Creative Volume Needs After an Ad Network CPI Reset
AppLovin and Unity's Q2 2025 results reset ad network CPI assumptions, and creative volume planning now needs a formula, not a guess.
AppLovin's Q2 2025 results, reported Aug 6, showed advertising revenue of about $1.259 billion, up 77% year on year, with net income around $820 million and Q3 guidance of $1.32 billion to $1.34 billion at an 81% adjusted EBITDA margin. Unity's same-day report put revenue at roughly $441 million, down 2% year on year but ahead of expectations, with the Unity Ad Network up 15% sequentially and now 49% of Grow. Both numbers describe the same shift from a buyer's seat: automated bidding is absorbing more of the auction, and the ad network CPI a campaign sees on a given day depends less on manual bid adjustments than on what the algorithm has to work with. What it has to work with is creative volume, and most teams are still sizing that output against last year's assumptions.
The reset that earnings made visible
Axon 2.0 and Unity Vector are not new as of this quarter. What Aug 6 changed was the confidence with which both companies described how far the automation has moved. AppLovin's growth rate at that scale is not explainable by budget reallocation alone, and Bromberg's description of Unity's ad business as reaching an inflection point was a direct claim about the model, not the market. For a buyer, the practical read is that CPI is increasingly a function of how many distinct creative signals the algorithm can test per dollar, not how well a media buyer manually shapes an audience. Creative volume ad network CPI dynamics used to be a soft correlation. On these networks, it is closer to the mechanism itself.
Why "more creative" is the wrong instruction
Telling a creative team to produce more assets after a CPI shift usually produces more of the same asset. A studio that made twelve video variants a month before Q2 does not improve its position by making twenty variants of the same three hooks. Automated bidding systems reward variance in the signals they can read: hook type, pacing, sound treatment, UGC versus studio-shot, gameplay capture versus stylised. Volume without variance just gives the algorithm more copies of the same experiment, and the CPI curve does not move. As we argued in Creative Hook Variety: Designing Ads for Mobile Games, the unit that matters is the distinct concept, not the rendered file count.
A formula for sizing creative volume against ad network CPI
A workable planning method separates three numbers rather than guessing at a single "how many ads" target.
- Active concepts. The count of genuinely different hooks running at once, not variants of one hook. Most mid-size UA teams can sustain six to ten live concepts across their top three campaigns without diluting signal per concept.
- Variants per concept. Aspect ratio, first-frame, sound-on and sound-off cuts, and length variants of the same idea. Three to five variants per concept is typically enough for the network to find its preferred cut without burning production hours on redundant work.
- Refresh interval. How often a concept needs a genuinely new idea behind it, not just a re-edit. Under the current CPI environment, hybrid-casual and puzzle titles are seeing faster decay on individual hooks than kingdom-builder and survival titles, where a slower-paced proof point holds longer.
Multiplying active concepts by variants per concept gives a working asset count, and the refresh interval tells a producer how often that whole set needs to turn over. A team running eight concepts at four variants each, refreshing every three weeks, is producing on a materially different cadence than one making twenty ads a month with no concept discipline behind the number, even though the second team's raw output looks larger on a spreadsheet.
A worked example
Take a hypothetical hybrid-casual studio spending most of its budget across AppLovin and Unity, running six active concepts before the Q2 results landed, at three variants each, refreshed roughly monthly. Applying the formula above after the reset might mean lifting active concepts to eight, holding variants at three to four rather than inflating that number, and tightening the refresh interval to every three weeks rather than four. That is a production increase of perhaps 40 to 50% in raw asset count, considerably less than the "double everything" instinct a CPI headline tends to provoke, and concentrated in the number that actually moves an auction outcome, which is distinct concepts, rather than the number that is easiest to inflate on a status report, which is total rendered files. The studio's producers can cost that increase against a concrete headcount or freelance budget line, rather than an open-ended request to "make more."
Where teams cut corners and pay for it
The most common failure after an earnings-driven CPI reset is treating the extra spend on testing as a one-off catch-up rather than a permanent step change in production cadence. Teams that ramp concept count for one sprint and then relax back to the old cadence see CPI drift back up within a month, because the algorithm's baseline expectation of fresh signal has already moved. The other common failure is outsourcing volume to a single UGC vendor without keeping a studio-shot or motion-graphics concept in the mix; networks weighting hybrid-casual and mid-core titles differently reward a spread of production styles more than a deep bench of one style done well.
A third failure worth naming is measuring the wrong thing once the new cadence is running. A producer who tracks total assets shipped per month, rather than active concepts and refresh interval specifically, can hit every target on their dashboard while the actual signal variety the network sees stays flat, because variant count is the easiest number to inflate and the least connected to the CPI outcome a team is trying to fix. Reporting on the three numbers in the formula above, rather than a single blended output count, keeps the incentive pointed at the thing that actually moves cost per install.
What this means for the next planning cycle
Media buyers reading AppLovin and Unity's results as a pricing story alone are missing the operational implication. The networks are effectively telling buyers what kind of input improves the auction outcome, and that input is concept variety delivered on a steady cadence rather than in bursts. Teams that rebuild their production brief around active concepts, variants and refresh interval this quarter will be working with the CPI environment these results describe rather than reacting to it every time a new number surprises them.
