Measuring marketing in 2027: MMM, incrementality and attribution for growing Shopify brands

Add up the conversions Meta, Google and your affiliate network each report for last month and the total will usually be higher than the orders in Shopify. Each platform counts the sales it touched, and they touched many of the same customers. Platform ROAS is a useful dial for running campaigns day to day. It is a poor basis for deciding where next year's money goes.

Tools that measure what your marketing caused have become far cheaper since 2025. Google cut the entry budget for Conversion Lift studies, Meta and Google both publish free marketing mix models, and post-purchase surveys cost almost nothing to run on Shopify.

Below: what each method answers, which combination fits your revenue band, how to run a first incrementality test, and how to turn the results into a 2027 budget. It is written for founders and marketing leads in the US, UK, Canada, Australia and Asia, so the privacy section covers each market.

52%

of US brand and agency marketers use incrementality testing (EMARKETER and TransUnion, July 2025)

47%

plan to invest in marketing mix modelling over the next 12 months (same survey)

US$5,000

minimum spend for a Google Ads Conversion Lift study under the Bayesian method

What does each measurement method answer?

Most measurement arguments start when someone uses one method to answer a question it was never built for. Four methods are worth knowing, and each answers a different question at a different speed.

Attribution (platform reporting and multi-touch tools) tells you which ads and clicks sat on the path to a purchase. Incrementality testing tells you how many of those purchases would not have happened without the ads. Marketing mix modelling (MMM) estimates how each channel contributes to total revenue over time, including offline and hard-to-track channels. Post-purchase surveys ask customers directly where they first heard of you.

Method	Question it answers	Speed	Main blind spot
Platform and multi-touch attribution	Which ads and clicks were on the path to purchase?	Daily	Counts correlation, not cause; each platform over-claims
Incrementality (lift tests, geo holdouts)	How many sales would not have happened without this channel?	2-6 weeks per test	One channel and one period at a time
Marketing mix modelling (MMM)	How much does each channel add to total revenue, and where is the next dollar best spent?	Monthly or quarterly refresh	Needs 1-2 years of clean weekly data; can be confidently wrong without test calibration
Post-purchase survey	Where do customers say they discovered us?	Daily	Relies on memory; under-reports channels people do not notice, such as search

The four measurement methods comparedSpeeds and data needs are typical working ranges, not platform rules.

Most measurement arguments start when someone uses one method to answer a question it was never built for.

Which measurement stack fits my revenue band?

You do not need every method from day one. The right stack depends on how much you spend, how many channels you run, and whether anyone on the team can maintain a model. The bands below are our guidance from working with Shopify brands, not hard thresholds. A brand at US$4M spending heavily across five channels may need the next band's stack early.

Annual revenue	Core stack	Testing cadence	Watch out for
Under US$5M	Platform reporting, GA4, a post-purchase survey, Shopify order data by region	One lift test or geo holdout on your biggest channel per half-year	Treating Meta ROAS as truth when it is your only channel
US$5-20M	All of the above, plus a lightweight MMM (Meridian or Robyn, in-house or via a partner)	Quarterly tests, rotating across your top 2-3 channels	Building an MMM before your weekly spend data is clean
US$20M+	Calibrated MMM refreshed monthly, a standing test calendar, multi-touch or modelled attribution for daily optimisation	Always-on testing, with every major channel tested at least yearly	Three tools giving three answers with nobody owning the final number

Suggested measurement stack by annual revenue (guidance only)Revenue bands are D2C Prominence guidance. Convert to local currency (GBP, CA$, A$, S$) at roughly equivalent scale.

How do I run my first incrementality test?

An incrementality test splits people or places into a group that can see your ads and a group that cannot, then compares sales. The difference is the lift your ads caused. There are three practical routes for a Shopify brand.

Google Ads Conversion Lift: Google moved these studies to a Bayesian method that uses results from comparable past studies, which lets them run on less data. Google's help centre lists a minimum spend of US$5,000 and shows results once certainty of lift passes 50%. Ask your Google rep or check your account's Experiments area for eligibility.
Meta Conversion Lift: Meta randomly holds back a control group of accounts that never see your ads and compares their conversions with the exposed group. Some advertisers can set this up in Ads Manager under Experiments, while others still need a Meta representative, so check your own account.
Geo holdout: switch a channel off (or hold it flat) in a set of matched regions, keep it running elsewhere, and compare Shopify orders by shipping region. It works on any channel, including affiliates, influencers and podcasts, and does not depend on platform tracking.

Read: how to pay affiliates for incremental sales, not coupon clicks

How should you use post-purchase surveys alongside the data?

A post-purchase survey is the cheapest measurement you can add, and the one most likely to catch what your tracking misses. One question on the Shopify thank-you page or order status page, such as "Where did you first hear about us?", picks up podcasts, creators, word of mouth and TikTok videos that never produced a tracked click. Several Shopify apps add this in an afternoon, and the answers sit next to the order, so you can cut them by product, market and order value.

Keep the question about first discovery, not the last thing the customer clicked, because your attribution tools already cover the last click. Randomise the order of the answer options so the first option does not win by default, include an "Other" field, and read the free text every month. Customers will name a creator or a newsletter you did not know was sending you sales.

Treat survey data as a direction finder. Customers forget, and they rarely credit search or retargeting even when both played a part. Its value is in the gap: when 15% of new customers say they found you through a channel that your platforms barely credit, that channel is the first candidate for a geo holdout or lift test.

Is open-source MMM worth it for a Shopify brand?

MMM used to mean a six-figure consultancy project. Two free, open-source models now cover much of that work.

Robyn comes from Meta's Marketing Science team and is available in R and Python. Meta still labels it experimental. It suits digital-heavy advertisers with plenty of weekly data and is built to explore how long ad effects last (adstock) and where returns start to flatten (saturation).

Meridian is Google's model. Google opened it to everyone on 29 January 2025 after a period of limited access, alongside a programme of certified measurement partners. It uses Bayesian methods, can take your lift test results as inputs, and in February 2026 Google added Scenario Planner, a no-code interface for testing budget scenarios without a data scientist.

You need at least one to two years of weekly spend and revenue by channel, plus promotions, pricing changes and stock-outs logged.
Someone has to own the model. A free model that nobody maintains is worse than none, because your team will stop trusting the numbers.
Calibrate with lift tests. An MMM fed with real experiment results is far more reliable than one guessing from correlations alone.
Remember who built it. Both models are honest open-source code, but neither vendor is neutral about its own channels, so check results against your tests.

Are third-party cookies going away in Chrome, and what else changed?

No. After years of planned removal, Google said in April 2025 that it would not launch a new third-party cookie prompt and would keep Chrome's existing cookie settings. On 17 October 2025 it went further and retired most Privacy Sandbox technologies, including the Attribution Reporting API and Topics, citing low adoption. Third-party cookies still work in Chrome unless a user blocks them.

Tracking is still getting harder. Safari and Firefox block third-party cookies, Apple's App Tracking Transparency still requires permission before apps track users, and privacy law keeps tightening. Click-based attribution will keep losing signal, which is why tests and models matter more each year.

Market	What applies	What it means for measurement
Global (iOS)	Apple App Tracking Transparency: apps must ask permission to track users across other companies' apps and sites	Fewer matched conversions from iOS users in Meta and other app-based reporting
US	No federal privacy law; a growing set of state laws	Opt-out rights in several states; plan for modelled conversions
UK	UK GDPR and PECR; the Data (Use and Access) Act 2025 added a narrow consent exemption for first-party statistical cookies, in force from 5 February 2026	Some first-party analytics can run without consent; ad tracking still needs it
EU	GDPR plus ePrivacy consent rules for non-essential cookies	Consent rates directly cap tracked conversions; use Consent Mode and modelling
Canada	PIPEDA remains the federal law after Bill C-27 lapsed in January 2025; Quebec's Law 25 fully in force since 22 September 2024	Stricter consent and transparency for Quebec customers
Australia	Privacy Act 1988, amended in 2024; a statutory tort for serious invasions of privacy started on 10 June 2025, with further reforms still pending	Review tracking disclosures now; expect tighter rules on targeting
Asia	India's DPDP Rules notified 13 November 2025, with most duties applying from May 2027; Singapore's PDPA requires consent and notice; Japan's APPI restricts passing cookie data to third parties that can identify users	Consent flows need to be set up market by market

Privacy context by market, high level (not legal advice)Summarised from official and legal sources listed below. Check with your own adviser before changing tracking.

How should I set 2027 budgets from tests instead of platform ROAS?

To connect tests with day-to-day reporting, give each channel an incrementality factor. Divide the incremental conversions your test measured by the conversions the platform reported for the same period. A factor below one means the platform over-claims, above one means it under-claims, which can happen with upper-funnel video or search that feeds branded demand.

Multiply platform ROAS by that factor to get incremental ROAS, and set budgets on that number. Your team can keep optimising inside each platform every day, because the factor converts their dashboard into something you can compare across channels.

Start with your largest channel, because a wrong factor there costs the most.
Set a target incremental ROAS per channel from your margin, not from the platform's default goals.
Move budget in steps of 10-20% and re-test before moving more, since returns flatten as spend rises.
Re-test each major channel at least once a year, and after any big change in creative, audience or bidding.
Feed every result into your MMM so the model and the tests agree.

What does an incrementality factor look like in practice?

Take a hypothetical brand spending US$40,000 a month on Meta prospecting. Ads Manager reports 1,000 purchases and a 3.0 ROAS. A geo holdout over the same four weeks finds that the regions with ads produced 600 more orders than the matched regions without them. The incrementality factor is 600 divided by 1,000, or 0.6, so the incremental ROAS is 3.0 multiplied by 0.6, which is 1.8.

Now compare that with the margin. If the brand needs a 2.0 ROAS to break even on a first order, Meta prospecting at this spend loses a little money on the first purchase. That might still be acceptable if repeat purchases pay it back within a few months, so the next step is to check the 90-day value of the customers the test brought in. Without the test, the dashboard said 3.0 and the budget would have gone up.

Run the same sum for search, affiliates and email, and you have a set of numbers you can compare across channels, each tied to a real test instead of each platform's own claims. The figures in this example are invented to show the method, so use your own results.

Your next step

Want a 2027 measurement plan built around your channels?

We will map your current reporting, pick the first test worth running, and show how to turn the result into next year's budget.

Book a strategy call Book a strategy call

Key takeaways

Platform ROAS shows what each network touched, not what it caused. Use it to run campaigns, not to set annual budgets.
Pick a stack that fits your scale: surveys and one test a half-year under US$5M, adding MMM between US$5M and US$20M, and always-on testing above that.
Google Ads Conversion Lift now starts at US$5,000 of spend, and geo holdouts work for any channel using Shopify order data.
Meridian and Robyn are free, but only worth it with clean weekly data, an owner, and lift-test calibration.
Chrome kept third-party cookies, yet iOS, other browsers and privacy laws keep eroding tracked signal in every market.
Turn each test into an incrementality factor and budget on incremental ROAS.