What Is a DSP?

A demand-side platform is software that buys advertising automatically. The advertiser sets who they want to reach, what they will pay, and where they will allow it to appear; the DSP works out the rest and places bids in auctions in the time between a page loading and an ad being needed.

The core idea is latency

The whole model exists because of a technical problem with a deadline. When a page loads, an ad slot becomes available for a few tens of milliseconds. Nothing human could negotiate in that window, so the buying is delegated to software on both sides: the DSP bidding for the advertiser, and a supply-side platform offering the inventory. That deadline is why programmatic exists and why every difficulty in the ecosystem traces back to it.

What happens in the auction

When a page loads, the publisher's inventory is offered to a real-time bidding exchange. Advertisers bid, usually in a few milliseconds, and a winner is chosen on price subject to the targeting each bidder submitted. The bid is what the advertiser is willing to pay for that specific impression, priced according to how likely it is to work.

What a DSP is actually for

Three things, and the third is the one people miss. It manages campaigns across many exchanges. It applies targeting and frequency controls. And it optimises: it learns which combinations of audience, context and creative produce results, and shifts budget toward them during the campaign rather than only between campaigns.

How a DSP differs from TikTok or Meta inventory

A DSP buys across the open web, assembled from many publishers. TikTok and Meta are single closed platforms where the inventory, the user data and the auction are all owned by the same company. That is why targeting on those platforms can be more specific: the platform knows what individual users watched. It is also why the two are harder to compare directly, because the inventory is not the same kind of thing.

The honest limitation

Programmatic is often described as precise targeting, and that overstates it. The ad is placed against a probabilistic prediction built from cookies, device identifiers, site context and modelled interest. It is good at scale and increasingly good at relevance, but it is inference rather than identification, and any claim to the contrary should be treated with suspicion.