Four systems, one customer: who gets credit for the sale
Search, social, affiliate, and CRM each report the same sale. Tests suggest some of that credit was never earned, though not every time.
By Nathan Barling Product Marketing Principal 4 min read
In July, Hello Partner reported a Bloomberg test that found the shopping app Phia's mobile extension adding its own referral code at checkout on more than 50 websites. Phia told Engadget a recent release was “causing misattributions from a subset of users.” The everyday version of that problem needs no faulty software at all.
Illustrative path to a purchase
Each system reports the sale. Only one of them changed the outcome, and possibly none did.
Four systems now claim the sale. Search reports it, the social platform reports it, the affiliate network pays a commission, and the CRM program logs a win. Add up the reports and you have more credit than customers.
A system that's rewarded for conversions it touches will find the people who were already about to convert. Reported performance rises, the system earns more budget, and growth stays flat. Only a holdout, where some people never see the ad, shows which touch changed anything. Even then, cutting isn't the automatic answer.
Figure 1
Branded search returns the least incremental revenue per dollar
Median incremental ROAS by channel, 225 geo tests
Bar chart of median incremental ROAS by channel, with branded search lowest at 0.70x
Why every system claims the sale
Each system is rewarded for the conversions it can see and touch. Bidding models learn to find people close to buying, because those people convert most. Affiliate and coupon programs are often paid on last click. Retention flows send to anyone who might leave.
Marketing attribution assigns credit to the touchpoints before a sale. That's a different question from whether any of them caused it, and none of these systems is built to ask.
Count the claims
Before you run any test, there's a check you can do with data you already have. Pick a month. Add up the conversions each platform, affiliate network, and CRM program claims for it. Then divide by the orders you actually took.
We call that the claim ratio. At 1.0, every order is claimed once. At 1.6, your systems are reporting 60% more sales than happened. Some overlap is normal, since one customer often touches several systems. But the ratio shows how much credit is in circulation, and a ratio that rises while orders stay flat is the loop at work.
Split it by new and existing customers if you can. Existing customers are the people closest to buying, so that's where you'd expect over-claiming to show up first. There's no reliable public benchmark, so your first number is the baseline.
What the tests show
Incrementality testing asks the missing question by holding some people out and comparing results. Stella, a testing vendor, analyzed 225 geo-based tests (see geo-lift testing) run on its platform between August 2024 and December 2025. The typical campaign returned $2.31 in incremental revenue per dollar. Branded search returned $0.70.
Advertisers who run geo tests are measurement-minded and self-selected, so treat the ranking as direction. Stella itself suggests discounting results by 15 to 20% when planning new tests.
Affiliate and coupon programs
Haus switched off all US affiliate loyalty partners for a global DTC brand for seven weeks. The result was no change in revenue, even though affiliate looked like one of the brand's top channels in Google Analytics. Haus concluded the program had been claiming credit for purchases that would have happened anyway. It's a single brand, with no revenue figures published, but the test is easy to copy.
Don't cut brand search on reflex
The same kind of testing complicates the story. Haus found that branded search showed significant lift for 82% of brands facing high competition and 35% of brands facing low competition. A brand with many bidders after its name may need to defend it. A brand with none may be paying for clicks it already owned.
Figure 2
Whether branded search pays off depends on who else is bidding
Share of brands measuring significant lift from branded search
Bar chart: 82% of high-competition brands saw lift from branded search versus 35% of low-competition brands
So the honest reading is: test, then decide. Measurement itself is thin on this. The IAB's 2026 survey found 76% of buy-side decision-makers use incrementality tests, 73% use attribution, and 67% use marketing mix models. Only 39% use all three together. Three methods that never meet give three answers and no referee.
The platforms offer some controls. Google describes New Customer Value and New Customer Only bidding modes for Performance Max, plus brand exclusions. Meta's documentation says the existing-customer budget field is not available for new Advantage+ campaigns, and suggests two ad sets instead: one for new customers, one for existing customers with a spending limit. Each control covers one platform. Coordinating across search, social, affiliate, and CRM is a separate job, and it needs an owner.
What to do about it
Count the claims
Add up the conversions every system claims, including affiliate and CRM, and compare the total with real orders.
Test, then cut
Run holdouts or geo tests on brand search, retargeting, and affiliate. Cut only what the test shows isn't incremental.
Set the rules
Use new-customer goals, brand exclusions, suppression lists for CRM audiences, and written affiliate terms.
Judge on one number
Calibrate platform reports against incremental results, and agree who owns the reconciliation.
Where Alli fits
Alli Alerts watches the competitive pressure that decides whether branded search is worth defending. Strategic Auction Intelligence flags drops in brand impression share and new competitors entering an auction. Unified Data Layer brings signals from across platforms into one governed view, so a team can compare channels with the same logic. Scenario Planner uses diminishing-return curves to show where more spend stops producing efficient growth.
None of these runs a holdout test for you or decides what to cut. Your team still owns the test design and the decision.
Pick one customer path and trace it. Count how many systems claim the order.
Frequently asked questions
What is marketing cannibalization?
It happens when several channels or programs claim credit for the same customer, or when spend reaches people who would have bought anyway. Reported performance rises while real growth doesn't. Incrementality testing is the usual way to find it.
Is branded search worth paying for?
It depends on competition. A Haus study found significant lift for 82% of brands facing high competition and 35% facing low competition. Pause brand terms in some markets and compare results before you cut or defend.
What is incrementality in marketing?
Incrementality is the extra result caused by marketing that wouldn't have happened otherwise. You measure it by holding a group out, such as a region or audience, and comparing it to an exposed group. Platform-reported ROAS usually can't show it.
Do affiliate programs cannibalize sales?
They can. In one Haus test, a DTC brand turned off affiliate loyalty partners for seven weeks and saw no change in revenue. That's one brand, so test your own program and review partner terms, including browser extensions that can override referrals.
How do I stop Performance Max from targeting existing customers?
Google describes New Customer Value and New Customer Only bidding modes and brand exclusions. Check which options your account has and test the effect. Pair them with customer list exclusions so recent buyers aren't retargeted.
Sources
- Stella, 2025 DTC Digital Advertising Incrementality Benchmarks, May 26, 2026. Stella sells incrementality testing.
- Haus, Why last-click attribution is easy to game, 2026. Haus sells incrementality testing.
- Engadget, Phoebe Gates' AI shopping app Phia reportedly claimed unearned affiliate sales through fake clicks, July 11, 2026.
- Haus, When Is Branded Search Worth the Investment?, September 29, 2025. Haus sells incrementality testing.
- IAB and BWG Strategy, State of Data 2026, February 2026 (n=430).
- Meta for Developers, Advantage+ campaign experience, updated May 21, 2026. Meta sells ads.
- Google Ads Help, Answering your top questions about Performance Max. Undated; Google sells ads.
Last updated October 19, 2026. Statistics are quoted from the sources above and were checked on that date. Survey results are self-reported unless noted.