Measurement and attribution

Deduplicated attribution: why your platforms claim 300% of your sales

Every ad platform counts a sale as its own if its ad was anywhere near it, so the same order gets claimed two or three times. Deduplicated attribution hands each real order out once across all channels, so the totals match your store. It is the right base for budget decisions, but it still shows who was present, not who caused the sale.

A customer sees your serum in a TikTok video on Sunday. On Tuesday she clicks a Meta ad, browses, and leaves. On Thursday she Googles your brand name, clicks the top ad, and buys one $70 bottle.

You made $70. TikTok reports a $70 sale. Meta reports a $70 sale. Google reports a $70 sale. Your dashboards now show $210 of revenue from one bottle.

Do that across every order with more than one touch, and it's common for a brand running four or five channels to see platforms claim two to three times the sales it actually shipped. Deduplicated attribution is how you get back to one order, one credit.

One order, three receipts

Duplication isn't a bug. It's how each platform is built.

Meta, Google and TikTok each see only their own ads. When an order comes in, each one checks a simple thing: did this person click or see one of my ads inside my window? If yes, it counts the sale. None of them can see the others. None of them has any reason to hand credit away.

Three things make the overlap worse:

  • View windows. A platform can count a sale if someone merely saw an ad and bought within a set time. Meta's attribution setting can include a 1-day view window, for example. Someone who scrolled past your ad and bought because of a Google search still counts.
  • Long click windows. A 7-day click window means a click on Monday gets credit for a purchase the following Sunday, even if three other ads came in between.
  • Retargeting and brand search. These ads sit right before the purchase. They touch almost every buyer, so they claim almost every sale.

Here's why it matters for money. When every channel looks like it returns 5x, you can't tell which one to cut. Worse, you end up scaling the channels that are best at being near a sale, not the ones creating it. Platform-reported totals also make the whole program look far more profitable than your bank account says. That gap is how brands overspend for months before the P&L catches up.

What deduplicated attribution actually does

Deduplicated attribution looks at every order once, across every channel, and splits the credit so the total adds up to one. Ten real orders produce ten conversions of credit. Not twenty-five.

To do that it needs three things:

  1. A source of truth for orders. Your store or backend, not an ad platform. This sets the total.
  2. A path to conversion. The list of tracked touches before each order: clicks with UTMs, sessions, sometimes views.
  3. An attribution model. The rule that decides how to split credit along the path. Last click gives it all to the final touch. Data-driven models spread it based on patterns in your data.

The most common free tool for this is GA4. It applies one model across every channel it can see, so Meta and Google can't both claim the same order. GA4 currently offers data-driven attribution and two last-click models. The older first-click, linear, time-decay and position-based models were removed in 2023.

The model choice changes how credit is split. It doesn't change the core benefit: the total is capped at reality. That's the whole point. We explain where this fits in the bigger picture in unified measurement.

A worked example: 18,000 claimed orders, 6,000 real ones

Say a skincare brand spends $150k a month and ships 6,000 orders at a $70 average order value. Real revenue is $420k. Here's what each platform reports next to a deduplicated view that sums to 6,000.

ChannelSpendPlatform-claimed ordersDeduplicated ordersInflation (claimed ÷ dedup)
Meta$70,0006,2002,1002.95x
Google brand search$8,0002,4001,3001.85x
Google non-brand and Shopping$32,0002,6001,2002.17x
TikTok$25,0003,1005006.20x
Affiliates$15,0003,7004009.25x
Email, organic, directn/an/a500n/a
Total$150,00018,0006,0003.00x

The platforms claim 18,000 orders, or $1.26M in revenue. The business made $420k. That's 300%.

Now turn orders into ROAS. ROAS is attributed revenue divided by spend. Revenue here is orders × $70.

ChannelPlatform ROASDeduplicated ROAS
Meta6,200 × $70 ÷ $70,000 = 6.2x2,100 × $70 ÷ $70,000 = 2.1x
Google brand2,400 × $70 ÷ $8,000 = 21.0x1,300 × $70 ÷ $8,000 = 11.4x
Google non-brand2,600 × $70 ÷ $32,000 = 5.7x1,200 × $70 ÷ $32,000 = 2.6x
TikTok3,100 × $70 ÷ $25,000 = 8.7x500 × $70 ÷ $25,000 = 1.4x
Affiliates3,700 × $70 ÷ $15,000 = 17.3x400 × $70 ÷ $15,000 = 1.9x

Next, the profit line. Say the brand keeps 60% of each sale after product cost, shipping and fees. Break-even ROAS is 1 ÷ 0.60 = 1.67x. Below that, each dollar of ads loses money. (Our break-even ROAS guide walks through this math.)

On platform numbers, every channel clears break-even by a mile. On deduplicated numbers:

  • Meta at 2.1x and non-brand search at 2.6x are profitable.
  • Brand search looks fantastic, but read on: it's the classic case of a channel touching buyers who were coming anyway.
  • Affiliates at 1.9x barely clear the bar.
  • TikTok at 1.4x is below break-even.

Look at the inflation column too. It tells you where duplication hides. Meta and search are inflated two to three times. TikTok and affiliates are inflated six to nine times. Those channels mostly show up in paths that other channels also touched. Affiliates, in particular, often catch shoppers at checkout with a coupon code.

The decision changes completely. On platform data, you might add budget to TikTok, since 8.7x looks like the best prospecting channel. On deduplicated data, TikTok is the first channel to question.

Deduplication fixes double counting, not cause and effect

This is the part people skip. Deduplicated attribution answers "who touched this order?" It does not answer "who caused it?"

Take TikTok. Its deduplicated ROAS is 1.4x, but that number comes mostly from tracked clicks. TikTok is a channel people watch and then search for later. A lot of its value may land in the "brand search" and "direct" rows. So the deduplicated view could be underselling it.

Brand search is the reverse. It gets 1,300 deduplicated orders because it sits at the end of so many paths. But many of those buyers typed your name because they already wanted you. If you paused brand ads, plenty of them would click the organic link right below.

So treat deduplicated attribution as the honest starting line, not the finish. It removes the obvious overcount. Then you check the channels you doubt with lift tests or other validation. That's what turns a deduplicated number into an incremental one.

Where deduplicated views still go wrong

Even a clean deduplicated view has blind spots. Know them before you trust it.

Click-only tracking misses views. GA4 sees sessions and clicks. It does not see someone who watched a video and never clicked. Upper-funnel video and CTV tend to look worse than they are.

Untagged traffic falls into "direct." If your ad links lack UTMs, or a redirect strips them, paid sales land in direct or organic. Fix tagging first with a consistent UTM naming convention and the UTM builder.

Privacy limits break paths. Since iOS 14.5, apps must ask permission before tracking people across other apps and sites, under Apple's App Tracking Transparency rules. Cookie consent and short cookie lifetimes cut paths too. Cross-device journeys, like phone then laptop, often split into two strangers.

The total doesn't match your store. If GA4 shows 5,100 purchases and Shopify shows 6,000, your deduplicated view is missing 15% of orders. Usually that's a tagging gap, consent loss, or a broken purchase event. Audit with the GA4 purchase event QA checklist.

Offline and untracked channels vanish. Podcasts, influencers without links, direct mail and billboards rarely appear in any path. They need surveys, tests or MMM.

Event deduplication is a different thing. If you send purchases to Meta through both the pixel and the Conversions API, Meta needs to dedupe those two copies of the same event. That stops Meta from double counting itself. It does nothing about Meta and Google claiming the same order. Both matter. Our CAPI deduplication audit covers the first kind.

How to build one without a data team

You don't need a warehouse to start. You need a few disciplines.

  1. Tag every paid link with UTMs that identify platform, campaign and ad. Same format everywhere.
  2. Pick one cross-channel tool for the deduplicated view. GA4 is fine for most brands. Your own warehouse is better at scale.
  3. Pick one model and stick with it. Switching models every month makes trends meaningless. Data-driven or last click both work as a base, as long as you know their bias.
  4. Reconcile to real orders weekly. The deduplicated total should land within a few percent of your store. If it doesn't, fix that before reading channel numbers.
  5. Report platform numbers next to deduplicated ones. The ratio between them is useful. A rising inflation factor means more overlap, often from new retargeting or a new channel joining the same paths.

What to do Monday

  • Write down last month's real orders and revenue from your store. That's the denominator for everything.
  • Sum what each platform claims. Divide by real orders. That's your overclaim rate. Share it with the team.
  • Build the two tables above for your own channels: claimed vs deduplicated orders, then platform ROAS vs deduplicated ROAS.
  • Compute break-even ROAS from your contribution margin. Mark every channel above or below it on deduplicated numbers.
  • Flag channels with an inflation factor above about 4x. Those are mostly riding other channels' paths. They're your first candidates for a holdout test.
  • Don't cut a below-break-even channel on deduplicated data alone if it's a view-heavy channel like TikTok or YouTube. Test it first.
  • Set the rule: budget moves between channels use deduplicated or incremental numbers. Platform numbers are for optimizing inside each platform.

An AI ad agent needs the same discipline. If it reads each platform's ROAS at face value, it will reward whichever channel claims the most overlap. It should judge spend on numbers that add up to real orders.

Back to one bottle, one sale

That $70 bottle of serum was one sale. Your dashboards turned it into $210. Spread across a month, that's how a brand doing $420k in real revenue reads $1.26M in its ad accounts and feels rich.

Deduplicated attribution won't tell you everything. It will tell you the truth about the total. And you can't make a single good budget call until the total is true.

Meta says 120 sales, Shopify says 70: why they disagree and which number to trust

Meta and Shopify count sales from your ads differently, so their numbers never match. See the six reasons for the gap, a step-by-step bridge from one number to the other, and which number to use for which decision. Free Excel reconciler included.

Unified measurement: why no single ad number tells the truth

Unified measurement combines deduplicated attribution, MMM, lift tests and surveys into one KPI you can budget on. Here's how it works, with a worked example.

Calculate break-even ROAS with contribution margin

Calculate an advertising break-even ROAS from retained revenue and variable costs, then separate that threshold from overhead, profit targets, and attribution claims.

Use attribution and incrementality for different decisions

Separate advertising credit assignment from causal lift, and choose reporting, experiments, or modeling based on the budget question you need to answer.

Have a correction or a question about the workflow? Contact GaaS. Read our editorial standards for sourcing and example conventions.