Anatomy of a mediation switch

Mediation switches get decided in monetisation meetings and felt in UA meetings. That mismatch explains why the team that made the call so often counts the switch a success while the team that buys users treats it as a mystery, and it's why this piece argues that a mediation switch is a media-buying decision wearing a monetisation badge. The eCPM uplift a vendor promises is real. You can measure it. The change in how the large networks value your users, and therefore what they'll pay to acquire them, is equally real; almost nobody measures it.

The claim is worth stating sharply because it cuts against the pitch. Every mediation vendor sells on fill and eCPM, with operational simplicity as the closer. Those are the metrics the monetisation lead owns, so the switch gets scored on them. But the two dominant mediation stacks in mobile games belong to the two dominant ad networks, and each network sees the inventory it mediates with a clarity it lacks on the other. Switching mediation changes which network has the better view of your users; that changes bidding behaviour on the acquisition side, and the effects there can outweigh the eCPM gain in either direction.

The mechanism

A network buying installs for your game through its own in-app bidding makes two connected judgements: what a user in your game is worth as an ad viewer, and what a user acquired for your game is worth as a conversion. When the same company mediates your inventory, it holds first-party data on the first question. It sees impressions, fill, eCPM and, through its SDK, session behaviour. When it doesn't mediate your inventory, it sees only what the MMP and its own conversion signal tell it.

The practical effect: the network whose mediation you use tends to be more confident about your users, and confident bidders bid more precisely. Sometimes that means cheaper acquisition for users it has strong signal on; sometimes it means a flat refusal to bid on users it has learned are low-value ad viewers, and the UA team can't tell which from the outside. Either way the curve on that network shifts in the weeks after a switch, for reasons the UA team didn't cause.

This has become more pronounced in 2026, not less. Unity shut the ironSource Ads network at the end of April and redirected its demand through Unity Vector, which concentrated the LevelPlay mediation relationship on a single AI-driven buyer. AppLovin has tied MAX to Axon for years. Unity's second-quarter results in early August reported Vector passing a one-billion-dollar annual run rate two quarters ahead of plan. The mediation layer and the acquisition layer are now more tightly coupled at both companies than at any point since the mergers that created them.

The second-order effect most switches miss

The obvious trade-off is that switching mediation away from a network may reduce that network's willingness to buy your users. The less obvious one is what happens to your ad revenue attribution, and through it to your ROAS targets.

Most hybrid-monetisation games feed ad revenue back to the MMP so that campaigns optimise on total return rather than IAP alone. That feed comes from the mediation platform, and the two platforms don't report impression-level revenue identically. Granularity differs. So does latency, and so does the handling of unfilled impressions. A switch therefore changes the ad-revenue signal every network is optimising on, at the same moment it changes which network has first-party visibility. For a few weeks the UA team is buying against a return figure whose definition has silently shifted, and any comparison to pre-switch performance is a comparison of two different numbers.

The fix is procedural rather than clever: freeze ROAS targets for a defined window after the switch, then re-baseline them from the new feed before anyone gets to conclude that acquisition got better or worse.

An illustrative sequence

Consider a hypothetical hybrid-casual publisher moving a portfolio of titles from one of the two major mediation stacks to the other, motivated by a promised eCPM uplift on rewarded video.

Week one to two: the switch goes live. Rewarded eCPM rises roughly as promised and the monetisation team records a win. UA notices nothing yet; the networks are still bidding on cached signal.

Week three to five: the network that lost the mediation relationship begins bidding less aggressively on the portfolio. Its cost per install falls slightly but volume drops more. The network that gained it starts bidding more confidently, and its volume rises, but its cost per install rises too, because it now knows which users are worth paying for. Blended cost per install moves up. The UA team, still reading the pre-switch ROAS feed, sees return falling; reallocation begins.

Week six to eight: somebody has re-baselined the ad-revenue feed, and the picture clarifies. Total return per user is actually higher than before the switch, because the eCPM uplift is real and the network with better signal is acquiring better ad viewers. But the volume mix has shifted heavily toward one network, and the portfolio now depends on a single buyer more than it did.

In this illustration the switch paid, but on a different line from the one it got scored on, and nobody recorded the concentration risk.

A decision framework

For a team weighing a switch, four questions to answer before the eCPM comparison gets a vote.

• Which network currently buys the largest share of your installs, and does the switch move mediation toward or away from it? Moving toward it deepens a dependency. Moving away from it may cost volume on your best channel.
• How will the ad-revenue feed to the MMP change, and who owns re-baselining ROAS targets afterwards? If the answer is nobody, the UA team will misread the first two months.
• What does the switch do to your negotiating position on the acquisition side? A publisher whose inventory is mediated by network A has a lever with network A that it loses on switching. That lever has a value, and it belongs in the comparison.
• How long is the migration window on the old mediation contract, and what happens to historical revenue data at the end of it? When Unity closed the ironSource network in April, advertisers had until the end of May to pull historical data. Mediation contracts have their own cut-offs, and the re-baseline needs that data.

Only after those four have answers does the eCPM figure earn a place in the decision, weighed against the volume and cost changes on the acquisition side. That means the monetisation lead and the UA lead in the same meeting with the same spreadsheet.

Who should own the decision

Whoever owns total return per user should own a mediation switch, which in most studios is nobody and in well-run ones is the growth lead sitting above both monetisation and UA. Where that role doesn't exist, the UA lead should insist on a seat in the decision and on the re-baselining window as a condition of the switch. The vendor will present the eCPM uplift as the whole story. It's the easiest part to measure. In a market where the mediation layer and the largest buyers are the same companies, it isn't the part that determines whether the switch made money.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• Buying rewarded video: a UA-side view
• Mobile ad networks explained for beginners