Dynamic audio ads: slot design, matrix build and clearance

Dynamic audio earns its production cost when the variable is something a listener can act on within the hour: the weather where they are standing, the branch three miles up the road they are driving, the price of the thing sitting in their basket. Twelve to twenty-four variants carries almost all of the completion difference we have been able to measure against a static control. Matrices in the hundreds do not clear, do not get checked properly, and do not repay the studio time. Signals we will use: Coarse location from the ad request. Local time of day. Device class, meaning in-dash, smart speaker or headphones, which changes what a call to action can reasonably ask for. Weather from a live feed. Sequence position, so the second spot can assume the first one happened. Your own segments matched in on a hashed identifier. Live price or stock feeds where you can stand behind the endpoint's uptime and where we have agreed the fallback line for the night it fails at three in the morning. Signals we refuse: Anything bought from a data broker that implies a listener's health, finances or household situation. Anything that addresses the listener as though the ad knows them, because the effect in audio is not persuasion, it is the sensation of being overheard by a machine, and it costs you the listener you just paid to reach. Naming an abandoned basket item in a spot a passenger can also hear. Recording the matrix: All variable stems in one session, one microphone, one distance, one room. Levels matched, room tone captured and laid under every seam, splices placed on consonant onsets with short crossfades so the join sits under the ear's threshold. The standard failure in this work is a stem recorded three weeks later on a different microphone, and the assembled result reads like a ransom note. When a new variant is needed after the fact we re-book the same voice and the same room, or we do not ship it. Clearance is the real ceiling: In the UK every version has to clear Radiocentre. So the campaign is written as one script with named slots and an agreed vocabulary for each slot, and the structure is submitted rather than a pile of unrelated reads. That is what makes a matrix affordable to clear at all. Build the variants first and you will pay to clear them one at a time, which is where most dynamic audio budgets quietly go. Proving the variants earned it: A holdout arm running the single generic spot against the dynamic build, on the same supply, under the same cap. Without it you are measuring audio rather than dynamic audio, and the report will flatter us both. We read variant-level completion, then the delta between dynamic and generic on the conversion endpoint agreed before launch, and we retire slots that show nothing. On most builds a third of the slots are gone by week six and the work is better for it. Not for: Brands with no reliable feed and no store estate. Dynamic assembly across three near-identical variants is a production line with nothing moving along it. Advertisers running one national message with no local, temporal or personal hook, where a well-made static spot beats a badly motivated dynamic one every time we have tested it. Anyone whose media weight sits below the point where a listener hears the same slot twice, because sequencing and variant learning both need frequency before they mean anything. We will tell you which of these applies to you in the first conversation rather than the fourth invoice.