Types of Programmatic Deals Explained: PMP, PG, Preferred and Curated

There are four ways to transact programmatically: open auction, PMP, Preferred Deal, and Programmatic Guaranteed. They differ on exactly two variables. Is the price fixed in advance, and is delivery guaranteed.

That's the whole taxonomy. Everything else you'll read about deal types is a longer way of saying it.

These aren't marketing categories, incidentally. They're encoded in OpenRTB, the bidding specification maintained by the IAB Tech Lab, which is why a deal behaves consistently whichever platform you buy it through.

But the taxonomy has a blind spot, and it's the reason this guide exists. Knowing you're buying a PMP tells you how the transaction works. It tells you nothing about what's inside the deal, or who put it there. And on most of what you buy today, that second question matters more.

The four types, quickly

Open auction. Any eligible buyer can bid. Nothing negotiated, nothing reserved. Publishers set a floor and take the best bid. Maximum scale, minimum control, lowest CPM.

Private Marketplace (PMP). An invitation-only auction. You bid against a negotiated floor, competing only with other invited buyers. Better inventory, more transparency, and you still have to win the auction. Delivery is not guaranteed.

Preferred Deal. A fixed price, non-guaranteed, served ahead of private and open auctions in the publisher's priority stack. You get the impression offered at the agreed rate, you take it or you pass. No competing bidders, but no volume commitment either, so fill rate is the risk you carry.

Programmatic Guaranteed (PG). Volume and price agreed upfront, delivery committed. No auction at all: the deal clears at the negotiated CPM. This is a direct buy running through programmatic pipes.

The pattern usually holds: control and CPM move together. Each step from open auction toward PG buys more certainty, and certainty tends to cost more.

Usually, not always. On inventory that struggles to sell, a PG can clear below what the same impressions fetch in a contested PMP, because the publisher is buying certainty too. The rule is a tendency, not a law, and the exceptions are worth looking for.

Why the choice is harder than the table suggests

Most guides stop at the comparison table. The hard part isn't knowing the definitions, it's knowing which certainty you're actually paying for.

The common mistake is buying certainty you don't need. A PG deal on inventory that never sells out is money spent on a guarantee the market would have given you for free. Locking a fixed price on supply whose CPM is falling means you carry the downside without the upside.

The opposite mistake is subtler. Running everything through open auction to protect efficiency, then spending the saved budget on verification and blocklists to clean up what you bought. The efficiency was real and the cost moved somewhere else.

There's also a mechanical reason open auction doesn't automatically mean more delivery. Bidders operate under queries-per-second limits, and when those limits bind, the open auction gets sampled rather than fully evaluated. Deals carry priority through that, so the requests reach you. Which means a narrower curated pool can out-deliver a wider open-market setup, and the assumption that scale lives in the open exchange quietly stops being true at volume.

So the question isn't which type is best. It's what this specific campaign cannot afford to get wrong. If it's delivery on a fixed date, PG earns its premium. If it's inventory quality, a PMP or a curated deal gets you there for less. If it's neither, open auction is doing exactly what it's supposed to.

In practice, of course, the choice is often made for you. Publishers only offer their best formats on PG. Your SSP pushes its own packages. The client has commitments that predate the brief. Most of the time you're not picking from four options, you're deciding whether what you've been handed is worth what it costs.

Which is arguably the more useful version of the question anyway.

Where curated deals fit, and why the four-type model breaks

Here's what the standard breakdown gets wrong.

Curated deals are usually presented as a fifth type, or lumped in as a flavour of PMP. Neither is right. A curated deal is not a transaction type at all. It runs on the same Programmatic Guaranteed or Preferred Deal infrastructure as anything else. The deal ID behaves identically in your DSP.

What's different is upstream. Someone assembled a pool of inventory, often across several publishers, sometimes with audience or contextual data attached, and packaged it into one deal ID. The mechanics of how you buy it are unchanged. What changed is who decided what's in it.

And the specification makes this concrete. In OpenRTB, a Deal object carries an identifier, a bid floor, a currency, an auction type, and the list of seats permitted to bid. That's the substance of it. There is no field describing what inventory sits behind the deal, because that was never what the protocol was designed to convey.

The deal ID carries the terms. It does not carry the contents. That's not an oversight by sellers, it's the architecture.

Which means the four-type model is answering one question when there are two:

How am I transacting? Open auction, PMP, Preferred, PG. Fixed price or not, guaranteed delivery or not.

Who selected the inventory? The publisher, an exchange, a data company, a curator, or you.

These are independent. A curated deal can be a PG or a Preferred. A publisher's own PMP involves no third-party curation at all. Treating "curated" as a fifth item on the list hides the fact that it answers a different question entirely.

And it's the question that has quietly become the important one. Most of what's available to buy now sits in deals somebody assembled in advance. The transaction mechanics are the part you understand. The selection is the part you don't see.

What people get wrong

Assuming the deal type says something about quality. It doesn't. A PG deal on poor inventory is a guaranteed way to buy poor inventory at a fixed price. The deal type describes the contract, not the goods.

Paying PMP premiums for open auction inventory. This is the most expensive version of the mistake. If the same impressions are available in the open auction, a private wrapper around them is a fee, not a service. Worth checking, and easy to check: compare the domains in your deal against what you're already reaching.

Treating deal IDs as static and forgetting them. A deal was configured at a moment, against criteria that applied then. Six weeks later it's still serving whatever it was set up to serve, whether or not that's still the right thing. The deal type has nothing to do with this, and no deal type protects you from it.

Judging a curated deal by its name. "Premium sports audience" is a label someone chose. It tells you what they intended, not what's in the pool.

What to do next

Two questions on every deal, kept separate:

What am I paying for in the transaction? Certainty of price, certainty of delivery, or neither. Then check whether the campaign actually needs it. Most of the time, one of the two is enough.

Who assembled what's inside, and when? If the answer is a third party, ask what shaped the selection and whether anything has updated it since it was packaged. Most of the time nothing has, which is worth knowing before you commit six weeks of budget.

The first question is well documented and every guide covers it. The second is the one that decides what your campaign actually delivers, and it's the one nobody puts in a comparison table.