Dynamic creative gives you an outcome. Manual variations give you an explanation. If you need this month's CPA to be as low as the assets you already have can make it, dynamic creative optimisation will beat you at that job. If you need to know what to make next, DCO cannot answer, because the thing that makes it good at allocating is exactly what makes it useless at explaining.
What each approach can and cannot tell you
DCO takes a pool of hooks, bodies, CTAs and formats, assembles combinations at delivery time and moves spend toward whatever converts. Manual variations are finished ads you briefed deliberately, one variable changed against the others.
The difference is not automation versus effort. It is confounding. DCO earns its result by giving the promising combination more delivery, so the winner's numbers reflect both the creative and the better audience slices the algorithm fed it, and you cannot separate the two afterwards. A manual test holds delivery roughly equal on purpose, which is why it can attribute a difference to the creative.
| Dynamic creative | Manual variations | |
|---|---|---|
| Question it answers | Which combination should get budget in the next hour | Why did this argument beat that one |
| How delivery is allocated | Unevenly, on purpose, toward the current leader | Roughly evenly, on purpose, to keep the comparison clean |
| What reporting gives you | Asset-level breakdowns showing which assets appeared in winning delivery | A per-ad CPA and hook rate you can rank directly |
| Main confound | Allocation bias: winners got the better impressions | Sample size: each ad carries fewer impressions |
| What it cannot do | Tell you the reason, or invent the next claim | React within the day to a combination going stale |
| Time to a usable answer | Continuous, no clean stopping point | One to two weeks per question |
| Fails when | The asset pool has one claim in it and nothing to choose between | You run out of ideas worth briefing |
| Right choice when | You have a proven claim and want the last 15 percent out of it | Performance has plateaued and you do not know why |
Run only DCO for two quarters and the account arrives somewhere specific: performance exists, and nobody can say what the winning argument is. That is fine until the argument stops working, when you have no thesis to fall back on and no record of what has been tried.
How many assets does DCO need at your budget?
This is where most DCO setups go wrong. People upload five hooks, four bodies and three CTAs because the interface has slots, without checking whether the budget can deliver 60 combinations. The arithmetic is the same as any test:
cost per readable combination = read impressions × CPM / 1000
At a 5,000 impression read and a EUR 12 CPM, a combination needs about EUR 60 before its numbers mean anything. Divide weekly budget by that and you get the number of combinations the campaign can actually sample before it starts concentrating.
| Weekly test budget | Combinations it can sample at EUR 60 each | Asset set that produces it | Live variations in rotation |
|---|---|---|---|
| EUR 350 | 6 | 3 hooks x 2 bodies x 1 CTA | 6 |
| EUR 700 | 12 | 3 hooks x 2 bodies x 2 CTAs | 12 |
| EUR 1,400 | 23 | 4 hooks x 3 bodies x 2 CTAs | 24 |
| EUR 2,800 | 46 | 4 hooks x 4 bodies x 3 CTAs | 48 |
Substitute your own CPM: at EUR 20 a combination costs EUR 100 and every row halves. Pick asset counts so the product is at or below what your weekly budget can sample, not at whatever fills the form. Six differentiated combinations that each get delivery beat 60 that each get a trickle. Most accounts run best with 8 to 20 live variations per active campaign, and the 48-combination row is a large-budget setup where 48 permutations is not 48 ideas, it is 11 ingredients multiplied.
Why DCO is not a production strategy
DCO looks like it reduces the amount of creative you need. It raises it, for a reason that only shows up in month two.
Every asset is load-bearing across multiple combinations. In a 4 x 3 x 2 set one hook appears in six permutations, and when that hook fatigues all six go with it. You do not lose one ad, you lose a quarter of the campaign, and replacing it means briefing and shooting a new hook, not adjusting a bid.
Permutations also fatigue on the ordinary clock. Our 2026 fatigue benchmark, a synthesis of published platform and agency figures rather than our own experiment, puts most creative effectively dead within three weeks: CTR falls 15 to 20 percent in the first two weeks, then week three arrives as a cliff at 45 to 70 percent below launch. Decline on Meta prospecting begins around a weekly frequency of 2.5, and days to a 40 percent CTR decline run from 9 for food and beverage to about 28 for B2B SaaS.
DCO therefore sets a floor on throughput rather than lifting one. It burns the pool combinatorially and needs it refilled on the vertical's clock. The benchmark concludes that throughput, not talent, is the bottleneck, and DCO makes that more true.
The hybrid workflow most accounts should run
Neither approach is the workflow. The workflow is manual for learning, DCO for exploiting, on a split.
- Reserve 20 to 30 percent of creative budget for manual tests. The rest goes to DCO on claims that have already won. This is the only structural change that matters; everything below is detail.
- Ask one question per manual cycle. Three arguments by three hooks, nine ads, one variable per axis and everything else held constant. Standard creative testing discipline applies: if two things moved, you learned nothing.
- Read the margins, not the cells. Row means tell you which argument won across hooks. That is the finding you carry forward.
- Promote the winning argument into DCO as a family. Build four to six hooks and three bodies that make the winning promise differently, sized with the table above. Our Meta ad creative generator workflow covers producing those as 9:16 and 4:5 exports.
- Let DCO allocate, and stop interpreting it. Read asset-level breakdowns as inventory, not as test results: which assets still get delivery and which have gone quiet.
- Refresh on the vertical's clock. Waiting for CPA to move means you are already three weeks into decay.
- Return to step 2 when refills stop working. New hooks on a dead argument will not save it, and that is the signal the manual layer owes you a new claim.
The failure mode this avoids is common: DCO running for a year, every refresh a new cut of the same claim, performance drifting down while nobody can name what is no longer true.
Where Genyad fits, and where it does not
Genyad is our product, so discount accordingly. It turns footage you already own into finished video variations: upload once, and each variation is a fresh script, shot selection, voiceover, caption set and export rather than a re-cut of one timeline. Nine variations for a manual matrix is nine credits, one credit each, and Growth is €99 for 65 credits, roughly €1.52 a variation.
What it is not is a DCO engine. Genyad does not render modular hook, body and CTA slots recombined at delivery time; for that mechanism, Sovran ($99 a month, as publicly listed in August 2026) is built for it. No product-feed or CSV-driven template rendering either, which is Plainly's territory ($69 a month, same date), and prices move. No predicted performance scores, so nothing tells you which variation wins before you spend. It does not publish to Meta or TikTok, so you export and upload the files as ads or DCO assets yourself. No AI avatars, no static banners.
Frequently asked questions
Is dynamic creative better than manual variations?
Neither is better in general; they answer different questions. DCO reliably extracts more performance from an asset pool you already trust, because it reallocates within the day. Manual variations are the only way to find out why something works, because they hold delivery roughly equal instead of concentrating it on the leader.
Can I use Meta's asset-level reporting as a creative test?
Not as a clean one. The breakdown tells you which assets appeared in delivery that converted, but the algorithm chose that delivery, so a winning asset's numbers include the better audience slices it was given. Use it to see which assets have gone quiet, and run a separate controlled test when you need a reason.
How many assets should I upload to a dynamic creative campaign?
Divide your weekly test budget by the cost of one readable combination, which is read impressions times CPM divided by 1,000. At a 5,000 impression read and a EUR 12 CPM that is EUR 60, so a EUR 700 weekly budget supports around 12 combinations, which is 3 hooks by 2 bodies by 2 CTAs. Uploading more than the budget can sample spreads delivery too thin to act on.
Does dynamic creative reduce how much creative I need to make?
No, it increases it. Each asset appears in several combinations, so one fatigued hook takes every permutation containing it down at once, and permutations decay on the same schedule as any other ad. Our fatigue benchmark notes that brands shipping 15 to 50 variants a month see 3 to 5 times longer campaign lifespan than quarterly refreshers, and a DCO pool has to be refilled at least that often.