By Ryan Lee, Trading Director,Nano Interactive
Picture an ad for a savings account running in a documentary about a food bank. The keyword “bank” can get you into both conversations, but it doesn’t tell you whether the context fits your campaign. Go the other way with broad exclusions and you hit a different problem; for instance, a blocklist containing “shoot” can rule out fashion shows and football highlights along with crime coverage.
In our last CTV spotlight, we looked at why the programme is now as important as the person watching it. But working out what a programme is about is harder when each streaming service describes its content differently.
Vectorisation can help media planners match content by its meaning instead of relying on a keyword, and in this article, I’ll explain how it works, how to describe the content you want to reach, and what to check before you buy.
Put meaning on the map with vectorisation
In simple terms, vectorisation converts text and the surrounding context into a long string of numbers to represent the meaning. These numbers, known as “vector embeddings”, let a planning tool compare different pieces of content based on context.
Think of those numbers as points on a map, with similar meanings close together. The context helps place a financial bank near savings, while a food bank sits closer to charitable support, even though both share the word “bank”.
The planning tool turns a description of the content you want into a vector, places it on the same map and looks for content nearby, so you can find relevant matches even when they use different words. With CTV, the same principle can apply to a programme’s title, synopsis or genre description, even when two services label the same show in different ways. And because content is analysed in advance, matching is fast enough for live ad buying.
Vectors also give planners an alternative to familiar approaches like lookalikes. A lookalike model stretches a seed list of known users, whereas vectors built from content don’t need personal identifiers at all. That’s useful on CTV, where one household can appear as several audiences. That said, some suppliers build vectors from identity signals too, so it’s worth asking which data powers your match.
Describe the conversations you want to reach
With vectors, your brief can describe what relevant content actually looks like. Keywords can still be a starting point, but a fuller description gives the system more context to match against. The trick is to write your intent prompt as a sentence that could appear in the content you want to reach, rather than a description of the person watching it.
For a premium beach holiday campaign, that means describing the holiday rather than the traveller’s income. For example, “a quiet coastal resort offers spacious rooms overlooking the Mediterranean, where guests can unwind on a private sandy beach beside clear blue water.”
To broaden reach, add other relevant conversations, each with its own prompt. A skiing trip or a city break will bring in different content from a beach holiday. Rephrasing the same idea five ways won’t, because similar prompts are grouped together and matched as one.
Check the fit before you foot the bill
The next step is to check how closely the content matches your topic. Each page or programme gets a “similarity score”, and you set your own threshold to control how close a match needs to be. A narrower setting keeps you to closely-related content, which can help with brand fit, while a broader one widens reach for an awareness campaign.
Then preview the matches to check you’ve got the right balance between relevance and scale. It’s worth asking for examples alongside their scores, so you can see what a good match looks like. Remember that a similarity score tells you how well the content matches, rather than how well the campaign performs, so measure results against the goal in your brief.
Finally, ask how often content is reanalysed and, for CTV, what programme information is available for matching.
A keyword list will only ever see “bank” as four letters. Vectorisation looks at the company a word keeps, so your savings account ad finds conversations about mortgages and money, rather than a documentary about a food bank.
Describe the content you want in your brief, check the matches before you buy, and measure results against your goal. That way, whether on a webpage or the big screen, you’re buying the conversation, rather than just the word.
