The bottleneck was never the optimization engine
Most conversations about AI for dynamic creative optimization focus on the wrong half of the problem. The pitch is familiar:
dynamic creative optimization is a programmatic advertising technology that automatically assembles and serves personalized ad variations in real time, using audience data, context, and performance signals to pick the best creative combination for each impression.
The machine learning that picks the winning combination is genuinely good and getting better.
But picking is only as good as the menu. DCO can only assemble variations from the modular parts a team has already produced — the headlines, images, CTAs, and offers loaded into the template.
DCO breaks an ad into modular parts and lets machine learning assemble the right combination for each user as the ad loads, replacing manual A/B testing with automated, continuous optimization.
If a team feeds the system three headlines and four images, the algorithm optimizes within twelve combinations. No model, however sophisticated, can test a variation that was never built.
That is the quiet tension underneath most DCO programs. The optimization layer is automated and infinite; the creative supply feeding it is manual and scarce. Teams buy a powerful engine and then starve it. The interesting question for buyers is not "how good is the optimization AI" but "what is actually limiting the number and quality of variations my AI has to work with."
The market keeps optimizing a menu that's too short
The DCO category has matured around the assembly-and-serving step.
Although DCO has existed as a marketing technology since the early 2010s, its adoption has surged in recent years thanks to AI and machine learning advancements.
Vendors compete on how fast they assemble, how precisely they match, and how tightly they integrate with the major buying platforms. Those are real improvements. They also address a stage that was never the constraint for most teams.
Two structural patterns explain why creative supply stays thin. The first is rule-based assembly.
DCO comes in two main types: rule-based and AI-driven, where rule-based DCO follows predefined conditions set by advertisers.
Rule trees still depend on a human authoring every branch and every asset that branch points to, which caps variety at whatever the team had time to produce. Industry analysts have flagged exactly this ceiling:
the problem with DCO is that it is based on humans creating a series of decision trees, which is what drives the different creative units developed.
The second pattern is the production process itself. Producing on-brand variation is slow, and the slowness sits before the ad ever reaches a DCO platform.
Even a single campaign asset like an email or an ad can turn into weeks of tickets, design requests, and legal reviews, blocking marketers from executing the way they want while designers get stuck on busywork like resizing assets and swapping copy.
When each new variation costs a design ticket, teams ration. They ship the safe four images, the algorithm optimizes among them, and the ceiling on performance is set long before any impression is served.
This matters more now because the buying platforms have changed what they reward.
Ad platforms are explicit about what drives performance: variety, volume, and relevance. Meta tells advertisers to develop ads that are truly different in look, feel, storyline, and message; TikTok wants participation in trends; Google wants freshness across every campaign. Manually producing the volume of unique creative the algorithms actually reward is impossible.
A short menu does not just limit DCO — it actively underperforms against the automated bidding systems now consuming most of the budget.
What to look for: supply, signal, and brand control
The most useful evaluation criterion for any AI for dynamic creative optimization purchase is whether the system addresses creative supply, not just creative selection. A buyer should pressure-test three things.
Can it generate new on-brand variations, not just recombine existing ones? Generative AI promised infinite creative and mostly disappointed, because the output looked generic.When AI tools for content production first emerged, everyone was excited by the possibility of unlimited content and then quickly disappointed — the generated assets rarely captured a company's visual language, tone, and design standards, making them unusable.
The fix is grounding. Platforms like Hightouch take this approach with Content Assembly, where
agents search a company's systems such as DAMs, Figma, Adobe, and Google Drive and intelligently assemble content from existing assets, including images, templates, and style guides.
The variation is new, but the raw material is the brand's own approved work.
Does it connect to real performance data, or optimize in a vacuum? Good creative supply should be informed by what already works.Strong systems build on complete analytical context, surfacing strategic insights from all of a team's data and integrating with the entire data stack, including ad platforms and the data warehouse, so teams analyze their best and worst-performing ads and use that to create winners.
This is the link DCO usually misses: the optimizer reports which combination won, but nothing feeds that learning back into producing better next variations.
Does the AI understand the brand well enough to be trusted at volume? This is where most tools fail. Producing a thousand variations is worthless if a fraction go off-brand. The standard worth holding vendors to is operational brand knowledge — not a static guideline PDF, but a queryable layer the AI reasons against.Most AI tools only know brand guidelines such as font, colors, and tone; but AI needs to understand what a design team actually does with those components to capture the feeling of a brand.
When that grounding exists, governance stops being a brake.
When AI is grounded in existing assets, guided by brand rules, and informed by what's worked in past campaigns, the question of whether output will be on-brand no longer exists.
The full loop: from signal to variation to served impression
Here is what the complete cycle looks like when supply and selection are connected rather than handed off between disconnected tools.
It starts with data, and the quality of that data sets the ceiling.
Strong systems run on a few core elements: data is the backbone, customer data provides insight into needs, preferences, and past actions, and a full view of the customer makes it possible to design ads that truly fit.
A unified, identity-resolved view of the customer is what lets the system know which audiences and journey stages even need distinct creative.
The second input is identity, which keeps personalization coherent across a fragmented media landscape.
Identity resolution makes sure the right person gets the right ad across devices and channels, keeping personalization consistent and avoiding missed or fragmented experiences.
This is the role a customer data warehouse plays underneath the creative layer — and a structural reason warehouse-native architectures matter here. Hightouch keeps this foundation in the customer's own warehouse as a Composable CDP, so the data informing creative never has to be copied into a separate proprietary store to be used.
From there, the production step generates variation at the volume the algorithms reward. Agents take a brief, pull approved assets, and produce dozens or hundreds of on-brand options, with compliance checked before a human ever reviews.
Custom agents grounded in legal and brand guidelines perform an initial review and catch issues early, and then legal, brand, and design teams approve the work.
That moves brand and legal review from a weeks-long gate to a fast confirmation step.
Then the optimization engine does what it does best — assembling and serving the right combination per impression and reporting which elements drive results.
As each variation is served, AI continuously tests and optimizes creative combinations, gathering insights advertisers can use to refine future campaigns and improve performance.
The difference is that those insights now flow back into production. Instead of a one-way report, the loop closes: performance signals shape the next batch of variations, which feed the optimizer, which produces sharper signals.
Notably, the conversion signals a team sends back to the platforms shape this loop as much as the creative does.
Conversion signals are the real-time fuel for the ad platforms' AI systems, and better signals improve budget allocation, bids, and targeting, resulting in higher ROAS.
A team optimizing creative against weak signals is sharpening the wrong target — which is why the data foundation and the creative engine belong in the same architecture, not stitched across separate vendors.
What success actually looks like
The outcome state is not "prettier ads." It is a shorter cycle time and a creative supply that finally matches the algorithms' appetite. The clearest published signal of the gain comes from teams running this connected workflow:
Early adopters are seeing reducing campaign production time by up to 70% while also seeing measurable performance gains.
Time-to-launch is the metric that compounds.
An integrated, agentic workflow takes a team from insight to idea to launch in minutes instead of weeks, and everything actually looks on-brand.
Faster cycles mean more tests, and more tests against a richer menu mean the optimization engine has more to learn from. The DCO algorithm that was previously starved now has enough variety to do its job.
The reframe for buyers is straightforward. The constraint on dynamic creative optimization is rarely the optimizer's intelligence. It is the rate at which a team can produce trustworthy, on-brand, data-informed variations to feed it. Solve supply, and selection gets dramatically more valuable. Leave supply unsolved, and even the best optimization AI is rearranging the same short menu.
Choosing tools that fix the real constraint
When evaluating AI for dynamic creative optimization, buyers should weight the upstream questions heavily. Does the tool generate genuinely new variations from approved brand assets, or only recombine a fixed set? Is it connected to live performance data and a unified customer view, or optimizing blind? Does it embed brand and legal review into the workflow so volume does not become risk? And does the customer data underneath stay in the team's own infrastructure, or get copied into another vendor's store?
The best results come when these pieces work together rather than as isolated point tools. As one industry framing puts it,
when a CDP, DCO, and the rest of the martech stack all work in concert, teams move beyond fragmented campaigns toward a truly integrated, one-to-one marketing strategy.
The category is moving from optimizing a static menu toward continuously producing and refining the menu itself — and that shift, more than any improvement to the selection algorithm, is what will separate the programs that scale from the ones that stall. Teams weighing the move can start by examining how an agentic advertising workflow connects production, data, and delivery in a single loop.