Unboxing Amazon Ads / DSP
Amazon Ads has been on quite the meteoric 12 months, further expanding its ads solutions within Sponsored Ads / DSP, enriching its analytics through Marketing Cloud (AMC) and even making moves in the cloud space with AWS through initiatives like AWS Clean Rooms, Retail Media Ad Service & more recently RTB Fabric. But throughout all of this, the ads solutions, specifically its DSP, can be seen to lose out when compared to its competitor set.
Don’t get me wrong, Amazon deserves a lot of credit for a lot of the product + partnerships updates and the core O&O inventory/audience power their ad platforms provide. But there are still ultimately areas where if your job is to be in the platform daily running campaigns, it still remains a challenge. This blog will deepdive into some of these areas where the biggest challenges / weaknesses are on Amazon Ads.
To preface:
I do not have TTD stock
I do not have any bias towards a certain platform
I speak from direct hands on experience of Amazon, both in-platform and via API
I am ignoring any special alpha / beta functionality available to certain Amazon partners
I am talking from a global perspective vs an US only approach
Access
Much like any ad platform, the first step is getting access. And with Amazon, that is not as easy as it sounds, specifically around the application of vendor central accounts for advertisers and access to Amazon DSP.
Currently if you want to sell products on Amazon as an endemic brand, you have two options when it comes to fulfilment. This involves Seller Central or Vendor Central. The latter is the prized version, given Amazon acts as the fulfilment partner. But the challenge remains that in order to get your own Vendor Central account, it is invite only. Therefore 3rd parties exist to provide this to advertisers, which in tandem implicates how Sponsored Ads accounts are created.
Amazon DSP has also had its own challenges. Previously you had to be a very big brand or agency to gain access to it. Nowadays, Amazon is taking a move straight out of the Google playbook by pushing DSP seat creation onto certified resellers, all of which have varying levels of minimum spends to work with. The reseller model for sure is still in its infancy and likely Amazon can learn a lot from Google here over time. But tying this back to the creation of new Amazon DSP entities and contracting is still a bit of an inefficient process on the Amazon end if you ever try it for yourself. The one positive however is that Amazon Marketing Cloud (AMC) as a clean room is now more accessible to all as of October 2025.
UI
Ask any trader what is the main problem with Amazon DSP? 99% it will be the UI. From a senior stakeholder level, this may be something taken for granted but in reality, this is where the main action is. If you do a poll on who has the worst DSP UX, likely Amazon DSP would come out on top still. Whether it is the amount of clicks taken to do certain actions, the inconsistent nature of its hierarchy and just lack of functionality, it has been a pain point for a long time.
Amazon has made big strides to improve it, especially over the past 12 months where earlier in 2025 a brand new UX was released and with the emergence of Campaign Manager to unify ad products into a single space with a single login, the future looks bright. But don’t think this suddenly gives away the enterprise DSP access to all users, there are other contractual elements to enable this.
At the same time, cramming different things into a single place can also be a recipe for disaster. You can say this is where the likes of Pacvue / Skai have taken full advantage of, to provide more enterprise Amazon users the ability to run campaigns from a slightly nicer experience. Though in reality, the UI may be playing the stage not to humans but AI agents if we are to go by the latest Ads Agent announcement.
Scale
As with any ad platform, scale is a component that it must be graded on. For sure, Amazon’s inventory & data has significant power as well as reach. But it is also reminding yourself that Amazon DSP only really operates where the Amazon marketplace exists. Unlike a Google DV360 or TTD or actually most other DSPs which cater truly globally (with some exceptions), you cannot run a campaign in every market, even via 3P exchanges on Amazon DSP.
Then factor in that certain functionality will always remain tied predominantly to the US market, which makes sense as the biggest and aligns to what TTD / DV360 do too. But if you are looking for a truly global DSP, Amazon is not quite there yet nor likely will it be a core focus.
Cost
Of course for any ad platform, the actual cost to the end advertiser is important. For the Sponsored Ads side of things, this remains on the freemium model much like its Search competitors. But on DSP, this is where a lot of buzz has been focused on through 2025, especially when comparing Amazon to other leading DSPs.
For this it is important to break out the different types of costs as well as the buying strategy. Much like any DSP, there is a platform fee to use Amazon DSP. This can range from anywhere from 5% to 15%, with the latter more likely if you are being resold via a reseller taking an additional margin. But this platform fee is not always applied, depending on the type of buy. Primarily around the use of Programmatic Guaranteed (PG) or inventory like Amazon Prime, where this is often dropped completely or minimised. This is a tactic Google also leverages on DV360 for PG / YouTube. This take rate element is always going to be a consideration for O&O buying.
But at the end of the day, the majority of buying still sits on the open auction, where the DSP fee will be very much apparent. And then add on other fees like Audience (as we will cover shortly), suddenly that take rate % becomes no longer a strength of Amazon. Amazon also like Google have to abide to Digital Service Taxes (DST) in certain markets, meaning another chunk of budget disappears from a media budget.
Activation Targeting
The core power of Amazon Ads is in its ad targeting, specifically around the ability to serve ads on premium inventory alongside use of Amazon’s heavily commerce intent powered dataset. The latter is often compared to a Google In-Market / Affinity audience or Social Interest / Keyword targeting in competitor platforms. But there is 1 major difference here: Amazon charges an additional CPM for the use of this data. A charge that can get more expensive the more custom you go. Whilst it is understandable that Amazon wants to monetise their most premium features like this, the other implication is exactly where your ads end up serving, which for the majority of cases means back onto the Amazon O&O. Is this much different to Google where you ultimately see most delivery through its own SSP when targeting Google data? No, not really. But it is an important consideration to know when it comes to how much investment goes to raw media vs audience fees. Thankfully also Amazon have introduced multi-country audiences, to make it more efficient in targeting the same audience across multiple markets. This directly correlates to where exactly on Amazon O&O you end up serving. For example, did you know that a fairly significant amount of delivery occurs on the confirmation page / thank you page of Amazon for both Sponsored Ads / DSP? The very moment someone has bought something, ads are served to the same user, which is the subject of my next blog to look out for. Incrementality for sure is a question to consider here!
Other forms of targeting are still fairly new to Amazon DSP still, namely things like contextual targeting, brand suitability & even to some extent device targeting. Amazon have had to repurpose this to fit a modern DSP but also making it 1st party to themselves. This also translates to more standard concepts like the idea of an inclusion list or blocklist being a bit more of a clunky object to implement. The split between web & app here is quite rare to see compared to other DSPs, which has caused a fair amount of confusion for traders.
Finally it is worth talking about Geo targeting. Now credit to Amazon, this has gotten a lot better over the years & we are even entering a world where more granular location targeting can be leveraged. Historically this was a challenge due to the marketplace nature of Amazon. But location is one of the more important targeting levers nowadays, not only due to the challenges it faces from an IP address approach but more from a measurement standpoint in doing incrementality testing with geo. This is moving in the right direction for Amazon DSP but Sponsored Ads remains impossible to do geo testing with on a granular level, which is the complete opposite when comparing to Google Ads for Paid Search.
Reporting
For any ad platform, the insights of what exactly the campaigns are delivering and performing is critical to determining success. Amazon has plenty of reporting options, both via the UI and API. More recently, the omnichannel reporting suite is a testament to improving this within the platform. For the more advanced advertisers, this is where AMC as a data clean room really is now mandatory to be using, which is also helped by the fact it is now available to all Sponsored Ads advertisers & Agentic support for writing queries.
But reporting is also a bit of an Achilles heel for Amazon. In comparison to other DSPs, the level of report granularity is not the same & you are generally pushed towards exporting data vs looking at it in-platform. On the Sponsored Ads side, there are certain metrics that are seemingly only available via UI vs API, making it quite the challenge to automate reporting out of Amazon, especially as scheduled reports now sit behind additional authentication vs an attachment in an email.
A good example of Amazon’s questionable approach to reporting can be viewed in reaction to the Adalytics research around delivery of ads on CSAM websites. Much like most ad platforms, the reaction from Amazon was to release more transparency in where ads were being served, notably on an URL level. The problem however? The only way to pull this report is via an API that barely anyone knows exists or uses. And when you do, the data is aggregated due to “privacy”, which defeats the purpose of URL transparency. The same goes if you ever want to pull a detailed report on where exactly ads are going through Amazon Publisher Services (APS). You can’t either through UI or API, which should be a red flag.
To some extent the reporting future is clear: either buy into the agentic model where AI agents will likely do all the heavy lifting with prompts or go use AMC to understand what is going on. As a trader, detail is key and aggregation can often cause more questions to be left unanswered.
Log Level Data
With any ad platform, data & ultimately transparency is quite critical to understanding exactly what it does. We have already touched on some of the gaps in standard reporting that Amazon suffers from, but there is also the elephant in the room of log level data.
Log level data or event level data is the ability to see a single row represent a single impression with more characteristics. This is something offered today by DV360 (DT Files) & TTD (REDS). Whilst log level data can have its own flaws & is a level of data that most advertisers / agencies cannot handle easily, it still unlocks a huge amount of power for analytics / activation. And as of present day, Amazon DSP does not want to give this away in its raw form.
Instead you will be pushed towards playing around in AMC or some of their lesser known APIs. Which to some extent provides similar capabilities to log level data, which is comparable to Google pushing users to use ADH instead of DT Files. Though ultimately, this is a gap which in a privacy centric world is likely the reason for it. But for a mature marketer, I’m sure this is an area that hopefully one day Amazon looks at opening up.
Measurement & “Closed Loop Attribution”
Amazon’s measurement offerings on the surface sound absolutely incredible. The promise of linking Amazon’s authenticated ID graph to ad exposure either through attribution or through AMC led enrichment of 1st party data, there is no denying that. This has now supposedly hit 90% of addressability in the US market.
But like many things in measurement, you should have some reservations about it, especially when it comes to “closed loop attribution”. That is, the art of trying to deterministically match a person to an ad exposure. Amazon for sure can do it on their marketplace / Prime and even more recently with partners like Spotify. But can they truly do it on the open web in a deterministic manner? Absolutely not. This goes back to legacy IDs like 3rd party cookies or Mobile Device IDs. Even Amazon’s TAM / APS can try as hard as it wants to get a view of the open web, but unless Amazon is doing something unethical that neither Google or Meta ever did, it cannot claim to do attribution accurately here. Amazon actually attempted to do this when they acquired Sizmek as an adserver, which was a very smart move at the time and had a lot of similarities to Meta acquiring Atlas. But we all know how that ended up.
The other big gap with Amazon is their actual in-platform measurement capabilities. Granted DSP is still behind here vs others, but the fact you still cannot do a simple AB test without doing a hacky workaround that only 1% of Amazon users know how to do but is still flawed, is a problem. More advanced lift studies are still sitting behind closed doors for now vs available to everyone. Measurement is ultimately king and Amazon knows this from the moves they have made with AMC. It is now about the platform keeping up.
Endemic vs Non-Endemic
This one may be more of a controversial take. Amazon at the end of the day makes more money from Endemic brands i.e. brands that sell on Amazon. The non-Endemic brands i.e. brands that do not sell on Amazon is where 1) the incremental ads opportunity is for Amazon & 2) is the sole competition of TTD / DV360 etc to compete with. Whilst endemic brands may run a bit of YouTube or Social to supplement their Amazon sales, it is still largely impossible (with a few Prime exceptions) to optimise non-Amazon to an Amazon sale easily, even with tools like Amazon Attribution in the mix. On the other side, if you use the entire Amazon stack as an endemic brand, you can see a lot of success.
The non-endemic piece is the biggest question mark here, especially where we get to more outcome based goals. In part this comes down to a couple of things. The ad tracking is still lightyears behind its competition where the pixel (or AAT as part of Events Manager as it is now known) has only recently overhauled itself to make it somewhat usable. The previous iteration couldn’t even passback revenue as a parameter, which is quite wild for an ecommerce platform! This also translates back into reporting, where off-Amazon conversions have finally got to a better place. The server side / conversion API (CAPI) is finally moving but again when paired AAT feels like it needs another evolution to compete with other players. AWS surely is the answer to some of this for sure.
Most enterprise non-endemic brands are likely using an adserver like Google CM360 or Innovid, both of which are supported within Amazon DSP. But again there are some challenges with even the basic application of applying click macros in ad trafficking. And this 3rd party tracking is also not supported on all Amazon buys.
This also nets out with some of the product functionality, which is pretty much tailored towards endemic brands (as you would expect). But you can argue this is where the competitor DSPs are able to take advantage of.
Closing Thoughts
Amazon Ads as a complete adtech / martech ecosystem is very powerful. The DSP is getting stronger by the year. And I hope it continues to, even in this new theme of agentic powered ads both from an activation / analytics standpoint but also generative AI creative approach.
But in order for Amazon DSP especially to take the crown away from DV360 / TTD or even a Yahoo DSP, there is still plenty of room for improvement and transparency. Amazon could truly differentiate itself by making their newer AI campaign types in Performance + / Brand + actually transparent in comparison to a Google PMAX approach. The true understanding of whether that works will be if non-endemic outcomes do see an uplift either through attribution or through incrementality testing, as well as the ultimate end cost to the advertiser.