Zee 5 Bets on Federated Audience Intelligence for Its Ad Platform
ZAP rests on an audience intelligence model where the query travels and the data stays. The more interesting part is what the partner refuses to do.
Zee 5 has launched an advertiser platform called ZAP, built with Saptharushi, a Singapore-headquartered data collaboration company that trades as The Audience Company. The pitch is that advertisers can build audience cohorts from what people watch on Zee 5 combined with consented behavioural and transactional signals from partners in quick commerce and digital payments, without any of that partner data leaving the environment it sits in. The company’s summary is the clearest thing in the release: the data does not move, the query travels. Zee 5 reaches viewers in more than 190 countries across seven languages and carries premium originals, a free tier, live sport including the FIFA World Cup, and a children’s slate, which gives it unusually varied viewing signals to work with.
That variety is the commercial premise: a sport viewer during a tournament is worth more to a beverage brand than a library viewer at midnight, and conventional demographic targeting cannot tell them apart. The gap between what a platform knows about viewing and what it can sell to an advertiser is the problem audience intelligence exists to close. Zee 5’s advertising business has been building toward it, because a platform holding several kinds of audience in one place needs a mechanism that lets a buyer pay differently for each.
What the architecture actually does
Saptharushi’s system, which it calls Atom, installs a component inside each participant’s own cloud. A query goes out to that component and only counts and hashed identifiers come back. Matching runs deterministically on those hashed identifiers rather than probabilistically, which matters because a probabilistic match on an audience is a guess that compounds across a large buy.
The stack combines three techniques. Differential privacy adds mathematical noise to outputs so nobody can reconstruct an individual from them. Federated querying and learning compute in place rather than pooling data centrally. A permissioned distributed ledger records consent and every match. Saptharushi argues the combination matters more than the parts, because federated learning alone still asks participants to trust whoever coordinates the system, and the ledger replaces that trust with a record every participant can check. The architecture holds granted patents in Singapore and the United States, with the founder named as sole inventor.
Saptharushi names the parts of its protocol after particles. Neutron resolves fragmented signals into an addressable audience. Proton curates and activates that audience where it already lives, across open web, connected television, owned channels and customer relationship management systems, without pooling or copying. Electron handles rewards, valuing attention and renewing the consent that keeps an audience usable. The sequence is the argument: resolve, reach, reward, with consent renewed at the end of each cycle rather than treated as a one-time permission.
Why federated beats pooling
Conventional data clean rooms pool data in a neutral environment and compute on it there. That works, and most large advertisers already use the model, but it requires every participant to hand over a copy. For a publisher sitting on first-party viewing data, or a payments company holding transaction records, handing over a copy is the decision that triggers legal review, board sign-off and, under India’s Digital Personal Data Protection framework, a clear account of who carries fiduciary duty.
A federated model removes that decision. Nothing leaves its owner, so nothing needs a new custody arrangement. That is the commercial argument, and it explains why a partner can sign publishers who would not have pooled. There is a second argument the release does not make. A pooled clean room creates a new dataset that somebody has to govern, secure and eventually delete, and each obligation carries cost and legal exposure. A federated model creates no such artefact. The governance burden shifts from protecting a copy to auditing queries, which is a smaller problem, and that is the claim every audience intelligence vendor now makes and few can evidence.
The trade-off is flexibility. A federated query cannot do everything a pooled dataset can, because only counts and hashed identifiers return, and an analyst who wants to explore data rather than ask a specific question finds the door closed. Saptharushi concedes as much by describing the clean room as one of several technologies rather than as the product.
The three gates
The most interesting thing about Saptharushi is not its technology but what it has publicly ruled out. Its own materials list three commitments: it will never centralise or host a client’s data, it will never act as a data fiduciary for someone else’s users, and it will never run ad operations, bidding or ad serving.
Each refusal costs revenue and each makes the neutrality claim credible. A company that hosts data becomes an arbitrageur of it. A company that acts as fiduciary for a publisher’s users carries a conflict when it also sells access to them. Moreover, a company that runs the auction has an interest in the outcome of the audience it just built. The position is unusual in advertising technology and it is what makes the architecture believable, because a federated system run by someone with something to gain from pooling is a federated system waiting to pool.
This also settles a question the release leaves open. Under the DPDP framework somebody holds fiduciary duty for the viewing data. Saptharushi says it will not be that party. That leaves Zee 5 holding the duty, which is the correct allocation for a publisher and also the arrangement that makes the partnership possible at all.
The industry-first claim
The release calls ZAP an industry-first offering, and that claim needs narrowing. Saptharushi launched the same underlying marketplace for the Digital News Publishers Association in May 2026 under the name ONAM, announced it publicly again in October, and runs it live on PubMatic with founding news publishers. ZAP is the first deployment on a streaming platform rather than the first of its kind.
None of which diminishes it. Applying a federated audience model to a video platform with seven-language content, live sport and a free tier is a different problem from applying it to news, because the signals are richer and the cohorts carry more commercial weight. But a reader should know that the infrastructure has been in market since May, that the audience intelligence layer is not new, and that ZAP is its third public instantiation rather than its first.
What is missing
The release gives no number for the one thing audience intelligence most needs to prove. Prashant Shetty of ZEEL states that Saptharushi’s previous deployments showed a significant, multifold uplift in click-through rate over campaigns activated against a brand’s existing customer relationship management audience. Multifold is not a figure. No client appears by name, no baseline accompanies it and no measurement window applies, so nobody considering the platform can check or repeat the claim.
The same applies to scale. No advertiser appears by name, the release gives no cohort sizes and it describes no pricing model. Saptharushi‘s own site reports five enterprise clients live, seven data partners in production, a twenty-two publisher collective and more than five billion profiles processed, alongside ISO 27001 certification with no nonconformities, SOC 2 Type II and a CERT-In penetration test. None of that appears in the Zee 5 release, which is an odd omission for an announcement whose credibility rests on the partner’s track record rather than on the publisher’s reach. One further omission deserves mention. The release describes co-branded attention and loyalty rewards as something the partnership will explore, particularly around premium content and sport. Exploration is not a product. No design, no partner and no reward mechanic appears, which suggests the rewards component is a later phase rather than part of what is launching now.
What to watch
Three markers. The first is whether an advertiser will go on record with a measured uplift, since the whole proposition rests on federated audience intelligence being better rather than merely more compliant. The second is how DPDP enforcement develops, because the architecture anticipates a law still taking shape and the fiduciary allocation is the part most likely to be tested. The third is whether Zee 5 opens the platform to other publishers, which would turn ZAP from a broadcaster’s tool into the audience network Saptharushi has been describing since its Aqilliz years.
What it means
The strategic read is that the privacy argument in advertising technology has moved from compliance to architecture. For a decade the industry treated privacy as a policy layer on top of collection; the current generation tries to make it a property of the design, so the compliant path is also the cheap one. The compliance cost of a clean room recurs with every dataset; the compliance cost of a federated query is an audit trail. Whether that holds depends on whether federated audiences outperform pooled ones commercially. Saptharushi is betting they will, and Zee 5 has now put a large streaming inventory base behind the bet.
For advertisers the practical question is whether a cohort built this way converts better than one bought on demographics. For publishers it is whether their viewing data becomes an asset they can price rather than a byproduct they hand to a platform. Both questions share an answer, and no announcement will settle either.