Why most D2C brands are measuring ROAS wrong — and losing budget because of it
Marketing Analytics · Performance Strategy
The Attribution Lie
Why most D2C brands are measuring ROAS wrong — and losing budget because of it
For many D2C brands, ROAS looks clean on paper and broken in reality. The channel that closes the sale gets all the credit. The channels that spent weeks building desire get nothing. That is not measurement. That is a polite fiction your ad platform tells you — and your budget decisions follow it off a cliff.
Consider a real scenario: a consumer in Mumbai sees a reel from a creator she follows. Three days later she reads a blog review. A week after that she clicks a Google Shopping ad and converts. In last-click attribution, Google Shopping gets 100% of the credit. The reel, the review, the brand awareness built over that fortnight — statistically, they never existed.
This is the attribution lie, and for D2C brands operating across India, the USA, or the Middle East, it has become one of the most expensive problems in growth marketing.
The journey your dashboard never shows you
Modern purchasing is non-linear, multi-device, and increasingly private. A customer in Riyadh might discover a brand through an influencer story on Instagram, research it through YouTube comparisons, revisit through a WhatsApp business catalogue, and finally convert after clicking a branded search ad. Each step mattered. Only one shows up cleanly in your ROAS report.
A single customer journey — and how credit gets assigned
Day 1
Day 4
Day 9
Day 12
Day 14
Full credit
Significant credit
Partial credit
No credit
Google’s own attribution documentation acknowledges that attribution reports exist precisely to show the paths customers take and how different marketing efforts work together — because last-click alone misses assisted conversions and cross-channel influence. In other words, even the platform that benefits most from last-click attribution quietly admits it is incomplete.
Why last-click keeps winning anyway
The persistence of last-click attribution is not a mystery. It is comfortable. It is dashboard-friendly. It produces clean, defensible numbers in weekly reviews. A search campaign shows a 4.2x ROAS, so you increase its budget. A social prospecting campaign shows a 1.1x ROAS, so you cut it. The logic feels airtight. The logic is wrong.
Paid search can look like the hero channel even when it is mostly harvesting demand generated elsewhere. That is not a media strategy. That is a reporting bias.
— The structural problem with bottom-funnel-only measurement
When a customer searches your brand name, they already want you. That branded search term did not create the desire — the video ad, the creator post, the word-of-mouth recommendation did. But the search ad intercepts them at the moment of conversion and collects all the credit. Over time, teams shift more budget toward these high-ROAS search campaigns. The prospecting investment shrinks. New customers stop entering the funnel. Brand search volume slowly stagnates, and everyone wonders why “great ROAS” is not translating into great revenue growth.
The canary in the coal mine: If your branded search volume is flat or declining while your search ROAS looks strong, your upper-funnel investment has probably been cut too far. You are harvesting a field you are no longer seeding.
What data-driven attribution actually changes
Data-driven attribution (DDA) is a meaningfully better approach. Rather than applying a fixed rule — last click, first click, linear — it uses the actual data from your account to determine which touchpoints contributed most to conversions.
Google’s DDA model compares converting paths against non-converting paths to identify which ads, keywords, and campaigns had the greatest impact. It looks at interactions including clicks and video engagements across Search, YouTube, Display, and Demand Gen campaigns, then identifies patterns in paths that led to conversion. The result is a model that assigns partial credit across touchpoints rather than treating the last interaction as the only one that mattered.
Google Analytics 4 also gives teams the ability to choose attribution settings for key event reports, and changing the reporting attribution model applies to both historical and future data. That is a practical tool: it lets teams compare models side by side and see how dramatically reported performance changes — itself a signal of how fragile the current view actually is.
But data-driven attribution is not the full answer. It is still limited to the signals the platform can see, which in 2024 and beyond is a shrinking slice of the actual journey.
The three forces breaking modern attribution
1. Privacy changes have reduced signal
Apple’s App Tracking Transparency (ATT) framework, introduced with iOS 14.5 in April 2021, requires apps to explicitly ask users for permission before tracking them across other apps and websites. Without permission, the device advertising identifier (IDFA) is set to all zeros — making deterministic cross-funnel tracking impossible for a large share of iOS users. Meta publicly disclosed that ATT cost approximately $10 billion in ad revenue in 2022, primarily due to attribution and targeting disruption across iOS.
2. Consent requirements create structural gaps
In web environments, consent management under GDPR and similar frameworks creates measurement gaps that are particularly acute in Europe and increasingly relevant in India under the forthcoming Digital Personal Data Protection framework. Google’s consent mode documentation is explicit: when a user does not consent to ads or analytics cookies, Google tags adjust behavior and advertisers experience a direct measurement gap. Consent mode can model some of those gaps using aggregated signals, but the original deterministic visibility is permanently reduced.
3. Cross-device journeys are poorly captured
A customer who discovers your brand on a mobile Instagram reel and purchases three days later on a desktop browser will likely appear as two separate users in your analytics. Their journey is invisible. Offline touchpoints — store visits, WhatsApp conversations, phone calls, in-store purchases — compound the problem further. No single ad platform has a complete view, and most have incentives to report the parts of the path they can see as more important than the parts they cannot.
The signal loss reality for D2C in 2024
iOS ATT has reduced deterministic tracking for a large share of mobile users. Cookie consent has fragmented web measurement. Cross-device identity stitching is imperfect at best. Any single attribution report is, structurally, an undercount of total touchpoints and an overcount of easily measurable ones.
This is not an argument against measurement. It is an argument for measuring in layers rather than trusting a single number.
The measurement stack that serious brands actually use
Platform attribution
Analytics attribution
Incrementality testing
Marketing mix modeling
Google’s Meridian — its open-source marketing mix model released in 2024 — describes itself as built for privacy-durable, advanced measurement and causal inference, with incremental outcome at the center of its ROI approach. The fact that Google built and open-sourced an MMM tool is itself a signal: even the world’s largest ad platform acknowledges that cookie-based attribution is not sufficient for strategic planning.
Incrementality: the test that changes everything
A geo holdout test works by dividing your market into test and control regions. You run your campaign normally in test regions and withhold it in control regions, then compare revenue outcomes. The difference is your incremental lift — the result your campaign actually caused, rather than merely attributed to itself.
What brands often discover is confronting. A retargeting campaign with a reported ROAS of 8x might show incrementality of just 1.2x — because most of those converted users would have purchased anyway. The reported ROAS was measuring intent, not creating it.
If a Meta campaign looks strong in-platform but a geo holdout shows little incremental lift, the budget decision should follow the incrementality result — not the vanity ROAS.
— The correct hierarchy of measurement evidence
A practical framework for India, the USA, and the Middle East
In India, mobile-first journeys dominate. WhatsApp commerce, influencer-led discovery, and UPI-enabled micro-conversions create touchpoints that rarely appear in Western-designed attribution tools. MMM and incrementality testing are especially valuable here because they work at the business-outcome level, not the cookie level.
In the USA, iOS ATT has hit hardest, and signal loss from consent is most acute in states with strong privacy legislation. The brands that moved fastest to server-side tagging, first-party data infrastructure, and MMM are the ones maintaining measurement quality today.
In the Middle East, particularly Saudi Arabia and the UAE, e-commerce growth has been rapid and the customer base is highly mobile-engaged. Attribution complexity is compounded by high WhatsApp usage for commerce and by the prevalence of cash-on-delivery orders that break digital attribution chains entirely.
Three non-negotiable foundations
Clean event tracking first. Server-side tagging and enhanced conversions reduce signal loss and improve the quality of data feeding every model above it.
Normalize KPIs by channel role. Judging a prospecting campaign by the same last-click ROAS standard as branded search is like judging a farmer by how many apples they picked in October without asking whether they planted any trees in spring.
Run experiments before scaling. Before increasing any channel budget by more than 30%, run a geo test or lift study to confirm the incremental signal is real.
The red flags that attribution is lying to you
New customer growth is flat while reported ROAS is strong. The clearest signal that downstream capture channels are being credited for demand they did not create.
Performance swings dramatically when you change attribution windows. If moving from a 7-day to a 28-day window, or from last-click to data-driven, shuffles your channel rankings significantly, your model is fragile — not informative.
Platform revenue diverges from commerce backend revenue. If your ad platforms collectively report 2,000 conversions but Shopify shows 1,400 orders, the platforms are double-counting. Attribution overlap — where Meta, Google, and Snapchat all claim credit for the same sale — is extremely common. The sum of platform-reported ROAS is almost always higher than business reality.
ROAS is up but contribution margin is flat. If spend rises faster than incremental revenue, the media plan is winning inside the dashboard and losing in the business.
What CMOs should demand from their agency or internal team
The right measurement partner is not the one with the most impressive dashboard. It is the one who can explain why the dashboard looks the way it does, what it is missing, and how to make better decisions because of it. These are the questions they should be able to answer without deflecting:
- What attribution model are we using, and why is it appropriate for our funnel and customer journey?
- How does our analytics attribution differ from platform-reported attribution, and which should drive budget decisions?
- What percentage of reported conversions come from brand search, retargeting, and existing customers versus genuinely new demand?
- When did we last run an incrementality test, and what did it tell us about the causal impact of our top three channels?
- What is our plan for maintaining measurement quality as signal loss from iOS, Android Privacy Sandbox, and consent requirements continues to grow?
- How does our measurement account for offline touchpoints, cross-device journeys, and marketplace overlap?
If the answer to most of these is “our ROAS is strong,” the measurement system is incomplete and the budget decisions that follow from it are less reliable than they appear.
The honest conclusion
The attribution lie is not that attribution is useless. It is that bad attribution creates false confidence — and false confidence costs money in ways that are slow to surface and hard to attribute when they do.
Last-click is too narrow for modern commerce. Data-driven attribution is better but still incomplete. Incrementality testing tells you what actually caused lift. MMM tells you how to plan at scale. Used together, these methods triangulate something close to truth. Used in isolation, each one has a blind spot large enough to mislead a serious business.
The D2C brands that will maintain growth in India, the USA, and the Middle East are the ones building measurement discipline alongside marketing capability. A dashboard that makes you feel good is not the same as a dashboard that makes you better. The goal is the second one.
Frequently asked questions
Not entirely. It is useful for understanding closing behavior and optimizing bottom-funnel campaigns. The problem is using it as the primary or sole measure of channel performance. It over-credits the final touch and structurally ignores every earlier interaction in the journey.
Yes, in most cases. Google’s DDA model compares converting and non-converting paths to assign credit based on observed contribution. But it is still limited to the signals the platform can see, which privacy changes have materially reduced.
Because ROAS measures attributed conversions, not incremental ones. It rewards channels that capture existing demand even when those channels did not create it. Privacy signal loss and cross-device gaps compound the overcount further.
At minimum: server-side or enhanced event tracking, GA4 with data-driven attribution, consent-aware measurement, periodic incrementality testing via geo holdouts, and marketing mix modeling for strategic budget allocation. Each layer answers a different question.
Because D2C growth depends on both creating demand and capturing it — and those two functions are typically split across different channels. A last-click dashboard cannot distinguish between the two, which leads to systematic underinvestment in demand creation until growth stalls.
ROAS Measurement
Incrementality Testing
Marketing Mix Modeling
D2C Analytics
Data-Driven Attribution
