Every company in ad tech has an AI story right now. Most of them started around the time the word began moving share prices.
Ours started earlier, and it was a lot quieter. Machine learning was doing real work inside ConnectAd long before “AI-powered” became the default line on every pitch deck. We just never saw the point of shouting about it. No rebrand, no sudden pivot, no buzzword bingo.
At the end of October we broke that silence, a little, when we launched our real-time AI Brand Safety Stack. This post is about the thinking behind it: why we run a deliberately small team, why AI has been part of our platform from the start, the rules we don’t bend, and where we are heading next.
ConnectAd is a small team, and that is not a phase we are trying to grow out of. We are bootstrapped and owner-operated, and we intend to stay that way.
That choice shapes everything. A small team means short decision paths: the person who talks to a publisher is never more than one conversation away from the person who can change something in the platform. It means low overhead, which is one reason we can run industry-low take rates. And it means we hire for quality, not headcount. Everyone here carries real responsibility, and partners deal with people who know their account, not a rotating queue.
The platform, on the other hand, is anything but small. As a fully independent European SSP, we process more than 7 billion ad requests a day. Behind every request sits an advertiser, a landing page and a creative, all of which ideally deserve a proper review.
No team can check all of that by hand, and hiring our way out of it would turn us into exactly the kind of company we never wanted to be.
Most of ad tech scales the same way: more volume, more people, more layers. We quietly went the other way. A handful of people running a platform of this size is not how the industry usually does it, and we would argue it is one of the strongest arguments for ConnectAd. It works for two reasons: an exceptional team, and the fact that automation and AI have been built into ConnectAd from the very beginning.
Our founders came up in the early days of programmatic, working with DSPs for agency trading desks. They saw first-hand how much of the open web’s value disappeared into inefficiency: bloated supply paths, opaque fees, junk traffic dressed up as inventory. ConnectAd was built as the answer to that, without a cent of outside funding.
When you are bootstrapped, every piece of technology has to earn its place. There is no investor money to burn on features that look good in a press release. So automation and machine learning were never a marketing decision for us. They were the only realistic way to run an auction of this size with a team of our size, and to keep it clean and efficient while doing so.
The hype changed the vocabulary around us. It did not change how we work. What it did change is the toolkit: today’s models can make judgement calls that were out of reach only a few years ago. That is what made our Brand Safety Stack possible.
Every idea has to pass the same test. Does it make an internal process better, or does it improve the service our partners get from us? If the answer is neither, we don’t build it.
That test has killed more projects than it has started, and we are fine with that. A feature that only exists so we can put “AI” on a slide helps nobody. Brand safety passed with ease. Here is why.
For years the industry has handled brand safety with two tools: static block lists and paid third-party add-ons. Both share the same flaws. Block lists are blunt. They reject legitimate campaigns because a declaration looks wrong, yet still wave through the fake shop that registered its domain yesterday. Add-ons cost money and slow the auction down.
The underlying problem is that a list cannot judge context. Our AI Brand Safety Stack can. For every bid, it looks at what actually stands behind it, along the entire demand path:
This is exactly the kind of work AI is built for, and it has nothing to do with milliseconds. The auction itself runs on rules and signals that have to be ready instantly. AI earns its place behind that, on volume: billions of requests produce a flood of raw data, from domains and landing pages to creatives and declarations, far more than any team could ever review. AI works through that mass of data continuously and turns it into verified signals the auction can act on.
For publishers, the result is fewer bad ads and fewer wrongly blocked good ones, because their bid and block rules now run on verified information instead of declarations alone.
And it is included for every ConnectAd publisher, at no extra cost. You will find all the details in our launch announcement.
We could have bought a verification service and passed the bill on to our partners. Much of the market does exactly that. We built the checks into our own auction instead, and there is a very practical reason we could.
ConnectAd runs its own on-premises data centres in Europe, Canada, the US and Singapore, with no cloud dependency. Owning the hardware means the analysis runs on our own machines, right next to the auction, so its results are there when the auction needs them without adding latency. It also means there are no cloud or verification fees to hand down. That is how advanced screening and low take rates end up in the same sentence.
Privacy is not something we bolt on at the end. ConnectAd is built GDPR by design, and our use of AI follows the same principle.
On the platform side, the Brand Safety Stack runs entirely on our own infrastructure, and EU traffic is processed exclusively in our EU data centres. European data stays in Europe. For the AI we use internally, we only work with set-ups where data stays under our control and is never used to train anyone else’s models. If a tool cannot meet that bar, we don’t use it, however good the demo looks.
You will notice we don’t name models, vendors or architectures. That is deliberate. Models change every few months; the principles we choose them by don’t. And frankly, how we build is part of what makes us different, so we would rather show you results than blueprints.
Internally, we have built an intelligent workspace that brings together the knowledge, context and know-how our teams have built up over the years.
It was not free. In a small company the real cost is people’s time, and we spend a lot of it checking what goes into the system, because what comes out is only as good as what goes in. It is already paying off on the repetitive, time-consuming tasks. The bigger win was one we didn’t plan for: writing down how we work forced every team to spell out its unwritten rules, and departments now understand each other’s processes far better than before.
Being small also keeps us honest. There is nobody further down the line to catch your mistake. So we check AI output in more detail, not less, and responsibility for the result always stays with a person.
The Brand Safety Stack is the first AI capability we have talked about publicly. It will not be the last.
Over the coming months, AI will move deeper into the ConnectAd platform, and increasingly in agentic form: systems that don’t just score or flag, but act, within clear rules, on behalf of our partners. We won’t pre-announce features here. But we can tell you how we decide where an agent belongs:
Just as important is where agents don’t belong. They won’t replace the person a publisher calls. They won’t make commercial decisions behind anyone’s back. And they will always operate within rules our partners can see and control, with a human accountable for the outcome.
There is a bigger shift behind this, too. Buying is becoming agentic: more and more media will be planned, negotiated and bought by machines talking to machines. In that world, clean and transparent supply paths matter more, not less, because an agent is only as good as the information it gets. Building for that future is not a side project for us. It is the logical next step for a platform that was built for efficiency and transparency from day one.
What you won’t see from us is an “agent” that exists so we can put it in a press release. Same test as always: does it make our service better, or a process better? If not, it doesn’t ship.
All of this serves one goal: giving our people more time for the work only people can do.
AI is excellent at working through masses of raw data. It is far less good at the judgement that needs experience, context and plain common sense, or Hausverstand, as we call it in German. That part stays with people: analysing in depth, asking the awkward questions, and making the call.
Our customer success team spends more time talking with, and listening to, our publishers. When a publisher needs a custom block rule, they speak to a named expert, not a ticket queue. Sales talks to more publishers, more often, and arrives better informed. Marketing has more room to create.
That is what lean by design means to us. The machines take the volume: more than 7 billion requests a day, every one of them checked. Our team takes everything that needs a conversation.
Want to see what that looks like for your inventory? Talk to us. You’ll get a person.