Marketing Attribution in 2026: Cookies, GA4, MMM and Incrementality

Honest attribution in 2026 combines three methods: GA4 and platform attribution for weekly optimisation, marketing mix modelling for budget allocation, and incrementality tests to prove what advertising actually caused. Chrome did not kill third-party cookies after all, but that did not restore old-style tracking, because consent rules, Safari and Firefox protections and self-reporting ad platforms still leave large gaps. This guide explains what changed, what each method is good for, and how a growing Indian business can combine them without an enterprise analytics team.

What did Google finally decide about third-party cookies?

Chrome keeps third-party cookies under users' existing settings, and most Privacy Sandbox replacements have been retired.

On 22 April 2025, Google's Privacy Sandbox team said it would maintain its current approach to third-party cookie choice in Chrome and would not roll out a new standalone prompt. On 17 October 2025 it published an update retiring most Privacy Sandbox technologies, including the Attribution Reporting API, Topics, Protected Audience, IP Protection and Related Website Sets, citing low adoption. CHIPS, FedCM and Private State Tokens continue, and Google said it would keep working on an interoperable attribution standard through the W3C.

So third-party cookies still work in Chrome for users who have not disabled them. That does not restore old-style tracking:

• Other browsers block them. Safari and Firefox restrict third-party cookies by default, so iPhone users browsing in Safari sit outside cookie-based tracking.
• Consent rules apply. India's DPDP Rules, notified in November 2025, bring most obligations into force 18 months after publication, as Storyboard18 reported; see our DPDP Act marketing compliance guide.
• Walled gardens do not share. Meta, Google and Amazon each report conversions by their own rules, and added together they often claim more sales than you made.

What are the three layers of modern measurement?

Attribution answers "what came before", MMM answers "how much did each channel contribute", and incrementality answers "did it cause extra sales".

No single layer is enough. Attribution guides weekly optimisation, MMM guides budget allocation, and incrementality tests keep both honest.

How does GA4 data-driven attribution work?

Data-driven attribution (DDA) uses your account's own conversion paths to estimate each touchpoint's contribution, instead of a fixed rule such as "last click gets everything".

Google removed the first click, linear, time decay and position-based models from Google Ads and GA4 in 2023, moving remaining conversion actions to data-driven attribution in September that year. Data-driven, last click and external attribution remain. To use DDA well:

• Tag everything consistently. UTM parameters on every non-Google link, including WhatsApp broadcasts, email and influencer bios, or those visits fall into "direct" and vanish.
• Define a few key events. A purchase, a qualified lead and perhaps a booked call. Too many key events dilute the model.
• Import offline outcomes. If sales close on the phone, push CRM stages back to GA4 or Google Ads so the model learns from revenue, not form fills.
• Remember what DDA cannot see. A podcast mention or a hoarding in Kochi leaves no click; DDA hands its credit to whatever was clicked next, usually branded search.

Where does Consent Mode v2 fit?

Consent Mode v2 makes Google tags respect consent choices while still allowing modelled conversions, and it is mandatory for advertisers reaching EEA and UK users.

Version 2 added ad_user_data and ad_personalization to the original ad_storage and analytics_storage signals. It became mandatory on 6 March 2024 for advertisers targeting the European Economic Area, including the UK, who want to keep remarketing and measurement features, as this technical guide explains.

In basic mode, tags do not load until the user consents. In advanced mode, tags load and send cookieless pings when consent is declined, which lets GA4 and Google Ads model missing conversions more accurately. For purely domestic Indian businesses the EEA mandate does not apply, but the same mechanism is a sensible way to honour consent captured under DPDP while preserving modelled measurement.

What is marketing mix modelling and is Meridian worth using?

MMM estimates each channel's contribution from spend and sales over time, needs no user-level tracking, and suits businesses with enough history.

Google's Meridian became available to everyone on 29 January 2025. It is open source, uses Bayesian causal inference, measures reach and frequency rather than raw impressions, and can take incrementality results as priors. Meta maintains its own open-source alternative, Robyn. Our rule of thumb for small businesses:

• You need about two years of weekly spend and sales data across several channels, with genuine variation in spend.
• You need someone comfortable with Python and statistics, or a partner who is.
• Below that threshold, a disciplined spreadsheet of weekly spend and sales by channel, plus occasional lift tests, teaches you more.

How do incrementality tests work?

An incrementality test compares a group that saw ads with a similar group that did not; the difference is the lift caused by advertising.

1. Platform conversion lift studies. Meta and Google split audiences into test and holdout groups. In November 2025 Google lowered its minimum incrementality test budget from nearly 100,000 US dollars to 5,000, using Bayesian methods, and recommends running studies for at least 14 days.
2. Geo tests. Turn a channel off in some cities while keeping it on in comparable ones, then compare sales. India's distinct metro markets make this practical for businesses selling in several cities, for example pausing in Kochi while running in Coimbatore.
3. Brand search holdouts. Pause branded search ads for a period and see how much traffic shifts to organic results instead of disappearing.

Plan each test with a hypothesis, a fixed duration and a decision attached. A test that cannot change a budget decision is not worth running.

How should a growing business put this together?

1. Fix the foundations. Consistent UTMs, a clean GA4 setup, a CRM that records lead source and outcome, and server-side event tracking.
2. Pick one source of truth for revenue. Your CRM or order system, not any ad platform.
3. Use DDA and platform data for weekly optimisation, accepting that each platform flatters itself.
4. Run one incrementality test per quarter on your largest or most questionable channel.
5. Graduate to MMM once you have the history and spend variation to support it.

Which method fits your stage?

• One or two paid channels, under a year of data: clean GA4, consistent UTMs, CRM source tracking and a monthly reconciliation. Skip MMM.
• Three or more channels and steady monthly spend: add one platform lift study or geo test per quarter, starting with the channel that claims the most credit.
• Two years of weekly data and real spend variation: pilot Meridian or Robyn, calibrated with your lift test results, and use it for annual and quarterly budget splits.

What should you do when the numbers disagree?

They will. Meta reports one figure, Google Ads another, GA4 a third and your CRM a fourth. Platforms count view-through conversions and use their own windows, GA4 sees only sessions it can track, and your CRM records what closed. Keep a monthly reconciliation that lists each source beside total revenue. Over time, the ratio between platform-reported conversions and real sales becomes a correction factor per channel, and a sudden change in that ratio is an early warning that tracking has broken.

Common mistakes

• Summing platform-reported conversions and calling it marketing revenue.
• Judging Performance Max or Advantage+ on their own ROAS without a holdout.
• Running a lift test during a festival sale, when everything spikes.
• Changing UTM naming every quarter, breaking year-on-year comparison.
• Counting WhatsApp enquiries as "direct" because click-to-chat links carry no UTM parameters.

How we approach measurement at Old Port

We set the revenue source of truth first, then fix tagging and offline imports, then add one lift test per quarter, and only recommend MMM once the data can support it. Better measurement feeds directly into how you allocate your marketing budget, and into honest judgement of automated campaigns such as Performance Max. Our MarTech and automation service sets up the tracking layer, and our performance marketing team uses it to plan spend. For brands also active in the Gulf, our UAE team's article on AI-run advertising with Meta Advantage+ and Google AI Max covers why holdouts matter more as campaigns automate.