Engineering the Ad Format Behind Creator-Brand Partnerships

Open almost any social feed today and the line between a personal recommendation and a paid placement has nearly dissolved. Creators sit at the center of how brands reach audiences, and the spending has followed them there. The global influencer marketing market reached roughly 33 billion dollars in 2025, more than triple its level five years earlier. Behind that figure is a change in what advertising buys. Increasingly, brands pay for the trust a creator has built with an audience, a currency that raw impressions never captured. The systems that serve ads, though, were not originally built to represent that relationship at all.

Kiran Kumar Manku, a software engineer with more than 11 years of experience across distributed systems and large-scale advertising platforms, spent close to a year building one of the first ad formats designed to surface those partnerships natively. He has spent that career inside the layer that decides which ad appears, who is charged for it, and how it renders on screen, and he carries that perspective into the wider field as a judge at the Beta University AI Super Hackathon. The format he led created a category of advertising the platform did not previously have, one that places a creator and the brand they promote in the same unit. That work sat exactly where the market was moving and the existing tooling was not.

The Gap Between How Brands Market and What Ad Systems See

For years, the way brands marketed through creators ran ahead of the way platforms monetized it. A creator would feature a product, tag the brand, and shape a purchase decision, all on the organic side of a platform where that behavior felt native. The advertising side treated creator content and brand campaigns as separate worlds. An ad was an ad, served and measured on its own, with no structural notion that a creator and a brand might be promoting something together. The result was a format gap: the most natural form of modern brand marketing had no first-class home in the ad system.

Manku owned core engineering across the ranking, delivery, and rendering paths for the new partnership format. Each path carried a distinct problem. A partnership ad had to be scored against every other ad competing for the same slot, delivered to an audience for whom the creator-brand pairing was relevant, and rendered so that it read as a real partnership rather than a banner dropped into a feed. None of those paths had handled a unit that represented two parties at once. Building the format meant teaching each stage of the serving stack to treat the creator-brand relationship as the thing being served.

“The hard part was conceptual before it was technical,” says Kiran Kumar Manku. “The stack assumed one advertiser per ad. A partnership is two parties with a shared interest, and almost nothing in the ranking or delivery logic knew how to express that.”

Teaching the Serving Path to Carry a Relationship

As creator marketing matured, it stopped being an experiment that lived at the margin of a brand’s budget. By early 2025, around 14% of marketers expected to allocate 10% to 15% of their total budgets to influencer marketing, with a similarly sized group planning to commit even more. Money at that scale changes what a platform owes its advertisers. A channel that absorbs a tenth of a marketing budget cannot be served by infrastructure that treats it as a special case. It has to be ranked, priced, and delivered with the same rigor as any other ad.

For Manku, that rigor started in the ranking path. A new format begins with no track record, so the model deciding when to show a partnership ad cannot lean on years of accumulated signal the way an established format can. The delivery path had to route these ads to audiences where the creator-brand pairing carried weight, and the rendering path had to present the unit cleanly across a global advertiser base from the first day it launched. He designed the serving components so they could be reused, letting future interactive formats be added without rebuilding the validation and delivery logic from scratch.

“You are serving something the system has never seen, at full scale, on day one,” Manku explains. “There is no quiet beta where the format earns trust slowly. It either holds up for every advertiser at launch or it does not ship.”

A Training Pipeline With No Foundation to Reuse

Ranking any ad well depends on a model that has learned from how people respond to it. For a format that has never existed, that history is absent. There was no prior version of the ad to learn from, and no internal pipeline already tuned to produce the models such a format would need. A brand-new format therefore carries a hidden dependency. Before ranking can work at all, something has to generate the training data and the models it relies on, and that machinery has to exist before the format can prove itself.

Manku designed and built that model-training pipeline from the ground up, with no existing foundation inside the company to extend. It produced the models specific to the partnership format and became the basis other engineers later built on. Building it meant making decisions about data handling, model evaluation, and retraining cadence that would hold up once the format was live and changing under real traffic, the same evaluation discipline he brings as a peer reviewer for the IEEE International Conference on Advanced Technologies and Innovations.

“Building the pipeline was the part that had no shortcut,” Manku notes. “You cannot borrow another format’s models and hope they transfer. The format only worked because the thing underneath it, the training system, was built deliberately for it.”

The Coordination That Decided Whether It Shipped

The category is still expanding quickly. The influencer marketing market is projected to reach about 121.81 billion dollars by 2030, growing at a compound annual rate of 31.42%. Growth at that pace pulls in scrutiny. When an ad surfaces a paid relationship between a creator and a brand, it carries disclosure and data-handling obligations that an ordinary banner does not. The faster the category grows, the less tolerance there is for getting those obligations wrong.

The hardest part of the project was not in the code. Because the format made a commercial relationship visible, it had to satisfy five separate organizations at once, including the teams responsible for privacy, legal review, and ads integrity, each with its own constraints and its own veto. A design that resolved a privacy concern still had to answer an integrity concern, and the same architecture had to be defensible to legal in the language legal cared about. Manku drove that coordination, resolving the data-handling and policy requirements that gated the launch. The technical design was finite. The cross-organization agreement was the work that actually determined whether the format went live.

“A design that only survives one review will not ship,” Manku reflects. “It has to mean the same thing to privacy, to legal, and to integrity at the same time. Getting five teams to agree on one architecture is slower than writing the code, and it matters more.”

What the Format Set in Motion

What began as a single new ad format left something larger behind. The serving components Manku built were designed for reuse, so later interactive and partnership formats could be added as increments rather than ground-up builds. The training pipeline became standing infrastructure for the category. And the platform gained a monetization surface tied directly to the way brands had already started to advertise, through the creators their customers trust.

The pattern matters more than the single launch. Building for a marketing behavior before the tooling exists has become a recurring demand, as creative tools and audience matching increasingly run on AI. When models generate and place creator content, the layer that verifies who is partnered with whom, and that the relationship is represented honestly, grows harder and more consequential together. The work Manku points to next sits in that layer: holding attribution and correctness intact as more of the format is handed to automated systems.

“The format people saw was the easy part to describe,” Manku concludes. “What made it work was treating the creator-brand relationship as a real object in the system, from the model that ranked it to the teams that signed off on it. That is the part that lasts, because the next format will need the same thing.”