Relevance has two inputs. The first is the model: the engine that reads a customer and picks the offer most likely to land. That half gets all the attention. Every quarter the engines get smarter, the press covers the models, the roadmaps fund the next version. The second input is the network the model has to choose from. That half gets almost none of the attention, and it does just as much of the work.
A model can't recommend an offer that doesn't exist. Point the best engine in the industry at a thin, stale network and it converges on the same handful of offers, week after week.
For a commerce media partner, the network is what to evaluate: how deep it is, where the quality bar sits, how fast new advertisers cycle in. All three are downstream of direct advertiser relationships. Affiliate sourcing can't deliver them.
A deep network only helps if the relationships are real
Most networks don't have their own advertisers. They pull from affiliate networks and run the same offers everyone else has. The economics are easy. The result is sameness.
Because every network drawing from that pool surfaces the same offers, they repeat everywhere: the same VPN trial, the same meal-kit deal, cycling across one merchant's checkout after another until the customer stops looking. Performance climbs for a quarter, then flattens. The repetition complaints come the quarter after. By the time conversion data confirms the drop, the offer is already wallpaper.
Rokt works the other way. We don't resell other people's offers. We work directly with the advertisers, as a partner and an advisor rather than a vendor selling impressions, helping them learn what works and improving it alongside them over time. The relationship goes well beyond the transaction.
That puts us in a specific spot: a trusted intermediary between the merchant whose checkout the offer runs in and the advertiser whose brand it carries. Both sides share data with us that neither could combine on its own.
An affiliate setup only sees what happens inside the placement: the click, the conversion if it lands in that session, nothing after. Because we work directly with the advertiser, we see what comes next. Whether the customer was still around at day 30. What they spent on the second purchase. What they were worth at month six. The model optimizes toward what the customer is worth over time, not what the click is worth for a day.
That changes which offer wins. A fitness brand cares about trial-to-paid conversion and retention. A clothing brand cares about basket size and repeat purchase. Same customer, same impression, completely different bid, because we can see what each advertiser got downstream and value the impression accordingly.
That visibility only exists on the back of a direct relationship. It's what Juliana Blazuk at Afterpay (Block) meant when she said Rokt's offers "feel native to the Afterpay experience." That fit comes from an advertiser and a placement aimed at the same outcome.
A good network keeps letting new advertisers in
The usual knock on large networks is that they get stuck. The offers with the longest track record win every auction, new advertisers never get tested, and depth stops turning into variety for the customer. That's a real risk. But it comes from a system that rewards history and nothing else, not from depth itself.
The fix is to judge a new offer by what it is, not just by its track record: what category it's in, who it's for, what it costs, what kind of offer it is. An advertiser the network has never run still has a lot in common with ones it knows well, so the model can predict how it will do while it earns its own history.
A network that does this keeps pushing fresh advertisers into live placements. Over the last year, Rokt signed roughly 104 advertiser deals a quarter on average, nearly half of them new advertisers or new product launches. That steady inflow is why a partner sees rotation instead of the same fixed set going stale.
"Premium" only means something if the bar holds
"Premium" is the word every network stretches furthest, so what counts is what it means in practice. Every advertiser clears a bar before it competes for a single impression: a legitimate brand, an offer that holds up at the customer level, a commitment built to last, and genuine relevance to that partner's customers.
That last one ties back to everything above. A strong brand in the wrong context is still the wrong offer.
The curation doesn't stop at onboarding. Advertisers that slip on quality fall out of rotation regardless of what they bid, and partners layer their own category filters, exclusions, and brand controls on top.
The bar holds for one reason: demand pulls supply. Premium advertisers compete for access to Rokt's partners and the billions of high-intent transactions a year that flow through them, across retail, payments, travel, and finance. An advertiser that won't meet the bar gets dropped, because one that clears it is already waiting.
That's also what produces the rotation a partner actually wants. Different customers see different offers, and the same customer doesn't see in month six what they saw in week one. The placement keeps working instead of fading into the background.
The model is only half the story
The Rokt Brain will process signals from more than 10 billion transactions in 2026, every one a Transaction Moment where intent is fresh and the next decision carries the most weight. The scale is real and the intelligence is real. It's also half the story.
The other half is the network the Brain has to choose from, and whether it's deep, genuinely vetted, and constantly refreshed. Without that, no amount of model investment lifts the ceiling.
Most of the industry's relevance conversation is about building a better model. The less glamorous work is building the network you point it at, and earning the trust, on both sides, that makes that network possible.
Models get the headlines. Networks do the lifting.
.png)
%20(1).png)
