In August, WPP stunned the industry with a 71% decline in ad revenue. For years, holding companies have relied on scale, data monopolies, and armies of strategists to stay ahead. That era is over.

The next generation of breakout agencies won’t lead with decks – they’ll lead with data.

For the first time, small and midsize agencies can access the same creative intelligence infrastructure that was once the sole domain of the big networks. The combination of language models, data clean rooms, and interoperable CDPs has eroded the old data moat. And the smartest independents are moving fast to seize the advantage.

The creative data paradigm shift

“Creative Data has never been on the table until language models were released. Creative Data is where financial data was in 1985 when people first realized computers were good with numbers. The same monumental paradigm shift is happening here… and grey areas are where art thrives.”
— Orlando Wood, CEO, KoobrikLabs

Language models don’t just tag content as “positive” or “negative.” They parse the nuance in billions of social posts, reviews, chats, and ratings.

Peer-reviewed studies show that models now excel at aspect-based sentiment analysis (ABSA) – identifying how people feel about different parts of a brand, product, or service within the same post (ACL Anthogoly; arXiv). Sarcasm detection, stance analysis, and intent modeling have also taken major steps forward

This turns what used to be marketing noise into actionable creative intelligence, in days rather than months.

The data moat has leaked

Big agencies no longer have a monopoly on data. Google’s reversal on third-party cookie deprecation has forced the industry to embrace a hybrid privacy model – legacy IDs + first-party data + clean rooms – standardized by IAB Tech Lab.

For independents, this is a once-in-a-generation opportunity. The infrastructure they can now rent and configure – CDPs, warehouses, clean rooms – was prohibitively expensive just a few years ago.

Meanwhile, holding companies are validating the direction with massive investments: Havas has committed €400M through 2027, and WPP is betting big on its “WPP Open” AI OS

But scale is no longer the determinant of insight velocity.

Early adopters that re-architect their workflows around AI are seeing 12–18% margin gains once reporting and creative cycles are integrated.

From polarity to precision

Classic social listening tools forced every post into positive / negative / neutral buckets. Not only is it not giving the full picture, to even get to that limited level of analysis required HUGE amounts of human input. LLMs break that trap. They can boil the oceans and structure the data and sentiment details around a creative output into actionable and analyzable data.

Imagine this tweet:

“I usually love Brand X, but this new product is a major misstep. I hope they get back to last year’s form.”

The creative implication? Message a “return to standards” and offer transparency to loyalists. Multiply that by millions of posts, and you get a creative analytics engine that feeds briefs and A/B tests every week

A 90-day plan to build a data-first shop

KoobrikLabs has distilled this into a pragmatic, three-month roadmap:

1. Days 0–7 – Stand up your warehouse & CDP, connect social/review/CRM data.
2. Days 8–30 – Ship your first LLM pipeline (entity/aspect extraction → sentiment →stance → emotion).
3. Days 31–60 – Wire insights into creative sprints, run 10–20 A/B/C tests.
4. Days 61–90 – Standardize briefs, reporting, QA, and model improvement loops

The future belongs to data-first creatives

This is about arming creativity with better intelligence. The agencies that win the next decadewill be the ones that treat creative data as seriously as media data—because the two are now intertwined.

If you’re a small agency, this is your moment. The tools are ready. The data is abundant. The incumbents are slow.

Get in touch with KoobrikLabs to start your 90-day data transformation.

Every few weeks, the tech world experiences what can only be described as a collective fainting spell.

A new AI model is announced (in this case, Google’s Nano Banana) and before the demo has even finished buffering on YouTube, a chorus of doom is already in full voice.

“This will kill creativity,” they say.  “Kill the human spirit… Kill Adobe!”

One can practically hear the tolling bells for entire industries.

There’s a ritual to this by now: the launch video lands; the breathless headlines follow; the op-eds sharpen their knives.

The technology is presented as a sort of messiah and executioner rolled into one, destined to both liberate and obliterate in equal measure.

And then… inevitably… people actually try it.

Have you?  Tried doing with Nano Banana what the demo promised?  Yes, so have I.  And no, it doesn’t deliver.  Not yet.

A Stanford CRFM study this year found that 92% of generative AI feature launches underperform their demo claims in early evaluations.  A reminder that Nano Banana isn’t a glitch in the system; it IS the system.

But this is all choreography.  And it’s been going on since the first PowerPoint pitch deck met a venture capitalist’s laser pointer.

CB Insights counted 387 GenAI product feature announcements between January and September 2025.  That’s more than one per day.

At that pace, even the most over-caffeinated futurist could be forgiven for feeling faint.

The Quiet Superpower

While the industry hyperventilates over each new feature, something less theatrical, but far more consequential, is happening beneath the headlines.

From their first day in the world, these models have excelled at a very particular task…

Drumroll please…

Structuring previously un-structurable DATA!

Yeah… I know… it doesn’t sound that exciting…. But it is.

They are prodigious librarians of chaos.  They can take a thicket of language, story, images, or dialogue and turn it into something you can query, map, and build on.

That’s their native skill.  And for the creative industries, it’s TRANSFORMATIVE.

If you’re wondering, creative data is almost entirely unstructured.

Are you seeing where I’m going with this?

IDC estimates that over 90% of enterprise data is unstructured, and that the figure is closer to 95–97% in media and creative industries.  A 2025 MIT Sloan survey found that companies using GenAI primarily to structure data, not to generate it, move from insight to action three times faster than their peers.

A screenplay has a semi-structure, sure, but nobody would confuse Casablanca with Anora.  Yes, both are love stories, but to equate their data structures is to miss the point entirely.

What unites them is not the structure, but the mess.  And AI is finally giving us the tools to turn that mess into something legible.

The Context Window Revolution

Take the recent release of Claude 4.5.  The headlines swooned over the usual laundry list of “smarter, faster, better” claims.

But the real breakthrough was almost buried: the context window finally works.

Claude 4.5 can process up to 1 million tokens, and Anthropic claims 98% accuracy in recall across large documents, compared to roughly 55–65% in previous iterations. Their internal benchmarks show a 70% improvement in mid-document reasoning, effectively closing what I’ve long thought of as the “amnesia gap.”

This is HUGE!

For years, feeding long documents into a model was like whispering to someone with short-term memory loss…  They’d remember the opening, maybe the closing, and little in between.  Now, for the first time, these models can hold and reason across large swaths of information in one go.

Powerful… and nobody’s talking about it.

From Scripts to Systems

At Koobrik, this is exactly where we’ve staked our flag.  While others chase the mirage of AI-generated genius, we’re databasing the data inside scripts for studios, independent producers, and agencies.

Every prop, scene, location, stunt, and character… structured, searchable, and ready.

On the analytical side, this structured creative data allows us to correlate box office performance, actor value by territory, story structures by genre, and surface development insights that once required armies of assistants and months of manual work.

McKinsey estimates that automating metadata extraction in media workflows can cut 30–50% of pre-production time, especially in script analysis, casting prep, and budgeting.

For the first time, creative data is genuinely on the table.

What once took thousands of human hours can now be synthesized in days… sometimes hours.

The Real Shift

So while the industry keeps peering anxiously at the horizon, waiting for some shiny new feature to either destroy or save us, the real story is already happening quietly at ground level.

The models are not yet the auteurs of our age.  They are the archivists.  The cataloguers.  The sexy Librarians.

The unsung data clerks making the creative industries legible to themselves for the first time.

And that’s not something to panic about.  It’s something to build on.

✨ Closing Thought

In time, Nano Banana will probably do everything its demo promised, just as Photoshop eventually did, and the iPhone eventually did, and every other “underwhelming” first draft of the future has done.

(Remember how shit the first iPhone was?  That beer drinking app was the only thing it did well…)

But the creative companies that win will be the ones quietly turning their unstructured creative gold into structured, queryable, actionable intelligence.

Forrester’s 2025 agency benchmark found that firms applying GenAI to data intelligence, not content gimmicks, saw 12–18% margin gains.

It’s not the flash that wins.  It’s the plumbing.

That’s the story worth paying attention to.

Los Angeles has always been the global capital of storytelling – a place where ideas, talent, and technology collide to shape culture. But in recent years, a new creative partner is changing the rules: large language models (LLMs). 

In studios, agencies, post-houses, and sound stages across LA, generative AI is transforming how stories are pitched, written, edited, and promoted. Tools like ChatGPT, Claude, and custom-trained models are helping creatives move faster – writing drafts, suggesting titles, localizing scripts, or generating entire brand narratives in minutes. 

But with speed comes exposure. As LLMs become embedded in day-to-day production, the conversation has shifted from what these tools can do – to how they’re being governed. 

Creative acceleration, legal lag

The creative sector has embraced LLMs faster than any regulatory framework can track. According to Artificial Intelligence Max, AI adoption across media and content production has grown over 1310% since late 2022. But for many teams, that growth is unstructured, unmonitored and legally risky. 

Creative professionals aren’t waiting for formal policies. Shadow AI is now a reality in agencies and studios, where individual team members plug content into free tools without oversight. Gartner flagged this in their 2025 Enterprise Risk Index, placing “Shadow AI” and “AI information governance” among the top five risks across all sectors. For a deeper dive on managing Shadow AI in creative environments, check out our blog on Shadow AI. 

These untracked uses go beyond data privacy concerns – they risk copyright violations, brand misalignment, misinformation, and bias. 

Why LA’s creatives are uniquely exposed

Los Angeles operates at the intersection of intellectual property, personality rights, and public influence. That makes unregulated use of LLMs particularly risky. Consider: 

In a city where intellectual property is both the product and the currency, even unintentional misuse of generative AI can lead to serious legal and reputational consequences. 

A shifting compliance landscape

Regulations are evolving rapidly. Beginning in 2026, California’s AI Transparency Act will require clear disclosure of AI-generated content across multiple industries, ensuring transparency in all uses of generative AI. 

The California Consumer Privacy Act (CCPA) strengthens consumer rights over how their data is used to train AI systems. Meanwhile, industry bodies like SAG-AFTRA and the MPA are revising contracts and usage clauses to address AI in writing, editing, and production pipelines. Globally, the EU AI Act has already introduced risk tiers for generative tools. Even at a national level, the FTC and Copyright Office are scrutinizing AI outputs for fair use, ownership, and ethical standards. 

Creative firms are finding themselves at a crossroads: embrace the efficiencies of AI, or risk falling behind – but only if they can do so with trusted compliance measures in place. 

The compliance gap inside creative workflows

What makes LLM compliance uniquely difficult in creative environments? It’s not just the speed or scale – it’s the ambiguity. 

Traditional legal and compliance teams often don’t have the tools to audit or evaluate LLM behavior at scale. Manual review is too slow. Risk scoring is inconsistent. And creative teams don’t want to be slowed down. 

Creative without compromise: building responsible AI workflows

To maintain trust with clients, audiences, and partners, creative firms need a system that balances innovation with governance. That means: 

And most importantly, doing it all without slowing down the work itself. 

Enter CALLM by Koobrik Labs

This is where CALLM comes in – a managed compliance framework for LLMs, designed specifically for creative teams navigating rapid AI adoption. 

CALLM gives studios, agencies, and production companies a simple way to test, monitor, and govern generative tools – without disrupting creative flow. It runs pre-use evaluations using 100,000+ prompts, scores risk across legal, ethical, and brand dimensions, and continuously tracks performance and compliance. 

No extra admin. No data leaving your environment. Just structured oversight that fits the way creatives already work. 

Want to see how CALLM can support your AI workflows? Let’s talk.