When I was running programmatic advertising operations in the early days of Telhaverde's predecessor company, the creative supply chain looked like this: a campaign brief went to an agency or internal creative team, came back as a set of master assets — usually a handful of sizes for display, a video spot, some copy variants — and then those assets fed into a distribution system that placed them programmatically. The limiting factor on creative performance was always the input side. You had ten creative variations. The algorithm could optimize across ten things.
Programmatic advertising taught media buyers to think about scale as an algorithm problem. Maximize impressions, optimize bidding, improve targeting. But the real performance variable was always the creative — and creative was frozen in a supply chain that was too slow, too expensive, and too linear to respond to the signals the distribution system was generating. You could see from your dashboards that audiences in one segment responded 3x better to a particular message frame, and you couldn't act on that insight because producing more variations of that frame took six weeks and a budget conversation.
That structural problem is what generative AI is actually solving in the creative supply chain. Not incrementally — structurally.
Where the supply chain is being restructured
The creative supply chain has several distinct stages: strategy and brief, concept development, production, review and approval, versioning and adaptation, distribution, and performance measurement. Generative AI is having meaningfully different impacts at each stage, and the honest answer is that some stages are being substantially restructured while others are more resilient to the change than the hype suggests.
Production is the stage with the most dramatic change. Writing copy, generating image concepts, producing video variations — all of these have gotten orders of magnitude faster and cheaper. A campaign that previously required a week of agency production time can have fifty variations ready in an afternoon. This is not an exaggeration. The companies in our portfolio that work in ad creative automation are running this at scale for growing brands that are too small to afford traditional agency production at the volume their digital distribution requires.
Versioning and adaptation is the second stage with significant disruption. Taking a master creative and adapting it to sixty different ad unit sizes, regional language variants, seasonal variations, and channel-specific formats was previously either expensive or simply not done. Most brands ran the same core creative across contexts with minimal adaptation. AI-assisted adaptation makes the right version for each context tractable — and early performance data from companies doing this seriously suggests the performance lift from proper contextual adaptation is real.
Strategy and brief is the stage I'd characterize as being augmented rather than restructured. The best creative strategy still requires human understanding of the brand's positioning, the competitive context, and the audience psychology. AI can accelerate research, surface patterns in what's worked historically, and help explore option space — but the judgment about what a brand should stand for in a given market moment is not being automated in any meaningful way.
The quality control problem at scale
The challenge that creative supply chain disruption creates — and that most of the coverage doesn't address — is quality control at scale. When you go from ten creative variations to five hundred, the review process that worked for ten doesn't scale to five hundred. Human review of every variation is no longer tractable. But completely automated quality control is not reliable enough to trust without human oversight for anything with brand risk.
The companies building serious infrastructure for AI creative are the ones that have thought hard about this bottleneck. They're building structured quality gates: automated brand safety checks that flag obvious violations before human review; confidence scoring that routes low-confidence outputs to immediate human attention; tiered review workflows where high-reach or high-risk placements always get human eyes, while lower-stakes variations get lighter review; and audit trails that allow post-deployment quality analysis to feed back into generation parameters.
This quality control infrastructure is not glamorous. It's not what gets covered in product announcements. But it's the reason some companies are able to actually deploy AI creative at enterprise scale while others remain in pilot purgatory indefinitely. We look specifically for this infrastructure layer when evaluating companies in the creative automation space — its presence is a strong signal that the founding team has dealt with production realities, not just demo performance.
The rights and provenance question is not settled
One area where I want to be direct about uncertainty: the intellectual property and provenance questions around AI-generated creative are genuinely unsettled. The legal frameworks in Europe, the United States, and most other jurisdictions have not fully resolved how AI-generated content is owned, what training data usage is permissible, and what disclosure obligations apply when AI-generated creative appears in regulated advertising contexts.
This uncertainty creates real risk for enterprise buyers and for the companies selling to them. A European financial services firm running AI-generated creative in advertising is potentially navigating Financial Conduct Authority guidance, local advertising standards, and GDPR simultaneously — with no clear regulatory precedent on several key questions. The companies we're most comfortable backing in this space are the ones that are building with the assumption that provenance transparency and content traceability will eventually be required — not as a compliance burden but as a product feature — and have designed their systems accordingly.
We're not saying the rights environment will be resolved favorably or unfavorably for AI creative companies — we don't know, and nobody does. The point is that the companies treating this as a solved problem are taking on a risk they're not pricing. The ones treating it as an unsolved problem and building for auditability are in a better position regardless of how the regulatory environment develops.
What the next three years look like
The near-term trajectory I'd describe for the AI creative supply chain has three phases. In the first phase — roughly where we are now — the primary value capture is in production speed and volume. Brands discover that they can generate more variations faster and cheaper than before, and the most sophisticated among them start measuring the performance lift from better creative-to-context matching.
The second phase is the measurement infrastructure build-out. The bottleneck shifts from "can we produce enough variations" to "can we close the loop between creative variables and performance signals." This requires investment in structured creative taxonomies, performance attribution at the creative element level, and feedback mechanisms that inform the next generation cycle. A small number of companies are already in this phase. Most are not.
The third phase is the one I find most interesting from an investment standpoint: brands with two to three years of closed-loop creative performance data begin to have a compounding creative intelligence advantage over brands without it. The data asset — knowing precisely how specific creative variables interact with specific audience segments in specific channels — becomes the real differentiation. The companies that built the measurement infrastructure in phase two are the ones who accumulate the data asset in phase three.
This is the long arc of what's happening in AI creative. The supply chain disruption is real and happening now. The measurement infrastructure is being built. The data flywheel advantage is where the durable value will live — and it's where we're directing most of our Fund II attention in this category.