AI creative moves past the biggest brands
For a while, AI creative production looked like a stunt reserved for the companies with the biggest media budgets and the most patience for experiments. Unilever, Coca-Cola, and L’Oréal were early to test the waters, which made sense. They run huge product portfolios, ship into many markets, and can’t afford to wait around for every banner, social cutdown, and paid variation to come out of one slow production line.
That original wave was never really about novelty, though. It was about volume. If you’re selling shampoo, hotel rooms, sports betting, or consumer health products, you rarely need one polished ad and a round of applause. You need dozens of versions, then dozens more. Different audiences want different copy. Different channels want different formats. Different markets need different language, legal disclaimers, imagery, aspect ratios, and sometimes completely different calls to action. Traditional workflows can do that, but they do it with a lot of handoffs, review cycles, and calendar shuffling that can make a simple resize feel like a small civic project.
In marketing, the hard part is often not making one good ad. It’s making the next 50 without slowing the whole team down.
That’s the spot AI has started to occupy for a wider set of brands. The conversation is no longer limited to the earliest adopters with giant budgets and internal labs. It has moved into the upper-middle tier of marketing teams, the ones large enough to feel the pain of content demand every day, but not so large that they can spend a year building a custom system from scratch. For those teams, generative AI marketing tools are less a moonshot than a practical response to a pile of unfinished briefs.
Wyndham Hotels & Resorts, Opella, and BetMGM sit in very different businesses, yet they’re running into the same basic constraint: how do you produce far more creative assets without turning the whole operation into a bottleneck? A hotel chain needs fresh imagery and promotions for different properties, seasons, and traveler segments. A consumer health company has to adapt content across languages, markets, and product lines. A sportsbook or casino brand faces relentless demand for new assets, but also more scrutiny than most advertisers because the category is watched closely. Different industries, same headache. Too much content to make. Too little time to make it.
That’s why the middle of the market has become so interesting. The companies pushing into AI creative production now are not all giant enterprises with sprawling research teams. Many are brands with strong in-house creative teams that still need help getting work out the door faster. They want systems that support the people already inside the shop, not a six-month software science project that requires a small army of consultants.
Platforms like Pencil and Adora are making that easier. They sit in the practical middle ground between “do everything manually” and “build a bespoke enterprise stack that nobody wants to maintain.” For brands that want speed without signing up for a full custom build, that matters. These tools can help teams generate variations, organize creative production, and keep the process moving without turning every campaign into an IT ticket.
The result is a broader rollout, not a brand-new idea. The early headlines belonged to the biggest brands because they had the scale, the budgets, and the tolerance for experimentation. Now the approach is spreading to marketers who simply have too much content to create the old-fashioned way. In the next section, the real question gets more interesting: once those teams adopt the tools, how are they actually using them day to day?
How Wyndham, Opella, and BetMGM are actually using it
Once you get past the broad enthusiasm, the real story is pretty ordinary in the best possible way. These teams are using AI to solve production bottlenecks. Not to make a splash. Not to chase novelty. They need more versions of good creative, faster than a traditional workflow can comfortably manage, and they need those versions to keep working once they land in different channels, markets, and regulatory environments.
The useful question isn’t whether AI can make something once. It’s whether it can keep making the right variations without turning the whole team into a production queue.
Wyndham gives a good example of what that looks like when a brand has a lot of inventory to promote and a lot of audiences to reach. The company, which brings in roughly $1.4 billion a year, has spent about a year working with Adora to generate stills and video for paid and owned Instagram activity. The starting point isn’t synthetic fantasy art or some abstract prompt about “travel vibes.” It’s real hotel imagery. That matters. A room, a lobby, a pool deck, a hallway shot that already belongs to the brand can be turned into a set of new assets without re-shooting the whole thing every time the team wants a fresh angle.
The workflow sounds fairly practical. Wyndham uses media mix modeling to understand where spend is moving the needle, then pairs that with feedback from Meta’s platform signals so the creative team can move faster on what to keep, what to change, and what to kill. In plain English, the loop is shorter. A variant that performs well can be pushed again with a different crop, a different line, or a slightly different visual treatment. A weaker one gets cut before anyone wastes another week polishing it.
The reported gains are not tiny side effects either. Wyndham says it now produces about fifteen times as many assets as before. Time spent on concepting, production, and approvals has dropped by roughly three-quarters. Site visits have also improved by about six times versus the benchmark. That last figure is easy to wave away if you’re allergic to marketing claims, but it still points to something useful: when creative testing gets faster and the asset pool gets wider, the team can learn more quickly which combinations actually move people.
Opella has taken a different route, and that difference is the interesting part. Rather than leaning on an outside vendor, it runs the work in house. The company’s 85-person team uses generative AI for strategy briefs, animated marketing assets, and mascot work. That includes Nigel, the owl tied to Xyzal, who now has an AI version. Which, to be fair, is exactly the sort of sentence that would have sounded like a weird April joke a few years ago. Yet here we are.
The setup lets Opella handle a huge amount of adaptation without rebuilding each asset from scratch. Output has climbed by around twenty times, and the team has produced more than 20,000 pieces of content. Most of that is not headline-grabbing campaign art. It’s adapted material for different languages and markets, the sort of work that used to eat up a lot of production hours and still had to be checked, approved, resized, and translated before it could run. That’s where AI fits neatly into digital asset production. It trims the repetitive stuff without asking the brand to abandon its own visual style.
If a team spends all its time recreating the same message in twelve formats, it stops having time for the message itself.
BetMGM sits on the more cautious end of the spectrum, which makes sense given the category. The company uses AI for imagery and short video spots, but it keeps some creative human-led because betting is heavily regulated and nobody wants to hand accuracy over to a machine that may get cocky about a number, a line, or a legal disclaimer. That restraint isn’t a sign that the rollout failed. It’s just how a regulated business behaves when it likes speed but likes compliance more.
BetMGM’s approach also shows that AI adoption doesn’t have to be all or nothing. Some parts of the process can be automated while others stay under direct creative control. The team can use AI to get visual options on the table faster, then reserve human judgment for the pieces that carry more risk. In a category where a single mistake can become a very expensive headache, that split feels less like hesitation and more like common sense.
Taken together, these three examples make the operating models pretty clear. Wyndham is using AI to speed up iteration across paid and owned social. Opella is using it as an internal production engine for multilingual, multi-market content. BetMGM is using it selectively, with guardrails built around a category that doesn’t forgive sloppy execution. Different businesses, different rules, same basic need: more creative output, less drag.
That is also why brand creative testing has become more practical. When production is no longer the slowest part of the process, teams can test more ideas before they commit budget to the ones that deserve a bigger run. The next question is why this shift is happening so quickly now, and why it is spreading beyond the biggest early adopters.
Why the shift is happening now
A year or two ago, AI creative production still looked like a luxury reserved for the biggest spenders, the brands that could afford custom plumbing, long vendor trials, and a few false starts along the way. That picture has changed. Part of the technical burden is now being handed off to platforms like Pencil, which counts some companies above the ten-billion-dollar revenue mark among its customers but is seeing its fastest growth among brands below the one-billion-dollar level. That detail matters. The people moving fastest aren’t always the giants with endless resources. They’re often the mid-sized marketers who need more output, want to test more ideas, and don’t have patience for a rigid enterprise package that takes forever to bend into shape.
When content demand keeps rising, the teams that win are usually the ones that can make more versions without turning the process into a small civil engineering project.
That appetite for speed has a lot to do with why the middle of the market is leaning in now. A brand like Wyndham Hotels, Opella, or BetMGM may not share the same category, but they all run into the same math problem. More channels. More audiences. More formats. More local variants. More approvals. The old workflow, where one concept gets polished, approved, resized, translated, re-approved, and then maybe shipped, starts to feel slow in a hurry. AI systems promise something else: a way to keep the creative team in motion without asking it to duplicate the same work fifty times.
The money side has improved too. Compute is cheaper than it was earlier in the year, and U.S. token prices are sitting roughly two-fifths below their spring peak. That does not mean generative production is cheap in an absolute sense, but it does mean the recurring cost of experimenting has come down enough for more teams to run with it. When every new image, script variation, or cutdown no longer feels like a budget event, people stop treating each test as a precious artifact and start treating it like part of the workflow.
That shift in economics would matter less if the tools still behaved like fickle interns with a design degree. They don’t always get everything right, but visual consistency has gotten better. Faces stay more stable. Brand colors are less likely to wander off. Layouts hold together more reliably than they did in the early waves of generative image tools. For creative leaders, that makes a big difference. The fear isn’t just “Will this look good?” It’s “Will this quietly drift off brand and make legal, brand, and regional teams lose their minds?” The answer is becoming less often yes, which is enough to change procurement conversations and shorten the distance between curiosity and adoption.
Demand is pulling in the same direction. Roughly four out of five CMOs expect they’ll need a lot more content soon. That’s not a small nudge. That’s a loud signal that the volume problem is not going away. Once the expectation shifts from “Can we make a few good ads?” to “Can we keep producing, localizing, and testing at a much higher pace?”, AI creative starts to look less like a novelty and more like a necessary production tool. The teams that can turn one concept into twenty usable variations without dragging the calendar into the mud have an obvious advantage.
A recent Kantar measure points to the same behavior on the ground. Early-stage creative tests rose by about a sixth between late 2025 and mid-2026. That suggests more brands are checking ideas earlier, before they pour real money into a concept that might flop after launch. Cheap and fast testing makes a lot of sense in that world. If you can try more versions up front, you can spend less time arguing about hunches later. Nobody enjoys a meeting where everyone discovers the campaign only worked on slide deck number one.
For marketers, the practical appeal is hard to miss. They want systems that move quickly, experiment quickly, and still leave room for something more tailored than a cookie-cutter enterprise setup. They also want a cleaner path from idea to output, because the bottleneck is rarely imagination. It’s production. That’s why the conversation has moved beyond whether AI can make something at all. The real question now is how much creative work it can absorb before the process starts to feel less like a custom art project and more like a repeatable operating model.
If you want a broader snapshot of how marketers are sorting through AI tools and use cases, HubSpot’s State of AI report is a useful benchmark.
The guardrails that still matter
The enthusiasm around AI creative production has moved past the “look what it can do” phase. The harder question now is simpler, and a lot less glamorous: where does it stop?
The fastest way to trust AI in creative is to decide, in writing, where it stops.
BetMGM gives a pretty clean example. It has a large in-house creative team, a serious media budget, and enough volume to make automation tempting. Even so, the company keeps some work human-only because gambling is a regulated business and nobody wants a machine freelancing near the parts that can cause real trouble. That includes betting odds, where the cost of a mistake is obvious. If a creative asset gets the number wrong, or presents the offer in a way that’s misleading, there’s no cute apology text that can patch over it later.
That caution matters because speed can make people sloppy in ways they don’t notice until a problem lands on legal, compliance, or customer support. In a category like BetMGM’s, the machine can help with imagery and short promotional clips, but it doesn’t get to own the whole job. The company’s approach is less “let the model cook” and more “you may stir the pot, but you are not in charge of dinner.”
Opella has drawn its own lines, and they’re just as practical. The company uses generative tools for strategy briefs, animated assets, and even mascot work, but it has put rules around medical imagery. It also uses AI disclosures where local rules require them, including in the European Union. That choice is easy to understand when you say it plainly: no fake doctors. A health brand can experiment with speed and scale, but it still has to treat medical trust like a live wire. If an image looks too polished, too synthetic, or too authoritative in the wrong way, it can create a problem that no amount of media efficiency will fix.
Wyndham is dealing with a slightly different version of the same issue. The company has already accepted that AI can help it produce more stills and video for paid and owned channels. What it has not settled is whether AI-generated people belong in hotel imagery at all. That sounds like a small detail until you think about what hotel marketing is actually trying to do. Guests want to picture themselves in a place that feels real, not in a brochure populated by synthetic smiles and impossible lighting. The debate is less about whether AI can make a person and more about whether that person belongs in the frame.
That’s the pattern across this upper-middle tier of AI advertising adoption. Brands are willing to use machines where the output is easy to check, easy to revise, and low-risk if it goes sideways. They are slower where the work touches regulated claims, medical trust, or anything that would make a compliance team reach for a stress ball. Reasonable, honestly.
As trust improves, the assembly-line style version of AI creative production will probably spread further. The companies that have already built internal rules will move faster because the boundaries are clear. The ones still testing the edges will keep asking awkward but useful questions about people, claims, and where a brand’s voice should never be left to a model.
Agencies won’t disappear from that picture. They’re still useful on the bigger, more traditional campaigns, and they still bring outside perspective that in-house teams can miss when they’re deep in the day-to-day grind. For now, that balance seems to suit everyone: machines handle more of the repetitive production work, while humans stay responsible for the parts where judgment matters and the jokes stop being funny.
