Moloco, Liftoff and Mintegral: one partner per market

Running Moloco, Liftoff and Mintegral at the same time in the same country is how you raise your own clearing price. They bid into the same exchanges for the same device, so three of your partners end up competing with each other on your budget. We run one demand-side partner per objective per market and prove the next one is better by replacing it for a fixed window, usually four weeks, rather than by adding it alongside. The sequence: We start by deciding what each market is for, then assign a single partner to it. Moloco where there is a clean event stream and enough volume to warm its model, which in practice means two to three weeks of spend before its pricing settles and enough real conversion events each week that the model is not extrapolating from a handful. Liftoff where iOS carries the weight and SKAN is the only signal. Mintegral where supply in Southeast Asia or Latin America decides the outcome, and only after we have read the source-app list. What we need from your stack: Postbacks configured to each partner with the events they can actually optimise on, an agency-neutral account structure opened in your name, and a data feed for Moloco if you want its model to price your users rather than your category. Where you sell subscriptions, we need renewal events, not just the trial start; a bidder trained on trial starts will find you people who love free trials. How we know it worked: Sequential holdouts. One market, partner off for a defined period, organic and other-channel volume watched against the pre-period trend. Where the platform supports a PSA or ghost-bid holdout we use that instead, because it is cleaner and cheaper. We publish the incremental figure alongside the attributed one every time, including the times the gap is embarrassing. Two of the last engagements ended with us recommending less spend in a market, not more. Not included: Creative production, which is a separate service. Mediation, monetisation and anything on the supply side. Fraud litigation and claims against a network; we will assemble the evidence and stop the spend, but chasing a refund through a partner's account team is your commercial relationship, not ours. Retargeting on iOS beyond consented traffic, which for most catalogues is too small to be worth building. Who this is not for: Anyone who wants a three-way bake-off in month one. Anyone who wants us to buy through our own seat so the media cost is bundled into one invoice; we buy at cost, in your name, and you see what was paid. Apps with fewer than a few hundred conversion events a week, which is below the point where any of these models can be trained and where the honest recommendation is to fix conversion before buying more traffic. Health apps handling sensitive categories in EU markets where an allowlist of source apps cannot be agreed. What the decisions actually are: Which single partner owns a market, when to swap it, what event to expose to the model, and when the incremental read says a channel is buying users who were coming anyway. Those four choices decide whether programmatic works for you. Dashboard access and a weekly call are not the product; they are what surrounds it.