Understanding containerization in ad tech
Historically in programmatic, most decisions happened on the buy side within DSPs, and the sell side was primarily responsible for making vetted inventory available.
An impression would originate with a media owner. An SSP would send a bid request across the internet to the buy side—often into public cloud infrastructure—where bidding decisions were made. Then it was a race to get that decision back in time for the auction.
That model is now evolving. With sell-side decisioning, decisions are also being made closer to where impressions actually originate, where rich and valuable signals live, making each impression smarter from the start.
Containerization is becoming one of its most advanced applications of sell-side decisioning today. It creates new opportunities to innovate, while reducing compute costs and latency. Containers allow companies to run custom code directly inside the auction environment, in real time, before bid requests reach DSPs.
What is a container?
A container is a lightweight, virtualized environment that packages and runs code. Containers historically have been used to move applications from on-prem servers to public clouds. In the same way, applications can now move from public clouds—which are increasingly cost prohibitive—directly into the auction environment, right within the exchange.
In programmatic, that means deploying custom logic—like targeting rules, bidding models, identity resolution, or supply curation—inside the auction environment the moment an impression becomes available.
Instead of sending impression data out to an external system, like the public cloud, for processing, containers bring the processing closer to the auction where all the valuable impression data originates. That reduces the need for outbound cloud calls, cutting latency and lowering infrastructure costs at scale.
When an ad opportunity first becomes available, the container runs its application or decisioning before bid requests are submitted to DSPs. The container receives impression-level signals, applies partner logic, and returns a decision. That decision might determine which DSPs receive the request, what enrichment is attached, what floor applies, or whether the impression qualifies at all for a given buyer or campaign.
It means we can now tailor the bidstream before the auction begins.
And all of that happens in about five milliseconds. This isn’t post-auction analytics. It isn’t asynchronous enrichment. It’s real-time sell-side decisioning at the moment the impression exists, before it enters the bidstream.
A programmatic auction flow with containers
What does this actually look like in an auction? Let’s walk through a simplified example:
1. First, a media owner’s ad server sends an ad opportunity to an SSP.
2. The SSP performs its quality checks—like confirming it’s a valid impression and filtering invalid traffic—then routes that impression opportunity to a container.
3. The container executes within milliseconds. It receives impression-level data, like device signals, content metadata, or publisher context. It processes this information and applies partner-defined logic.
4. Then it can return a targeting decision, or perform bid enrichment, floor adjustment, or filtration instructions.
5. The SSP then constructs the auction, taking into account any container output as it builds the bid request. Here’s the key point: All of these steps happen before any DSP is involved. That means DSPs receive more relevant, higher-quality inventory, improving campaign performance while respecting their QPS limits.
6. Only after that upstream intelligence is applied does the bid request go to qualifying DSPs, and the standard auction continues.
Whenever we talk about data and logic, security is essential. Containers are designed to run in strict isolation. Sensitive data stays inside the container. Access is tightly permissioned, and code is cryptographically signed to prevent unauthorized execution.
That means proprietary data and algorithms can be applied at the impression edge without being exposed in the bidstream. DSPs receive decisions, not underlying raw data.
Programmatic use cases for containers
Containers are flexible, leveraging a generalized framework, and the logic that runs inside them can vary widely depending on the use case.
Here are a few examples:
- Targeting, where impression signals are matched against audience taxonomies or segments in real time
- Bidding, where custom valuation models can decision based on floor prices, bid shading logic, or even calculate the bid price itself
- Scoring, where each impression is evaluated against defined criteria to prioritize high-value opportunities, allocate budget, or trigger different decisioning paths
- Supply curation, where inventory is filtered or routed based on brand suitability, contextual rules, or buyer-defined preferences
- Identity resolution, where available identifiers like device IDs, IP addresses, or other signals can be resolved against an identity graph inside the container, returning only the decision or deal ID—not the underlying identity data
- Invalid traffic detection, where additional custom fraud detection models can filter invalid impressions before a bid request is ever sent
These are just a few examples, and we’re continuing to see even more develop over time.
What containerization means for the future of programmatic
So, why is containerization such a meaningful evolution for the programmatic industry?
First, it marries real-time decisioning and the scale of the open internet with rapid innovation. Because containers execute in just a few milliseconds within the auction window, they enable capabilities that batch or post-auction systems simply can’t deliver, like dynamic floor pricing, impression-level scoring, real-time identity resolution, proactive curation, and more.
Second, it improves efficiency and opens the door to better outcomes. DSPs often throttle inbound requests to manage compute and cost, which means they evaluate only a fraction of total available impressions. When intelligence runs on the sell side, it applies to every impression seen, not just a filtered subset. That increases match quality and reduces waste.
Third, it reduces what we often refer to as the invisible cloud tax. Localized processing eliminates the need for outbound calls to external infrastructure, reducing latency and operating costs. That also means more simplified integrations, lowering barriers to entry for new companies to scale, accelerating innovation across programmatic.
Containerization brings logic closer to where impressions originate—closer to media owners, closer to consumers, and closer to the moment when value is created. It allows for true real-time decisioning.
It’s a new infrastructure layer for how programmatic operates and innovates, and we’re just beginning to see what becomes possible when intelligence lives at the impression edge.
See what’s possible when you bring intelligence closer to the impression.
Thank you to Michael Richardson, VP of product at Index Exchange, who also contributed to this video.
