Open any ad-tech glossary and you will meet the same wall of capital letters. DSP, SSP, DMP, ADX, CDP, ATD, RTB. The definitions are usually correct and almost never useful, because they explain each acronym in isolation and the confusion was never about one acronym. It was about how they connect. People who have worked in programmatic for ten years still mix up a DSP and a DMP. Vendors use the words loosely. Beginners give up.
So here is the trick that makes it stick. The whole stack divides into two sides. There is a buy side, where advertisers spend money, and a sell side, where publishers earn it. Between them sits a marketplace. Every box in the ecosystem belongs to one of those three places, and once you know which side a tool is on, its job becomes obvious. That is the entire map. The rest of this post just walks it.
Origin: why a stack exists at all
In the early web, an ad was sold like a magazine page. A salesperson at a publisher agreed a price, a quantity, and a run date with an advertiser, and that was the transaction. It worked until it did not scale. By the mid-1990s there were too many publishers and too many advertisers for hand-negotiated deals to cover everything.
The first fix was the ad network. Networks appeared around 1996, with DoubleClick among the early ones, and they did something simple. A network bought unsold inventory from many publishers in bulk, packaged it, and resold it to advertisers. Instead of a thousand advertisers talking to a thousand publishers, both sides talked to the network. That solved volume but created a new problem: the network bought low and sold higher, and nobody on either end could see the markup.
The ad exchange, which emerged in the mid-2000s, replaced the bulk repackaging with a live auction. Rather than buying a block of impressions at a fixed rate, advertisers could bid on a single impression at the moment it was created. Real-time bidding made that practical at scale once Google relaunched the DoubleClick Ad Exchange with RTB built in in September 2009. The instant a single impression became an auctioned item, both sides needed software to handle the speed. Advertisers needed a console to bid across many exchanges at once, so the demand-side platform was born. Publishers needed a tool to manage which auctions their inventory entered and at what floor price, so the supply-side platform was born. Audience data needed a home, which became the data management platform. Each layer of the modern stack is a response to a specific limitation of the layer before it. That history is why nothing in ad tech looks designed. It accreted.
Present: the two-sided map
Picture two columns with a marketplace in the middle.
The buy side: the demand-side platform. A DSP is the advertiser's console. Amazon defines it plainly as software for advertisers that provides automated, centralized media buying from many sources. You load a campaign, set targeting, set a budget, set a goal, and the DSP bids on individual impressions across dozens of exchanges on your behalf, thousands of times a second. The Trade Desk, Google's Display & Video 360, and Amazon DSP are the three that matter most. By one measure of US programmatic activity, DV360 sat near 47 percent of the market in early 2025, with Amazon DSP around 20 percent and The Trade Desk around 19 percent, so three platforms account for the overwhelming majority of buy-side spend. The DSP is on the advertiser's side. Its only job is to spend a budget well.
The sell side: the supply-side platform. An SSP is the mirror image. It is the publisher's console: software that manages and sells a publisher's ad inventory, connecting them to many exchanges, DSPs, and ad networks at once. Where the DSP tries to buy attention cheaply, the SSP tries to sell it dearly. It sets floor prices, decides which buyers to expose inventory to, and runs yield management so a publisher squeezes the most revenue out of every slot. Magnite, PubMatic, Index Exchange, and OpenX are the large independents. Google's Ad Manager carries the equivalent function for a huge share of publishers. The SSP is on the publisher's side. Its only job is to earn the most per impression.
The middle: the ad exchange. The exchange is the marketplace where the two sides meet. It is the auction house. SSPs list inventory into it, DSPs bid into it, and the exchange matches them and clears a price, all inside the roughly 100 milliseconds a page takes to load. If you want that auction traced impression by impression, we walk through it in how an ad gets bought in 100 milliseconds. The exchange is neutral by design. It does not have a side. It takes a cut for running the auction, and exchange fees commonly run from 5 to 20 percent of the transaction depending on platform and volume.
The decider on each end: the ad server. This is the box people forget, and it is the one that actually delivers the ad. An ad server does not buy or sell. It decides which creative to show in a given slot and then serves the file and counts the impression. There are two of them, one per side. The publisher's ad server, with Google Ad Manager dominant, sits on the page and decides whether a slot goes to a directly sold campaign or out to auction. The advertiser's ad server, such as Google's Campaign Manager 360, stores the advertiser's creatives, picks which variant to show, and counts impressions and clicks independently of the publisher. Because the two count at slightly different moments, their numbers never quite match, and that gap has a name in the trade: ad discrepancy. If a tool's job is to deliver and tally the creative rather than price it, it is an ad server.
The audience layer: the data management platform. A DMP is not a buying tool. Amazon calls it a place to store collected insights so they can inform campaigns. It collects data, sorts users into audience segments such as "in-market for a car," and pushes those segments to a DSP so the DSP can target them. The DMP supplies the audience; the DSP does the buying. That single sentence resolves the most common confusion in all of ad tech, so it is worth pinning down properly below.
The confusing pairs, pulled apart
Three pairs cause almost all the trouble. Here is each one, settled.
DSP versus DMP. They sound similar and they work together, which is exactly why people fuse them. They are not the same kind of thing. A DSP is a buying platform; a DMP is a data platform. Lotame puts it cleanly: the two are different but complementary, with the DMP feeding anonymous audience segments to the DSP and the DSP executing the actual media purchase. A DSP collects only the data it needs to make a bid. A DMP exists to collect, organize, and activate audience data as its entire reason for being. Think of the DMP as the address book and the DSP as the phone. One tells you who to reach; the other reaches them.
Ad network versus ad exchange. Both connect advertisers and publishers, so the line blurs. The difference is the business model. An ad network buys inventory in bulk and resells it as a packaged product, acting as a principal in the middle with a markup you cannot see. An ad exchange is an open auction where buyers bid on individual impressions in real time, with the price set in the open. A network sells you a pre-arranged bundle. An exchange sells you a live, per-impression market. Networks still exist and still suit premium or simple buys, but the exchange is the transparent, automated route, and it is what most people mean when they say programmatic.
DMP versus CDP. This is the one that has actually changed under everyone's feet. A DMP is built on anonymous data: cookies, device IDs, IP addresses, much of it third-party, and it typically holds that data for less than 90 days. A customer data platform is built on identified, first-party data: names, emails, real customer records, kept for the full customer lifetime. The DMP was designed for the third-party-cookie era. As that era ended, the DMP's foundation cracked. The two largest exited the category outright: Salesforce declared end of life for Audience Studio, formerly Krux, on 1 February 2024, and Oracle shut down its entire advertising division, BlueKai included, on 30 September 2024. The CDP, built on data a company owns and a user consented to, is the more durable replacement, and modern CDPs have absorbed much of the audience-building job DMPs used to own. We trace that whole shift in from DMP to CDP. The short version: if it runs on anonymous third-party data, it is a DMP, and the category is fading.
The supporting cast, in brief
A few more boxes round out the stack. The agency trading desk is a programmatic buying unit inside an advertising agency, a center of excellence for programmatic that runs campaigns across DSPs for the agency's clients. It sits above the DSP and charges its own fee on top of the DSP's fee, which is precisely why some advertisers now question whether they need one. The header bidding wrapper lives on the publisher side: a piece of code in the page, usually built on the open-source Prebid.js framework, that asks many SSPs and exchanges for bids at the same time before the ad server makes its call. Prebid supports more than 200 demand sources through interchangeable adapters. Header bidding is the publisher's tool for forcing a fairer auction, and it has its own piece if you want the full story.
Future and impact: the stack folds inward
Two things are happening to this map at once.
The first is consolidation. Every layer in the stack takes a cut, and the buy side has noticed. Across DSP, exchange, SSP, and data fees, intermediaries can absorb 30 to 50 percent of an advertiser's spend before it reaches a publisher. The pressure to collapse the chain is real and it now has names attached. Omnicom, one of the largest agency groups, told investors in 2026 it had executed live media buys for several clients using an agent-to-agent buying framework, software negotiating directly with publishers. Its CEO described ad-tech intermediaries as charging a toll the client ultimately pays and called more direct publisher relationships a deliberate objective. The goal is not to delete the DSP and the SSP but to shrink their role and their take.
The second is the agentic layer. A new tier of software is forming on top of the stack: AI agents that take a goal and plan, transact, and optimize across the boxes without a human approving each step. For those agents to talk to each other, they need a shared language, which is why a standard called the Ad Context Protocol launched in 2025. Built on the same Model Context Protocol that underpins much agentic AI, AdCP aims to do for agent-to-agent advertising what OpenRTB did for real-time bidding: give every machine a common way to exchange structured data about audiences, inventory, and campaign goals. This is where an implementation partner like Perform Digital does its work, wiring an agent layer onto an existing stack without losing the auditability that makes the stack trustworthy.
The honest risk is worth stating. Programmatic spent a decade adding transparency tools, header bidding and supply-path optimization among them, to drag hidden fees into the open. An agent layer that buys on your behalf could quietly reverse that, hiding which intermediaries it used and what data it touched behind a clean conversational interface. The two-sided map in this post is not a relic. It is the thing you need in your head to ask an agent, or a vendor, the only question that has ever mattered in ad tech: which side are you on, and what is your cut.
Council summary
The post argues that the ad-tech alphabet stops being intimidating the moment you sort every acronym onto a buy side, a sell side, or the marketplace between them, because a tool's job follows directly from the side it serves. It earns its keep with the confusing-pair section, settling DSP versus DMP, ad network versus ad exchange, and the genuinely shifting DMP versus CDP question with dated, verified specifics, including the 2024 exits of Salesforce Audience Studio and Oracle BlueKai. The forward look is honest: an agentic layer and pressure from buyers like Omnicom are folding the stack inward, but agents could just as easily re-hide the fees that header bidding dragged into the open. The reader's takeaway is a single durable test to carry into any vendor or agent conversation: which side are you on, and what is your cut.
Comments