Dynamic Creative Optimization looks lazy. Upload a pile of videos, headlines, and CTAs, let the algorithm assemble permutations, walk away. That is how most UA teams think about DCO, and it is why most UA teams do not get much out of it.
Used right, DCO is not a brainstorming tool. It is the best way to scale creative that is already proven.
The discipline is counter-intuitive. DCO is not where you find out whether a concept works. That happens in manual testing. DCO is where you take a concept that has already worked, break it into modular components, and let the algorithm find combinations that perform even better than the single winning ad you shipped. Use DCO for exploration and you waste budget on combinations the algorithm kills within days. Use it for exploitation and you compound the wins you already have.
Below: what DCO actually is (and is not), how to set it up on Meta specifically, and the campaign structures that make the difference between DCO as a shortcut and DCO as leverage.
Page Contents
What is DCO, and what it is not
DCO is a platform feature that automatically assembles and serves combinations of creative components (videos, images, headlines, primary text, CTAs) based on predicted per-user performance. Instead of you pairing a specific headline with a specific video, the algorithm tests combinations at a speed no human team can match.
This post focuses on Meta’s DCO specifically, because that is where the discipline matters most and where the creative modularity pays off. Meta calls it Advantage+ Creative (the current name for what used to be “Dynamic Creative”). You upload individual assets as components inside a manually targeted ad set, and the system serves different combinations to different users based on predicted response probability.
Meta DCO is not the same as TikTok Smart Creative or Google UAC
Conflating these three is the most common source of confusion in DCO conversations. They are not the same product, and the discipline does not transfer.
Meta DCO (Advantage+ Creative) gives you component-level control inside a manual or Advantage+ campaign. You decide the audience and bidding; the algorithm assembles creative. This is where the modularity discipline pays off most directly because you can see which component combinations drive which outcomes.
TikTok Smart Creative is a lighter version of the same idea. You upload multiple video and text variations; TikTok pairs them. It is newer, supports fewer component types (video and text only, no separate headline and CTA controls), and the learning signal is less granular than Meta’s.
Google UAC is not really DCO in the Meta sense. Everything in UAC is automated: targeting, bidding, and creative assembly. You cannot turn DCO off because there is no “manual” option. UAC is its own campaign type that happens to include creative optimization, not a creative feature you bolt onto a manually targeted campaign. Treating Google UAC as an extension of Meta DCO is how teams misallocate budget inside Google.
For this post, “DCO” means Meta DCO unless otherwise noted.
What we see on Meta DCO
DCO wins when fed proven creative, loses when fed brainstorming material. In our experience, DCO ad sets that combine 4 to 6 modular variations of an already-winning concept consistently outperform the original single ad. DCO ad sets built from 4 to 6 completely different untested concepts usually underperform, because the algorithm concentrates early spend on whatever shows the strongest CTR signal and kills the rest before they can learn.
The conversion threshold is not 50 per day or $500 per day. The real bar is simpler: each DCO ad set needs enough spend to generate at least 3 to 5 conversions per day. More is better. Below that, the algorithm cannot differentiate signal from noise across the combinations it is testing. Hit that minimum and the algorithm has something to work with. Far exceed it and component-level performance becomes readable.
Component modularity decides whether DCO ever learns anything. Every headline must work with every video. Every CTA must align with every primary text. The moment you ship a headline that references a specific video (“Watch how Sarah lost 30 lbs”), every other pairing of that headline in the set becomes nonsense. Modular components let the algorithm mix freely. Non-modular components poison the test.
Need help scaling your mobile app growth? Talk to RocketShip HQ about how we apply these strategies for apps spending $50K+/month on UA.
What we learn
DCO is not lazy. Using DCO the lazy way is lazy.
The right mental model is exploration vs exploitation. Manual testing is exploration: you run fully produced single ads against each other to find which concepts work. DCO is exploitation: once you have a winner, you break it into components and let the algorithm find the best combination within the winning framework. If you flip those two, DCO underperforms because the algorithm is making decisions on early CTR signal before conversion data stabilizes.
Three common mistakes follow from treating DCO as a brainstorming tool.
First, using DCO for creative testing instead of proven creative. Throwing 5 untested concepts into one DCO ad set and hoping the algorithm finds the winner does not work. The algorithm will kill 4 of them within days based on early signal, and you will never learn whether those concepts could have worked with different hooks or copy. Test concepts first in manual ad sets. Only promote proven winners into DCO.
Second, under-budgeting the DCO ad set. Forget the fabricated thresholds you see in most DCO guides. The real minimum is 3 to 5 conversions per day per ad set, and more is better. If your CPI is $5 and you are running a DCO ad set on a $30/day budget, you are generating fewer than one conversion per day. The algorithm has nothing to optimize against.
Third, designing non-modular components. Headlines that reference specific videos. CTAs that only make sense paired with specific primary text. Every combination the algorithm tries with those components is broken. Every component must stand alone and combine cleanly with every other component.
Campaign structures that make sense
- Validate concepts in manual ad sets first. Run 3 to 5 fully produced single ads for a week. Let them fight. The winners are the concepts that earn promotion into DCO. Everything else stays in manual testing or gets killed.
- Build DCO sets around one proven concept at a time. Take the winning concept and produce 4 to 6 video or image variations that all express the same underlying concept but with different hooks, opening frames, or pacing. Add 3 to 4 headlines (all written as standalone value propositions, not dependent on any specific video), 2 to 3 primary text variations, and 2 CTAs. Do not mix multiple concepts inside one DCO set.
- Budget for 3 to 5 conversions per day per DCO ad set, minimum. More is better. Below that floor, the algorithm cannot differentiate component performance from noise. If your CPI is $5, that is roughly $15 to $25/day per ad set as an absolute minimum to make DCO worth running at all. Scale the budget with your CPI.
- Ship all three major asset formats. Vertical (9:16) for Reels and Stories. Square (1:1) for Feed. Each placement performs differently, and the algorithm serves different combinations to each. If you only ship vertical, you are leaving Feed performance on the table. If you only ship square, you are not optimized for Reels, which is usually the cheaper placement. Produce every winning video in at least vertical and square; landscape is optional for most app install campaigns.
- Refresh components when signal stabilizes, not on a fixed calendar. Once component-level performance rankings settle (usually after the ad set has accumulated enough conversions that the rankings stop changing), replace the bottom-performing 20 to 30 percent with fresh variations inspired by the top performers. Do this when the data tells you to, not on an arbitrary two-week cadence.
- Read DCO results at the component level, not the ad level. The whole point of DCO is component-level learning. Use Meta’s Breakdown by Asset view to see which videos, which headlines, which CTAs drove conversions. Feed those insights back into your next round of manual testing and your next DCO build.
Frequently asked questions
When should you use DCO versus manual creative testing?
Use manual testing to validate new concepts. Use DCO to scale concepts that are already proven. The two work in sequence, not as alternatives. A concept graduates from manual testing to DCO once it shows a CPI within striking distance of your target for several consecutive days. At that point, the question becomes which component variations maximize it, and DCO is the tool for answering that question.
How many assets should you put in a DCO ad set?
Around 4 to 6 video or image assets, 3 to 4 headlines, 2 to 3 primary text variations, and 2 CTAs per ad set. That generates enough combinations for the algorithm to find winners without fragmenting delivery so thin that no single combination accumulates meaningful data. Overstuffing with the maximum 10 videos dilutes delivery and kills learning.
What budget do you need to run DCO effectively?
Enough to generate at least 3 to 5 conversions per day per ad set, and more is better. There is no universal dollar floor because it depends on your CPI. Below that conversion threshold, the algorithm cannot differentiate component performance from noise.
Can DCO replace a creative strategist?
No. DCO optimizes combinations; it cannot generate concepts. It tells you that Headline B paired with Video C outperforms other pairings in the current set. It cannot tell you that a completely different creative angle (say, an identity-transformation hook you have not tested) would outperform everything in the set. The strategic layer of identifying audience motivations, competitive gaps, and untested angles remains fundamentally human.
How do you design headlines and text for DCO interchangeability?
Write every headline and every piece of text as a standalone value proposition that works with any video in the set. Do not reference specific video content. Do not assume a reader has seen a particular opening frame. Each component is a building block, not a puzzle piece.
What asset formats should you include in a DCO set?
At minimum, vertical (9:16) and square (1:1) versions of every video. Vertical performs on Reels and Stories. Square performs on Feed. Each placement is a different inventory with different user behavior, and the algorithm serves different combinations to each. Landscape (16:9) is optional for most app install campaigns.
How does DCO work on TikTok and Google compared to Meta?
They are not the same. TikTok Smart Creative is a lighter version of Meta DCO (video and text combinations only, no separate headline or CTA controls). Google UAC is not really DCO in the Meta sense: it is a fully automated campaign type that includes creative assembly but does not let you turn automation off. Treat Google UAC as its own system, not as an extension of Meta DCO. This post focuses on Meta DCO specifically because that is where the creative modularity discipline matters most.
What are the three biggest mistakes in DCO setups?
Using DCO for testing instead of scaling proven creative. Under-budgeting so the ad set never hits 3 to 5 conversions per day. Designing non-modular components (headlines that reference specific videos, CTAs that only work with specific primary text). Any one of these is enough to make DCO worse than a well-crafted single ad.
Looking to scale your mobile app growth with performance creative that delivers results? Talk to RocketShip HQ to learn how our frameworks can work for your app.
Not ready yet? Get strategies and tips from the leading edge of mobile growth in a generative AI world: subscribe to our newsletter.
Related reading
