Most companies think they control their bidding. They have custom bidding algorithms, tuned models, first-party data feeding the decisions. It feels like control.
Then you look at where the algorithm actually runs, and who owns the ground it runs on.
There is a real and widely misunderstood difference between customising a bid inside someone else’s DSP and owning the bidder outright. The two get talked about as if they sit on the same spectrum. They do not. One is decorating a room you rent. The other is owning the building. This is what actually separates them, and why the difference decides what you can build.
What a DSP gives you, and where it stops
A demand-side platform buys ad space programmatically across exchanges and SSPs on your behalf. The good ones let you go well beyond preset bid strategies with custom bidding.
Take the most capable example in the market. Google’s DV360 lets advertisers upload their own bidding logic, including script-based algorithms written in Python. That algorithm goes through a training phase, learns from past campaigns, and scores impressions, assigning higher values to the ones more likely to hit your goal and lower values to the rest. You can weight signals. You can feed first-party data. You get, in Google’s own framing, granular control over what the algorithm values.
That is genuinely powerful, and for most advertisers it is the right tool. But look precisely at what you control and what you do not.
You control the scoring function: how impressions are valued relative to each other. You do not control the auction it plugs into, the infrastructure it runs on, the latency budget it operates within, the data pipeline that feeds it, the retraining cadence, or the model architecture underneath your script. You are writing the rules for a game whose board, referee, and clock belong to someone else.
Your custom algorithm is a passenger with a very good view. It is not driving.
What owning the bidder changes
A proprietary bidder is the engine itself: the server that receives the OpenRTB request, decides in tens of milliseconds whether and what to bid, and does it hundreds of thousands of times per second. Owning it means owning all the layers the DSP kept for itself.
You own the valuation model, not just its inputs. On a DSP you score impressions the platform surfaces to you, using the features the platform exposes, retrained on the platform’s schedule. Owning the bidder means choosing the model, the features, the training data, and the retraining cadence. When you learn something new about what predicts value, you rebuild the model. You do not file a request for the platform to consider it.
You own the data that makes the model better. Every auction generates signal: bid landscapes, win rates, what won, what lost, how the creative performed. On a DSP that exhaust is generated by your spend and lives in the platform’s infrastructure, exposed to you on the platform’s terms. The bidder owner keeps the full stream and trains on it directly. Since the quality of a bidding model is a function of the data behind it, this is the compounding advantage. Owned data feeds a better model, which wins better impressions, which generates better data.
You own the latency budget. The bidder has roughly 100 milliseconds for the full round trip. How you spend that budget, how much goes to model inference versus data lookups versus network, is an architecture decision. Own the bidder and it is yours to optimise. Rent it and you live inside someone else’s allocation.
You own the economics at scale. Buying through a platform means the platform takes its share of everything flowing through it, and that share grows exactly as you grow. Owning the bidder converts a variable cost that scales against you into a fixed infrastructure investment that scales for you.
You own the roadmap. This is the one that decides the next five years. On a DSP your capability ceiling is whatever the platform ships. Your differentiation is capped at features available to every other advertiser using the same platform. Owning the bidder means your roadmap is set by your engineering priorities, and a capability you build is a capability your competitors on the shared platform simply cannot access.
The test that cuts through it
One question separates real control from the feeling of it: when you need something the system was not designed to do, what happens?
On a DSP, even a sophisticated one, the answer is a support ticket and a roadmap conversation. Maybe the platform builds it. Maybe not. On their timeline, for their business, weighed against every other customer’s request.
Own the bidder, and the answer is that you build it. A non-standard optimisation goal, an unusual signal, a downstream event specific to one partner, an audience definition nobody else supports. The gap between “we will raise it with our platform” and “we will build it this sprint” is the entire difference between customisation and control.
This is not for everyone
Custom bidding on a strong DSP is the correct choice for the overwhelming majority of advertisers, and pretending otherwise would be dishonest.
Building and running a proprietary bidder is hard, slow, and expensive. It means owning uptime for infrastructure that cannot go down, capacity planning for traffic that spikes without warning, retraining pipelines, and the unglamorous cost of paying for bandwidth on requests you will never bid on. It produces no revenue for a long time before it produces an advantage. If you are running campaigns, not building a programmatic business, a DSP with custom bidding is not a compromise, it is the right tool, and you should use it.
The calculus changes when your bidding is your business rather than a means to run campaigns. At that point every layer the DSP keeps, the model, the data, the economics, the roadmap, is a ceiling on what you can become. And the only way through a ceiling is to own the floor it sits on.
This is the decision AdSpin made deliberately. We built our exchange and our oRTB bidder in-house, not because building is romantic, but because we could see that every capability worth having eventually ran into a layer a platform would not let us change. Our bidder is ours: the model, the data pipeline, the latency budget, the roadmap. Tuned for mobile in-app specifically, running on supply we own and operate, which means the signal it learns from is clean before the model ever sees it. When a partner needs something the standard playbook does not cover, we build it, because there is no platform between us and the decision.
Custom bidding lets you tune the algorithm. Owning the bidder lets you decide what the algorithm is. Everything downstream follows from which one you have.
Explore how AdSpin’s proprietary bidder and owned exchange give partners real control over programmatic outcomes at adspin.io.
