Why does radio look weaker in media mix models than in other measures? The IAB’s answer in a new guide is missing data, not missing results, and it puts the fix in the hands of advertisers, who pay for the models, rather than stations or agencies.
The IAB’s Amplifying Audio’s Impact in MMM, a follow-up to its June 2025 beginner’s guide, says digital audio doesn’t have an effectiveness problem but a data problem. It cites Nielsen Compass benchmarks that place digital audio in the upper half of channel returns and rank broadcast radio fourth in the US.
The guide says the data reaching a media mix model about audio is usually too thin for the model to measure audio with confidence, and a model unsure of a number returns a small one. It says planned delivery data barely varies, because a plan is an even weekly schedule, and that replacing it with a record of ads that actually aired is the key to visibility.
A 2012 study for the radio industry by Sequent Partners for Arbitron did that in three real-world remodels and raised measured radio return by 27 points to 78%, with model confidence rising from 92% to 97%. The guide notes the study covered local spot only. A commercial product launched around 2018 covered 45 markets and excluded network radio, and the IAB says one holding company submitted as-run data on network radio models for about two years without a material improvement.
The IAB cites four reasons: national ratings estimates are published twice a year, so a book can reflect listening from four months earlier; network schedules show little gap between bought and delivered weight; the agency added no fields; and the product never covered network radio. For digital audio, an ad server logs each impression, so audience lag and coverage gaps disappear, and what remains is detail lost when files pass between parties.
The guide traces eight hand-offs where detail is lost, from publisher feeds to agency verification tools to model intake. It says audio often enters a model as one line instead of broadcast, streaming, podcasting, and satellite, and that added value such as bonus weight can be counted at full value. An agency-supplied analysis of one five-week campaign found cost per acquisition varied roughly threefold across the four platforms, and that aggregating to one audio total reversed the finding.
The IAB says only the advertiser can change the model specification, since the advertiser commissions and pays for it. Its required delivery-feed fields include date, publisher, brand, campaign, audio type, format, line item, device, creative, impressions, cost type and rate, spend, and DMA-level geography. Recommended fields include daypart, creative length, ad position, and three new ones: audience basis, delivery record source, and whether weight was paid for or added value. SiriusXM Media’s first-draft data dictionary has since become a live feed of 21 fields.
The guide also says fewer than about 50 million national impressions a year is hard to read, and local campaigns should run in at least three to five markets. It recommends advertisers model audio as four channels and share results with publishers; agencies record the split when the buy is booked; publishers supply the required fields to every buyer; and measurement vendors publish their fields and minimum thresholds.
Contributors include Audacy, Cumulus Media/Westwood One, Nielsen, and Triton Digital, among others.



