Containerisation was the talk of ATS London last month, and with good reason, as it looks set to reshape programmatic as we know it. And just this week, Index Exchange and Nano Interactive announced that campaigns using Nano’s containerised intent signal on Index Cloud have shown over 30% improvement in CTR against control groups using conventional enrichment. It’s early data, but it points in the right direction.

So what is containerisation?

An auction is over faster than you can blink. An impression can come and go before an outside system has worked out what to do with it. Containerisation tackles that by putting selected intelligence inside the exchange’s auction, so it can act while the opportunity is still live. It’s part of a wider move towards sell-side decisioning, where more of the trade gets settled at the point of supply.

Below, I’ll explore how a container works inside the decision window, what buyers can build, where efficiency gains come from, and how open standards make it portable.

Every trip outside the auction costs you

For most of programmatic’s history, the clever decisioning has lived in the DSP. A media owner makes an impression available through an SSP, which sends a bid request across the internet to a DSP, often in a public cloud data centre far from where the impression started.

The DSP typically has around 100 milliseconds to respond, though publishers set their own timeouts. Every call to outside infrastructure eats into that window, and some take so long they miss it altogether. The signal suffers too. As Nano’s CEO Carl White puts it, “intent has a half-life measured in milliseconds.”

And of course, those calls also cost money, so DSPs routinely throttle incoming requests to keep their cloud bills and QPS in check, meaning they may only ever see a fraction of the impressions on offer.

Containers have been packaging code for years, but what’s changed is where they live, inside the exchange’s own auction infrastructure, so the logic never has to leave.

Inside the five-millisecond decision window

A media owner’s ad server passes an opportunity to an SSP, which checks if the impression is valid and filters out suspected IVT (Invalid Traffic), then hands it to a container before building the bid request.

The container sees only the signals it’s permitted to see, applies the partner’s logic and sends back an instruction. That might switch a deal on or off, adjust its floor, add a content identifier or choose which DSP gets the opportunity. On Index Exchange’s infrastructure, this takes roughly five milliseconds, up to 95% less waiting than a standard OpenRTB call.

Then it gets a second go. Once bids come back, another five-millisecond window opens for bid shading or margin adjustments. Every instruction is still just a request, and the host auction decides what to accept.

Build your own intelligence layer, bento-box style

Think of it like a bento box. Rather than one big bundle, a buyer picks smaller services that each do one job well, working alongside an existing DSP with no new SDK and no change to the trading set-up.

One service might match an impression to an audience segment. Another might estimate what it’s worth, while a scoring model pushes budget towards the most promising opportunities. Others could classify live content across STV (Streaming TV), web and app, or filter invalid traffic before it reaches a DSP.

The services can even work together. At ATS London, Index’s CEO Andrew Casale described containers starting to talk to each other, so a request missing a signal can be topped up by another partner’s container.

To keep proprietary logic private, containers run in strict isolation, access is tightly permissioned, and every piece of code is cryptographically signed. The exchange can check what’s allowed to run without ever seeing a partner’s models or data, and DSPs receive only the decision or Deal ID. It’s why Casale happily describes Index as “a dumb pipe”, leaving the clever work to its partners.

Why being close to the impression pays off

Keeping the logic inside the exchange cuts out the trips to external systems. That means lower latency, no external hosting bills and no public-cloud ingress or egress charges. A model also sees the full request before it’s pared back for downstream delivery, with every signal still intact. As Casale put it, “quite literally, time is money. In our world, we can repurpose time to make better decisions.”

And because it runs on the sell side, the intelligence gets applied to every impression the host sees, rather than the throttled slice a DSP gets round to. Index’s environment alone handles 700 billion requests a day across more than 3,000 media owners. Early results are promising. 

Of course, proximity on its own won’t deliver a worthwhile result, which is why clear permissions and auditable decisions are important. The logic still needs good signals, and measurement has to prove it’s improving matching or cutting waste. Nor does containerisation dictate an approach to identity. Some implementations run without cookies, device IDs or hashed personal data.

Build once, run on any compliant exchange

Programmatic companies haven’t always been good at working together. Casale summed it up bluntly: “We’ve operated as a thousand fiefdoms. Everyone knows a little piece of the truth.” But open standards could change that. The IAB Tech Lab’s ARTF (Agentic RTB Framework) gives containers one standard interface, so compatible services can run on any compliant host without rebuilding every connection.

Despite the "agentic" label, the agents running in these containers don't have to be LLMs. At auction speed, they’re more likely to be specialised models. Nano, for example, packs almost 100 million classified webpages, millions of CTV titles and a KPI-optimised ML model into one container that decides in under five milliseconds.

Making the decision while the impression is still there

In short, containerisation is changing where intelligence runs. Cutting out external round trips frees up the decision window, lets buyers build their own intelligence layer and applies it to every impression rather than a throttled few, while open standards carry that logic between exchanges. The auction will always be over in a blink, but the decision now gets made while the impression is still there to win.

Here at Netric, we've worked alongside Nordic publishers for decades, so we know their markets, their buyers and their inventory from the ground up. We'd love to help you work out what containerisation and Index Cloud could mean for your inventory, so do get in touch.