Programmatic DSPs for games, explained
Moloco, Liftoff and the wider DSP tier earn their margin on bidding technology, not inventory access, and deserve a fairer test than most buyers run.
Moloco and Liftoff sit at the top of a tier that also holds Bigabid and Remerge, along with a long tail of smaller names, and buyers tend to treat the whole tier as one undifferentiated thing: the DSPs. Ask what separates that tier from a self-attributing network like Meta, or from a supply-side platform the UA team never deals with directly, and the answers get vague fast. The distinction is worth getting right. It decides what a buyer is actually paying for, and it decides whether a head-to-head test between two vendors means anything at all.
What a mobile DSP is buying and selling
A demand-side platform buys impressions for an advertiser across a pool of supply it doesn't own, in-app inventory it reaches through exchanges and through direct integrations with publishers and ad networks, competing for every impression in real time against each other demand source bidding on the same slot. That's the structural difference from a self-attributing network like Meta or Google, which sells access to inventory it controls outright and then reports on the outcomes using its own measurement, leaving the buyer very little independent verification. The margin works differently too.
A DSP's margin comes from the spread between what it bids on a buyer's behalf and what it then charges that buyer, plus a management fee layered on top in most commercial arrangements. That spread is also the incentive to bid well, because a DSP that keeps overpaying for inventory relative to the value it delivers loses budget to a competitor that prices the same inventory more efficiently.
What "AI bidding" actually means here
Every DSP in this tier markets some version of AI-driven or machine learning bidding, and the substance behind the phrase is a real-time prediction problem: given a specific impression, a specific user signal and a specific app context, what is the probability this impression converts into whatever event the advertiser cares about, and how much is that probability worth bidding? The differences between vendors come down to training data volume and feature engineering, plus how quickly a model adapts to what a campaign does in its first few days. There is no categorically different technology that one company owns and the rest lack.
So bidding quality shows up in results, never in a vendor's description of its own model. Two DSPs can describe near-identical machine learning approaches and still produce materially different outcomes on the same campaign, because the quality of a bidding model lives in execution rather than in the pitch.
In-app bidding supply and why it matters
What a DSP can actually bid on depends on its integrations with in-app bidding exchanges and on its direct publisher relationships, and that varies more between vendors than the sales decks suggest, since all of them describe access to broadly similar supply. A DSP with weak access to premium rewarded or interstitial inventory in a given vertical simply has less good inventory to bid on, however strong its model happens to be. No amount of bidding sophistication fixes a thin supply pool.
This is part of why the mobile DSP tier has consolidated and repositioned itself over the past year, as UA Ledger reported when Liftoff withdrew its IPO and Moloco moved toward its own public listing process, a sign that the category's economics reward scale in supply relationships as much as in bidding data.
Running a fair test against self-attributing networks
The most common mistake is comparing a DSP's reported results straight against a self-attributing network's own numbers, as though the two measurement regimes were equivalent, when a network grading its own homework tends to look better on paper than it performs in an independent read. A fair test routes both channels through the same MMP, holds creative constant on both sides, then judges them on one shared event the MMP measures rather than on each channel's preferred definition of a conversion. Anything looser than that flatters whichever vendor reports most generously. Which is usually the network.
Budget allocation follows the same logic. A DSP earns a bigger share of spend by beating the studio's own measured retention and payback numbers across a full test cycle, not by winning an argument about whose targeting technology sounds cleverer in a meeting. The tier's whole pitch is that its bidding is worth the margin it charges. That's a claim a studio's own data settles far more reliably than any vendor's benchmark deck.
