Marketing Attribution for Video Campaigns

Add up the conversions reported by four ad platforms in the same month and you'll get a number larger than your actual order count. Sometimes 1.5x larger. Nobody is lying — each platform counts a conversion it plausibly touched, and the same buyer touched three of them. This is the first thing to internalize about video attribution: platform-reported numbers are claims, not accounting.

The second thing is that video makes this worse than any other format. A static ad is usually seen or not seen. A video is watched for two seconds, or eight, or all the way through, sometimes without sound, sometimes on a device that never converts. View-through windows stretch the credit further than click windows ever did. So a channel that "drives" conversions in the dashboard may be sitting downstream of demand something else created.

None of that means you should give up on measurement. It means you need a hierarchy: what you can trust exactly, what you can trust directionally, and what you use only for creative decisions. This digital marketing guide lays out that hierarchy for video campaigns and shows where each measurement type breaks.

What each attribution model actually claims

Every model answers a different question. Choosing one is choosing which question you care about this quarter.

Model	The question it answers	Where it lies
Last click	Which touch closed?	Ignores everything that created demand
First click	What started the journey?	Overcredits top-funnel, ignores closing
Linear / time decay	How do we split credit fairly?	"Fair" is an assumption, not a measurement
Platform-reported (in-app)	What did this platform touch?	Double-counts across platforms; view-through inflation
Post-purchase survey	What does the buyer say?	Recall bias, but immune to cookie loss
Geo holdout / incrementality	What happens if we stop?	Slow, expensive, and the only causal answer
Media mix modeling	How do channels contribute in aggregate?	Needs volume and history; low granularity

The practical stance: use platform reporting to compare creatives inside one platform, use post-purchase surveys and holdouts to decide budget between platforms, and never mix the two. Comparing Meta's reported ROAS to TikTok's reported ROAS is comparing two different accounting systems.

Why video breaks click-based attribution

Three mechanics specific to video make click paths misleading.

View-through windows. Most platforms count a conversion if someone saw a video within a window — often 1 day, sometimes 7. For a brand with existing demand, a large share of "view-through conversions" would have happened anyway. If a channel's reported performance collapses when you shorten the view-through window, that channel was mostly claiming credit.

Sound-off viewing. A meaningful share of feed video is watched muted. If your measurement assumes message delivery and your creative puts the message in the voiceover, reported impressions overstate delivered impressions. This is a creative fix, not a measurement fix: burn the key line into captions.

Dark social and search lift. People see a product video, don't click, and search the brand name later. That converts as branded search or direct. Video looks weak; search looks brilliant. The tell is a correlation between video spend and branded search volume — if branded search rises within days of a video push, your video channel is underreported.

The measurement stack that actually works

You don't need a data warehouse to do this well. You need four layers, each answering a narrower question than the one above it.

1. Business truth. Orders, revenue, new customers, from your own system. One number, no platform involved. Everything else is explanation.
2. Directional channel split. Post-purchase survey ("how did you hear about us?") plus branded search trend. Cheap, resistant to tracking loss, good enough to catch a channel that's dying.
3. Causal checks, occasionally. Geo holdouts or scheduled pauses on one channel for two to four weeks. Run these two or three times a year, not monthly.
4. Creative comparison, continuously. Inside one platform, one audience, one budget: which video wins. This is where platform data is genuinely reliable, because the bias applies equally to both variants.

That fourth layer is where most of your day-to-day decisions live, and it's the one to invest in. We've covered the metric selection in brand video metrics that matter and the deeper cut in creative analytics: what to track.

Tagging discipline you'll actually maintain

Attribution systems fail on naming, not math. A UTM scheme that nobody follows produces a report nobody trusts.

Keep it to five fields and enforce them with a spreadsheet template rather than good intentions:

• source — the platform, lowercase, no variants (tiktok, never TikTok and tik-tok)
• medium — paid, organic, email, affiliate
• campaign — the offer or launch, not the date
• content — the creative ID, matching your file naming
• term — audience or placement

The critical one is content. If your creative IDs in the ad platform match your production file names, you can trace a winning ad back to the exact hook, model, and reference set that produced it. If they don't, you'll rediscover the same winning hook three times a year. When you generate variants systematically — same units, one changed axis — this mapping stays clean automatically. That's part of why structured generation with an AI video generator beats ad-hoc production for teams that measure anything.

Reading results without fooling yourself

A few habits that prevent expensive wrong turns:

• Wait for the conversion window before judging. Killing a video on day two, when your median time-to-purchase is nine days, is guaranteed to kill winners.
• Compare within cohorts. A video launched into a retargeting audience will always look better than one launched cold. That's audience, not creative.
• Check the denominator. Reported ROAS improving while revenue is flat means spend fell, not that creative improved.
• Separate creative testing from budget decisions. Creative tests need matched conditions; budget decisions need incrementality. Using one for the other is the most common analytics mistake in performance teams.
• Log every change. Attribution shifts caused by a tracking change look identical to shifts caused by creative. A dated change log resolves those arguments in minutes.

For structured creative comparison specifically, the setup matters more than the analysis — see A/B testing video creative properly for how to build tests that produce readable results.

What to do when the data is thin

Small brands often can't run a clean geo holdout: not enough volume, not enough geography. That's fine. Substitutes exist.

Run time-based holdouts instead: two weeks on, two weeks off, repeated. Noisy, but over three cycles the pattern is usually visible. Add a post-purchase survey from day one — it costs one question at checkout and becomes the most valuable dataset you own within a quarter. And lean harder on the creative-comparison layer, where sample sizes are smaller because you're measuring engagement, not purchase.

The honest position for a brand doing modest volume: you can know reliably which creative is better, and you can know approximately which channel is worth more. Anyone promising precise multi-touch credit at that scale is selling a model, not a measurement.

FAQ

Why do my platform-reported conversions exceed my actual orders?

Because each platform independently claims any conversion it touched within its attribution window, and buyers touch several platforms. Sum the claims and you exceed reality. Use your own order system as the denominator and treat platform numbers as relative comparisons only.

Should I use a 1-day or 7-day view-through window?

Start at 1 day for decision-making and keep 7-day visible for context. A long view-through window flatters video channels because video reaches a lot of people who were already going to buy. If performance looks very different at the two settings, the gap is roughly the size of the overclaim.

How often should I run an incrementality test?

Two to four times a year per major channel is realistic for most teams. They're disruptive and slow, so run them when a budget decision genuinely depends on the answer, not on a schedule.

Does organic video need attribution too?

It needs measurement, but a different kind. Track branded search lift, direct traffic, and survey mentions rather than click paths. Organic video usually shows up as demand creation, which click-based models are structurally unable to see.

What's the minimum setup for a small team?

Consistent UTMs with creative IDs, one post-purchase survey question, a weekly revenue number from your own system, and in-platform creative comparisons. That's four things, all cheap, and it beats most complicated setups that nobody maintains.

If your bottleneck is producing enough distinct creative to test cleanly in the first place, start on the models hub and pick by rank, price, or speed rather than by brand name.