DTC gives the brand greater control over the experience, customer data, testing and retention.
Amazon and other marketplaces meet customers’ need for trust, comparison, availability and purchasing convenience.
Retailers and distributors can provide geographic reach, recommendations, physical presence and access to customers the brand would struggle to reach on its own.
The right mix depends on the product, the brand’s maturity, its margins, its organization and how customers actually buy.
Our position is this: performance does not come from accumulating channels. It depends on the role assigned to each one, pricing consistency and the brand’s ability to verify that its channels reinforce one another more than they cannibalize one another.
Before increasing a budget or closing a channel, we would examine four things:
- Does this channel create new demand or capture demand created elsewhere?
- How much financial contribution does the brand actually retain?
- How much work, inventory and resources does this result require?
- What would the overall business lose if this channel disappeared?
The last question is often missing from dashboards. Yet it can prevent a company from sacrificing a retail network because its brand impact is difficult to attribute, or from celebrating Amazon sales funded by a DTC campaign that does not cover its costs.
Customers will continue to buy wherever they choose. The brand’s job is to understand the role played by each point of sale, then decide where its next dollar and next hour of work will create the most value.
Not the most flattering ROAS.
The value actually retained.
View the main sources
- Fospha, The State of Retail Commerce 2026, a report based on data from 172 brands in North America and Europe.
- WorkMagic, Deciphering TikTok’s True Incremental Impact, a report based on more than 100 lift tests.
- TikTok for Business, Nordic Naturals case study, focused on measuring incrementality and sales completed on Amazon.
- Constantine Yurevich, The Myth of the “Halo Effect”, a critical perspective on margin leakage risks.