In October 2018 we published How Dynamic Ad Insertion Works. We had just cut the cord, and we’d noticed that streaming ads felt more like the web than like television: sometimes targeted, sometimes retargeted, sometimes the same spot over and over. So we dug into the plumbing.
The short version of that piece: streaming video is delivered in small fragments, described by a manifest. Dynamic ad insertion works by having a decision service read the manifest’s markers for ad breaks and hand each viewer a modified manifest that points to the ads chosen for them. The same approach can swap regional programming, handle blackouts and serve linear, on-demand and streaming from one architecture.
That description still holds. What changed is everything around it. Streaming stopped being an add-on to television and became television, and a lot of problems that were theoretical in 2018 became very practical.
How much attention it got
Stories in our archive that mention connected TV, by year:
| Year | Stories |
|---|---|
| 2017 | 22 |
| 2018 | 42 |
| 2019 | 57 |
| 2020 | 104 |
| 2021 | 144 |
| 2022 | 128 |
| 2023 | 76 |
| 2024 | 73 |
| 2025 | 63 |
Coverage peaked in 2021, which is about when CTV stopped being a novelty and became a line item. The drop since then isn’t a decline in importance. It’s the same pattern we’ve seen with every technology that wins: once it’s how things work, it stops being news.
What held up
Server-side insertion became the default. In 2018 we described the decision service stitching ads into the stream. That approach, server-side ad insertion or SSAI, is now the standard way premium streaming handles ads, because it plays smoothly and is much harder to block. AdExchanger published a full explainer on SSAI in 2022, which is a good sign it had become something buyers needed to understand, not just engineers.
One architecture for linear and streaming. The 2018 piece ended on the idea that linear, on-demand and streaming would converge on a common, software-based delivery stack. That’s largely what happened, and the industry worked on standardizing it. In 2019 AdExchanger argued it was time for standardization in linear addressable TV, and in early 2020 a TV-maker-backed group, Project OAR, launched specs for linear dynamic ad insertion.
The biggest platforms put ads in the stream. In 2018, the idea of ads on the major subscription services was still a prediction. By 2023 AdExchanger was running six things you probably didn’t know about Netflix’s ad business, and in January 2024 Prime Video went ad-supported by default. Live sports moved too. Amazon’s streaming of Thursday Night Football, which TechCrunch reported in 2018 as another two years of streaming, became a full exclusive package and, by 2022, a showcase for what streaming ads could drive in measurable response.
What broke
The plumbing scaled. The coordination around it didn’t, at least not right away.
Frequency. In 2018 we mentioned seeing the same ad over and over. That turned out to be one of streaming’s most persistent problems. In 2022 AdExchanger talked to a company working on CTV’s chronic frequency problem, and in 2023 it reported that Pluto TV’s ad insertion tech had a frequency problem.
The cause is structural, and it follows directly from how insertion works. Each ad break is often filled through a separate request, sometimes by different systems, and no single system sees everything a household has been shown. The decision service knows what it chose for this break. It doesn’t necessarily know what three other platforms chose in the last hour.
Ad pods. Television sells ad breaks as pods, several spots in a row. Early programmatic systems treated each slot as a separate auction, which made it hard to avoid two competing brands in the same break or the same brand twice. The industry answered with standards. Ad Tech Explained walked through OpenRTB 2.6 and pod bidding in 2022, the update that lets a whole break be bid on as a unit. And in 2025, the IAB Tech Lab introduced a playbook for live event streaming ads, because live sports puts millions of simultaneous ad decisions on the same few seconds.
Measurement. Our 2018 piece was optimistic that connected devices would return per-device data for tracking and attribution. Technically, they do. The problem is that it’s spread across devices, apps, platforms and operating systems that don’t share it. As early as 2017, agencies were experimenting with CTV measurement. By 2020 an AdExchanger contributor wrote that CTV measurement was doomed unless the big platforms collaborated. Measurement companies spent the next few years trying to align TV and CTV into something a buyer could compare. In late 2025, the free, ad-supported streaming channels that grew up alongside all of this were still described as underserved for measurement and targeting.
What we got wrong
In 2018 we were excited about ATSC 3.0, the next-generation broadcast standard, and its promise of an internet back channel for over-the-air viewing data. Digiday explained it that December, and broadcasters welcomed it as game-changing the next fall.
It’s still coming. In November 2025, MediaPost’s headline on it was why TV’s next big leap is falling flat. The lesson we take from that: the parts of the stack that rode on infrastructure people were already buying for other reasons (broadband, smart TVs, streaming apps) moved fast. The parts that needed a new antenna, a new tuner and a regulatory transition moved slowly.
What we’d add to the explainer now
If we were writing the 2018 piece from scratch today, the architecture diagram would look much the same. The text would spend a lot more time on three things.
The decision is only as good as what it can see. Dynamic ad insertion can put any ad in any break. Choosing well depends on knowing what this household has already seen, which competitive rules apply to this break and what the advertiser has agreed to pay for. Most of the industry’s CTV problems since 2018 are failures of that shared knowledge, not of the insertion itself.
Live is a different problem from on-demand. On-demand insertion can make its decisions with a little time to spare. Live sports asks the same systems to make millions of decisions in the same instant, at the most expensive moment of the year. That’s why live streaming has needed its own standards.
Data is now the differentiator. In 2018, being able to target a stream at all was the breakthrough. Now every major platform can do it. What separates them is whose audience data drives the decision: the platform’s own logged-in viewers, a retailer’s purchase data, or the advertiser’s first-party customer list. The insertion is plumbing. The data behind the decision is the product.
The plumbing we described in 2018 turned out to be durable. The interesting work now is everything that tells it what to do.
Every headline in this piece links to the original story. Read our original 2018 explainer, How Dynamic Ad Insertion Works, or browse the archive’s coverage of connected TV, server-side ad insertion and CTV measurement.
Kevin Reynolds is the founder of Advertising Perspectives, MadTech Consulting and data4orce.ai. He is a marketing technology architect and C-level advisor, with a focus on Salesforce and data-driven advertising.