Did the video, the podcast or the article drive the sale?
Use three measurements together, because each one sees a different part of the path. A multi-touch attribution model splits credit across the touchpoints you can tie to one buyer, which mostly means clicks and site visits. A marketing mix model estimates each channel's contribution from spend and sales over time, so it covers podcasts, CTV and other formats that rarely produce a click. A lift test, where a region or a random group sees no ads from one channel, checks whether that channel caused sales at all. Multimodal attribution adds one input to all three: what was actually in each ad and in the content around it, read from the video, audio and image files, so the answer names the creative and the context that worked as well as the channel.
Why do video, audio and text get credited unevenly?
Which attribution models are there, and which ones can I still use?
| Model | How credit is split | What it under-credits |
|---|---|---|
| Last click | All of it to the final touchpoint | Everything earlier, where video and audio usually sit |
| First click | All of it to the first touchpoint | Everything that closed the sale |
| Linear | Equal shares to every touchpoint | Differences in weight: a 6-second impression counts the same as a 20-minute podcast segment |
| Time decay | More to touchpoints closer to the sale | Early awareness formats |
| Position-based (U-shaped) | 40% to the first, 40% to the last, 20% shared by the middle | Mid-funnel content such as articles and reviews |
| Data-driven | Learned by comparing converting and non-converting paths | Touchpoints it never sees, such as audio and offline |
Google removed first-click, linear, time-decay and position-based attribution from Google Ads and Google Analytics 4 in 2023, citing low use, which leaves data-driven and last click in those tools (Google's announcement). The rules-based models are still easy to compute yourself from a table of paths, and other analytics tools still offer them.
How do I measure formats that never get a click?
Where does knowing what was in the ad come in?
Attribution tells you that a channel or an ad ID worked. To learn why, and to use it in the next campaign, each ad and each placement needs fields that describe it: the products and logos on screen, the words spoken, the text shown on screen, the type of scene, how soon the product appears, and the IAB Content Taxonomy 3.1 category of the content the ad ran next to. Join those fields to your conversions on creative ID and placement ID, and the same conversion data answers questions such as "do CTV ads that show the product in the first three seconds convert better?" or "which content categories do our podcast ads perform best in?".
Tagging a few dozen ads by hand works. Tagging thousands of creatives and every placement consistently is where teams stall, and it is where reading the files with models pays off, since every file gets the same fields from the same models.
| Step | What you do | Where it comes from |
|---|---|---|
| 1. Export conversions | Conversions or credited value per creative ID, placement ID, channel and day | Ad server, analytics tool, measurement partner or clean room |
| 2. Describe the creative | Scenes, objects and logos, transcript, on-screen text, time to first product shot | The video, audio and image files |
| 3. Describe the context | Content category and topic of each placement | The page, episode or program the ad ran in |
| 4. Join and compare | Add the fields as columns to the data-driven model or the mix model, or compare results across groups | Your warehouse or notebook |
| 5. Confirm | Test the finding in an A/B or lift test before moving budget | Your experimentation setup |
What should I watch out for?
How does Mixpeek fit?
Mixpeek reads your files where they already live, in S3, GCS or any bucket, and turns what's in them into data your software can search, classify and moderate. In an attribution setup it covers the creative and context side: scenes, speech transcripts, on-screen text, objects and logos, and IAB Content Taxonomy 3.1 categories for each video, image and audio file, stored with that file's ID so they join to your conversion export. Attribution credit, user tracking and identity stay with your analytics tool, measurement partner or clean room. On the pricing page, video is $0.05 a minute for scenes, speech transcript and descriptions, on-screen text adds $0.10 a minute, and taxonomy classification is included.
Related: how to fill content categories in a CTV bid request, the IAB Content Taxonomy 3.1 classification benchmark and the best ad tech AI platforms.
Frequently Asked Questions
What is multimodal attribution?
It is measuring how ads and content in different formats (video, audio, images and text) contribute to conversions, using what is in each piece as well as where and when it ran. In practice it is an attribution model or a mix model with extra columns describing each creative and placement.
Which attribution model works for video and podcast ads?
Click-based models on their own undercount both. Use a marketing mix model or lift tests to size the channels that rarely produce a click, and a data-driven model for the paths you can track, then check one against the other.
Can Google Analytics 4 attribute podcast or CTV ads?
Only through visits it can see. GA4 offers data-driven and last-click attribution across the touchpoints it tracks, so a podcast or CTV ad with no later tracked visit from that channel is missing from the path. Promo codes, vanity URLs, a mix model and lift tests fill the gap.
How do I find which creative elements drive conversions?
Describe every creative with the same fields (products on screen, spoken lines, on-screen text, time to the first product shot), join them to conversions on creative ID, compare groups, and confirm the strongest difference with an A/B or lift test.