Measurement and attribution

Attribution is a set of rules for deciding which ad gets credit for a sale — not the truth itself. Different rules look at the same account and produce different answers. Getting measurement right is not about picking the correct rule; it is about knowing where the rule is wrong.

How signal flows

Layer	Function	Typical loss
Browser pixel	Captures on-site events	Ad blockers, cookie restrictions, ITP
Server-side API	Sends the event from your server	Missing parameters, late delivery
Matching	Ties the event to a user	Weak customer data (no email, no phone)
Reporting	Credits the event to an ad	Attribution window and model differences

The browser pixel alone has not been sufficient for years. Server-side delivery (Conversions API on Meta, enhanced conversions on Google) both recovers lost events and improves match quality. If match quality is poor, server-side integration will not save you — send hashed email, phone, name and location fields as completely as you can.

Attribution window

The window answers: "how many days after seeing the ad should I still credit the sale to it?" Widening the window does not improve performance; it just assigns more sales to the channel.

Window	When it fits
1-day click	Short consideration, impulse purchases
7-day click	The default for most ecommerce
7-day click + 1-day view	Video-heavy accounts with upper-funnel investment
28 days or more	Long sales cycles, high-ticket products

UTM discipline

The platform report is the channel's own claim; your analytics is your record. To compare the two you need a consistent UTM taxonomy.

utm_source   = platform          (meta, google, tiktok)
utm_medium   = traffic type      (paid_social, cpc)
utm_campaign = campaign name     (winter-sale-2026)
utm_content  = creative identity (ugc-hook-a-9x16)

The rule is simple: lowercase, no spaces, one fixed abbreviation standard. If you do not break utm_content down to the creative level, your analytics will never tell you which asset actually sold.

The structural problem with platform reports

Each platform sees only its own impressions and claims credit on its own window. That has three direct consequences:

• Overlap. The same sale can appear in both the Meta and Google reports.
• Harvesting. A retargeting ad shown to someone who was already going to buy claims a sale it did not create.
• Invisibility. The upper-funnel video that triggered a branded search loses the credit to the search campaign.

This is why allocating channel budgets purely on platform ROAS systematically overfeeds the bottom of the funnel and starves the top.

Incrementality: measuring causation

The only real question is: if I stopped this spend, would these sales still happen? Attribution cannot answer it. Experiments can.

Method	How it works	When it fits
Geo test	Ads run in some regions, are held back in others; the gap is measured	Brands spending at national scale
Audience holdout	A percentage of the audience is never shown ads	When the platform has a built-in lift tool
Blackout test	A channel or campaign is fully paused for a set period	Small budgets — the most practical option

The cheapest starting point is a blackout test: pause the campaign you suspect for two weeks and watch total revenue. If total revenue does not fall, that spend was not producing sales.

Practical setup order

1. Set up server-side conversion delivery and check match quality
2. Choose one attribution window and lock it
3. Push the UTM taxonomy down to the creative level
4. Start weekly MER tracking
5. Run an incrementality test on your largest line item once a quarter