You have watched the same ad play three times in one break. Everyone has. Every repeat is an advertiser paying to irritate someone who already got the message.

Here is the part almost no one catches. Most DSPs report frequency as a single average across a week or a month. On the dashboard it looks healthy, say 8 times a month, so a large advertiser sees nothing wrong and moves on. The average hides the timing. The household showing 2.4 times a week may have taken two of those back-to-back in one pod while thousands barely saw the ad at all. The more you spend, the smoother that aggregate looks, which is why the biggest advertisers are the ones most likely to miss it.

We catch it because we are not reading someone else’s summary. GammaBurst, our own analytics and bidding platform, reads delivery at the household level, not as a rolled-up average. It shows how often a household was served, how close the impressions landed, and whether they stacked inside one break. The clustering that disappears in an average is exactly what surfaces in the real delivery.

A national direct-to-consumer retailer showed us how that plays out. One campaign reached two very different buyers, parents shopping for birthdays and holidays, and B2B buyers stocking events and schools, both running on the same frequency curve. The data made the gap obvious. Parents converted most efficiently at about one streaming impression a week, while B2B buyers needed six to eight, so no single cap could serve both without wasting spend somewhere.

We reset three things. We capped frequency separately for each audience, split CTV and display by role instead of running them in parallel, and reconciled reporting every week against what actually rang up at the register.

Seeing the problem is only half of it. A cap alone still lets impressions bunch, because caps are reactive. By the time three ads have run in one break, it is too late for most DSPs to step in. GammaBurst governs delivery at the bid, spacing impressions across breaks and days instead of letting them pile into one night. The cap sets the ceiling. The bidder controls the rhythm.

Within 30 days, streaming RoAS rose 48%, measured against the retailer’s own sales data rather than the platform’s self-report, from roughly $350K working harder over 16 weeks, not from spending more.

That is what a managed campaign looks like. The right frequency is never fixed. It shifts by audience, by region, and over the life of a campaign, so we keep adjusting in-flight instead of setting it once and walking away. Come see how we find the money others leave on the table, at DMWF North America, September 9 and 10 at the Javits Center in New York.