In signage and interactive video, AI means software that chooses what to show based on live data rather than a schedule someone set in advance. The recommendation engine, the audience estimate, the voiceover generated per viewer are all the same idea: the content is assembled at play time, not authored ahead of it.
That matters because the alternative scales badly. A network of two hundred screens across four countries cannot have a human deciding each playlist, and a personalised video for fifty thousand customers cannot be edited fifty thousand times. Systems that generate or select per context are how those numbers stop being absurd.
The caveat is that most of what is sold as AI in this industry is a rules engine with a better name. "If the weather API says rain, play the umbrella creative" is a conditional, not a model, and it works well, but it should be priced and evaluated as a conditional. The questions that separate the two are whether the system improves from outcomes without someone rewriting the rules, and whether it can explain a decision after the fact.
The second caveat is regulatory. Anything inferring age, gender or mood from a camera is processing biometric data under GDPR and KVKK, regardless of whether an image is stored. Aggregate counting is a different legal object from recognition, and vendors routinely blur the line.