Why so many platforms
Early online advertising had two parties: an advertiser and a publisher, with an ad network in between bundling many small publishers into something an advertiser could buy. Programmatic replaced the bundle with a market, and a market needs infrastructure on both sides because the two sides want different things. A publisher wants to sell every impression at the highest price the market will bear while keeping control over who appears on its pages. An advertiser wants to pay the lowest price for the impressions that matter to it and to count, independently, what it received. Neither trusts the other's software to act in its interest, so each side built its own.
Specialisation followed. Buying decisions (the DSP) were separated from serving and counting (the advertiser's ad server) so that the count would not depend on the platform being counted. Selling decisions (the SSP) were separated from the publisher's own scheduling of direct campaigns (the publisher's ad server), though Google folded both into Ad Manager. Audience data was separated from media so that the same segment could be used across several DSPs, which produced the data management platform and later the customer data platform. Verification was separated from everything so that a neutral party could say whether an ad was seen and whether the page was safe.
The cost of this structure is that every platform charges a fee and every hop adds latency and a chance of disagreement in the numbers. The benefit is that a buyer can choose each component, change one without changing the rest, and check one against another. The fundamentals hub shows the full journey of an impression through the chain; this guide takes each stop in turn.
Each platform and its job
The table lists the platforms a buyer will meet, which side they serve and what they do. Google's products are named because most readers will meet them first; independent equivalents exist for every row.
| Platform | Works for | What it does | Google's product |
|---|---|---|---|
| Demand-side platform (DSP) | Advertiser | Receives bid requests, applies targeting, frequency, pacing and brand safety rules, sets bids, reports delivery | Display & Video 360 |
| Advertiser ad server | Advertiser | Hosts creatives, serves them into any DSP or direct placement, counts impressions and clicks, records conversions, manages frequency across sources | Campaign Manager 360 |
| Supply-side platform (SSP) | Publisher | Packages inventory, sets floor prices, creates deals, sends bid requests, runs or forwards the auction, reports yield | Ad Manager (Ad Exchange) |
| Ad exchange | Marketplace | Runs the auction between many buyers and sellers; now usually part of an SSP | Ad Exchange inside Ad Manager |
| Publisher ad server | Publisher | Schedules direct campaigns, decides whether programmatic demand may compete, picks the final ad for each slot | Google Ad Manager |
| Data management platform (DMP) | Either side | Collects anonymous, mostly cookie-based audience segments and syncs them to DSPs and SSPs | Audience features inside DV360 and Ad Manager |
| Customer data platform (CDP) | Advertiser | Unifies identified first-party records into profiles and activates them across channels, including DSPs | No single product; audience lists in DV360 receive the output |
| Verification vendor | Advertiser, sometimes publisher | Measures viewability, brand safety, fraud and invalid traffic, pre-bid or post-serve | Active View for viewability; integrated third parties for the rest |
| Agency trading desk | Advertiser, through an agency | Operates DSPs on behalf of advertisers, sometimes with its own technology layer | Not a Google product |
Two patterns are worth noticing. The sell side has consolidated: the SSP, the exchange and the publisher ad server are often one product with three names, and Ad Manager is the clearest example. The buy side has stayed separate: DSP and ad server remain distinct products even inside the Google stack, and the separation is deliberate. A buyer who wants a count it can defend keeps a system that does not also make the buying decision, which is why the DV360 section and the Campaign Manager 360 section of this hub are written as two tools rather than one.
How a DSP and an SSP talk
A DSP and an SSP are connected by a server-to-server integration that both parties set up once and then leave running. The SSP sends bid requests; the DSP returns bid responses; both follow the OpenRTB protocol that IAB Tech Lab publishes, which defines the objects a request may contain (the impression, the site or app, the device, the user, the regulations, the source and its supply chain) and the shape of the response (a price, a creative reference, a deal ID where one applies). The protocol is a vocabulary rather than a contract; each SSP chooses which fields to populate and each DSP chooses which to read.
Three things happen around the request itself. First, the DSP decides which traffic to listen to. An SSP produces far more requests than any buyer can use, so DSPs filter at the edge, asking the SSP to send only requests from certain countries, formats or publishers, or sampling the rest. A line item that targets inventory the DSP has filtered out will never see a bid request for it, which is a common and invisible cause of under-delivery.
Second, identity is matched. A DSP can only recognise a person if the identifier in the request is one it has seen before. On the web that has meant cookie matching, where the SSP and the DSP exchange their own cookie values for the same browser through a sync, so the DSP's audience segments can be looked up from the SSP's request. The match is partial, decays and fails entirely where third-party cookiesCookies set by a domain other than the site visited, used to recognise a browser across sites. are blocked, which is why publisher-provided identifiers and consented first-party signals are increasingly used instead. For the impressions that arrive with no usable identifier at all, the guide to planning contextual targeting in DV360 turns what the audience reads into category, channel and keyword line items.
Third, creatives are reviewed. Exchanges scan or review creatives before they may serve, and a creative that has not been approved on a given exchange produces bids that are rejected or never made. Google documents that Ad Exchange reviews creatives for policy compliance; the review is per exchange, so a creative cleared on one may still be pending on another.
What happens between request and response, the ranking, the pricing and the win, belongs to the real-time bidding guide. How a negotiated deal is carried inside the same request and response is covered in the deal types guide.
The incentives of the two platforms at either end of this connection, and what an advertiser controls on each side, are set out in the DSP versus SSP comparison.
Ad servers on both sides
The publisher's ad server is the final arbiter of what appears in a slot. It holds the directly sold campaigns with their guarantees, decides whether and when programmatic demand may compete, and chooses the ad that serves. In Google Ad Manager, the help centre documents that direct-sold line items, programmatic guaranteed deals and auction demand are evaluated together under a priority system, with guaranteed line items protected and non-guaranteed demand competing on price. The practical consequence for a buyer is that the auction is not the whole market: an impression may never be offered because a direct campaign has claimed it, and the floor a publisher sets in the SSP reflects what its direct sales already earn.
The advertiser's ad server plays the mirror role. A creative is trafficked into it once, with its sizes, landing page, tracking and any verification wrapper; the ad server issues a tag that is loaded into the DSP, a direct placement or any other buying platform. When the creative is called, the ad server serves it, counts the impression and the click, and fires conversion tags on the advertiser's site back to the same record. The count is independent of every seller and every DSP, which is the whole point. It also gives a single place to manage frequency and to attribute conversions across platforms, within the limits that identifiers allow.
The two ad servers disagree, routinely, with each other and with the DSP. The DSP counts a won auction; the exchange counts a served creative; the advertiser's ad server counts a creative that began to render in the browser, which is the counting point the industry's measurement guidelines now use. The gaps between these counts are normal; a gap that suddenly widens is the signal to investigate. The ad serving basics guide in the Campaign Manager 360 section covers tags, trafficking and the mechanics of counting in detail.
Data platforms: DMP vs CDP
A data management platform (DMPPlatform that organises audience data for targeting.) and a customer data platform (CDPUnified first-party customer profiles.) both hold audience data, and the names are used loosely enough that it is worth being precise.
| Row label | DMP | CDP |
|---|---|---|
| Who the records are | Anonymous browsers and devices, keyed by cookie or device identifier | Identified customers and prospects, keyed by email, customer ID or login |
| Where the data comes from | Third-party data providers, publisher partners, the advertiser's own tags | The advertiser's CRM, website, app, transactions and support systems |
| How long it lasts | As long as the cookie or identifier survives, typically weeks | As long as the customer relationship, typically years |
| Main use | Building prospecting segments and syncing them to DSPs and SSPs | Unifying first-party records for analytics, personalisation and activation across channels |
| Dependence on third-party cookies | High | Low; activation to a DSP uses hashed identifiers or list uploads |
| Direction of travel | Declining as cookie-based matching erodes | Growing as first-party data becomes the main targeting asset |
The DMP was built for a world in which a browsing history could be bought, segmented and matched to a DSP's cookie. Browser restrictions on third-party cookies have cut the matchable share of that data, and the segments that remain are less accurate than they were. The CDP was built for the advertiser's own data: the customers who bought, the visitors who browsed a category, the subscribers who opened an email. Its output reaches a DSP as an audience list, either uploaded or synced through a hashed identifier, and reaches publishers through deals in which the publisher matches the list against its own logged-in users.
In the Google stack, DV360 receives first-party lists, offers Google's own audience segments built from signed-in activity, and gives access to third-party segments from data providers who are paid per impression when their segment is used. Which to rely on depends on what the plan needs: scale from Google's segments, precision from first-party lists, and third-party segments used sparingly and checked against results, since each one adds a fee to every impression it touches.
For the decision itself, including what has taken over the DMP's job, the comparison of customer data platforms and data management platforms sets the two side by side.
Reading a supply chain and its fees
Between an advertiser's budget and a publisher's revenue sit several platforms, and each takes a share. The DSP charges a platform fee, usually a percentage of media. Data segments, verification and ad serving are each charged per thousand impressions. The exchange or SSP takes a share of what reaches it. Any reseller between the exchange and the publisher takes another. None of these are published in one place; they are set by contract and discovered by asking. When a buy's price per thousand has come in above what the plan priced and nothing was changed on purpose, the diagnosis of a programmatic CPM higher than planned reads this fee stack and the reseller hops beside five other causes.
Three IAB Tech Lab standards make the path visible. A publisher's ads.txt fileA file on a publisher's site listing the companies authorised to sell its ad inventory. (and app-ads.txt for apps)
lists the companies authorised to sell its inventory and whether each is a direct seller or a reseller. Each
exchange's sellers.json file declares the identities behind the seller IDs it uses. The SupplyChain object,
carried in the bid request's source field, records every hop an impression took to reach the DSP. Together
they let a buyer confirm that an impression claiming to be from a publisher really was offered by that
publisher, and see how many intermediaries stood in the way.
Supply path optimisation is the practice of using that visibility: preferring the shortest path to each publisher, cutting resellers that add a hop without adding inventory, and consolidating spend on the exchanges that offer the best price for the same impressions. Most DSPs, DV360 included, report delivery by exchange so the comparison can be made; what is reported about resellers varies by platform. When a placement report shows trusted publishers at prices that do not match a direct path, the diagnosis of programmatic spend going to low-quality sites reads the seller path beside the goal, the list and the verification that let the spend drift there.
Comparing two paths to the same publisher
A travel insurer buys the same sports publisher's inventory through two exchanges and compares the paths. On
exchange A, ads.txt shows the exchange as a direct seller and the SupplyChain object has one hop; the
insurer's average winning CPM over a week is £5.40 for 1.2 million impressions, with a measured viewability
rate of 66 per cent. On exchange B, the publisher is listed only through a reseller, the SupplyChain object
shows three hops, and the average winning CPM is £6.10 for 400,000 impressions at 61 per cent viewability.
The insurer's DSP fee is 12 per cent on both; a third-party travel-intent segment at £0.90 per thousand is
applied on both. Removing exchange B from the line item's inventory sources costs almost nothing in reach,
because the same impressions are available on exchange A, and lowers the blended CPM by about 15 pence
per thousand while leaving the publisher's share unchanged.
The habit to build is a quarterly look at where the money went: by exchange, by seller type and by fee category. A plan that cannot say what share of its budget reached publishers is not in control of its economics, however well its campaigns appear to perform.
Common mistakes
- Treating the DSP's report as the count. The DSP records what it won; the ad server records what rendered. Reconcile them and report the ad server's figure.
- Confusing the SSP with the publisher. An SSP can carry inventory from thousands of publishers, some of
them through resellers. Check
ads.txtand the SupplyChain object before trusting a domain. - Buying third-party segments without a test. Each segment adds a fee to every impression; run it against a control line item before scaling.
- Expecting a cookie match to reach everyone. A large share of impressions arrive with no usable identifier. Plan contextual targetingChoosing where ads appear from the content of the page, app or video being viewed rather than from data about the viewer. for them rather than discovering a reach shortfall later.
- Ignoring creative approval per exchange. A creative cleared on one exchange can be pending on another, which looks like under-delivery and is not.
- Letting the DSP's traffic filters go unexamined. If the DSP never listens to the inventory a line item targets, no bid is ever made, and no report will say so.
- Running the same creatives in two DSPs without one ad server. Frequency, counting and conversions split across two systems that cannot see each other.
