Creative Asset Specs
You can find the creative specifications for each network here:
Creative Optimisation Best Practices
How to set up your creatives in your campaigns:
Include both video orientations (landscape + portrait) in creative pack to maximise reach:
Android: 4-6 creatives recommended per campaign (consider ~min 100 USD / creative).
iOS: No more than 4 creatives (consider creative + campaign SKAD limitation and 72 hr cooldown for unlocking slots).
Create separate testing campaigns for your new creatives with allocated budget + raise bids only if necessary.
Separate campaigns by ad type to optimize bids to the performance.
How to analyse and manage your creatives:
Focus on CVR, eCPM and number of starts and installs to understand your creative performance.
Add a new creative packs when:
An active campaign struggles to scale or perform.
When newly tested creatives have been vetted in same conditions as evergreen campaigns and are scoring strong results.
Keep a creative pack active on a campaign when:
It shows a steady level of starts and consistent performance.
Several creatives continue to receive a good number of starts but one of them shows a lower observed CVR, this specific creative should not be removed because it can still work better for some users.
Unassign a creative pack from a campaign:
The creative pack shows a downward trend in receiving starts, this might be a sign of viewer fatigue.
The creative consistently scores a low ROAS / Retention performance.
The creative is not picking up and consistently receiving a very low number of starts. This is a sign that the creative doesn't fit the targeted users in the campaign and only receives starts from random exploration.
Creative Learning Phase Explained
Conversion on Unity:
The Unity algorithm selects the most valuable creative for each user based on an individual prediction, taking into account all campaigns applicable to that user's targeting, rather than simply relying on the average conversion rate of a creative.
When you add a new creative, it will compete with other creatives in the same campaign as well as other campaigns with similar targeting.
If there are other high-converting creatives in the same campaign with the same cost-per-install (CPI), the new creative may struggle to compete against those established creatives with existing data. To address this, it may be beneficial to place the new creative in a separate campaign with slightly higher CPIs to provide a competitive advantage.
Removing a creative that receives a lot of starts can lead to the following:
The traffic will shift to SOME other creative. It will basically shift to whatever is second best for those users to whom the removed creative had been shown.
This second best creative can be another creative within the same campaign but it can also be a creative from another title of yours or from an entirely different advertiser.
We cannot know which scenario will happen. It is however highly likely that the traffic cannot entirely be caught by a creative from the same campaign or app.
Historical data:
What kind of historic data us taken into account? Simple answer: all that is available.
There is a certain hierarchy to how historic data is used. It moves from the most specific available level to the broadest:
creative → campaign → game → category → organisation.
In general, it applies that if and the more information exists on the smallest unit, the more this information is used. If information/history is not available on one level, the next best level where data exists is used.
Creative Optimisation Best Practices
How to set up your creatives in your campaigns:
Include both video orientations (landscape + portrait) and test all durations (15sec, 20sec, 30sec, 45sec)
4 is the magic number of creatives per campaign.
Creative language - there can be up to 4 creatives for each language in a campaign. At least one English creative must be included in each localised campaign.
Creative set combinations
When combining the best performing video with a new end card with a new creative set id, the model can serve it to users who saw the video in the past, and this could affect IPM. If this happens, try matching the new end card with a new video to re-assess the performance of the new end card.
It is recommended to test different combinations of videos and end card to find the best performing combinations.
How to analyse and manage your creatives:
The best performing creative has the highest SOV of impressions yesterday, per campaign, per channel. If there are less than 10K impressions yesterday, look at the last 3 days.
Go live with new creatives on the best performing campaign with high traffic, and then upload the successful creative to all other campaigns.
Add new creatives to the leading campaigns. A wide targeting audience and high scale helps to accelerate the learning phase. Once the creative is confirmed as a success add it to all campaigns.
For A/B testing, create a dedicated test campaign for concepts, end cards, orientation, and video length.
Pausing creatives:
Pause creatives that have been underperforming with low scale, low IPM and or low ROAS. These creatives can be replaced with new ones.
Creative Learning Phase Explained
The ironSource model:
The ironSource creative model learns the creative’s IPM during the learning phase to evaluate the performance potential.
The learning is done separately for each bundle per channel per OS.
The wider the targeting is, the faster the learning phase will be completed.
The more new creatives you have, the slower the learning will be per creative as the impressions are split between the new creatives.
The ironSource model continuously learns and updates the best way to serve creatives. The exploration logic allows different creatives to scale over time according to the performance.