Neither platform is universally better for e-commerce. Google Ads is the stronger first test when identifiable demand already exists, your product feed is clean, and query and category fit are the binding opportunity. Meta Ads is the stronger first test when the product needs visual explanation, creative supply is the growth input you can actually increase, and you have a reliable purchase and value event. Both platforms still have to answer to the same business denominator: contribution after the costs of serving the order.
That last sentence is where most comparisons fall apart. A platform ROAS of 4.1 on Meta and 6.3 on Google are not two measurements of the same thing. They are two vendor-reported numbers built on different attribution windows, different conversion definitions, different customer mixes, and different amounts of demand that would have converted anyway. Choosing a channel by comparing them is not analysis. It is a coin flip with a spreadsheet attached.
This article is a buyer-side operating framework: how to define the job you are hiring a channel to do, what has to be ready before either platform deserves budget, how to read the four common failure patterns, and how to reconcile both platforms against finance-owned economics without forcing them into one attribution model.
Write the channel job contract before you compare anything
A channel comparison without a defined job produces an argument, not a decision. Before you look at a single platform report, write a short contract — one page is enough — that fixes nine things.
Business decision. What will change based on this comparison? "Where does the next £40k test budget go" is a decision. "Which platform is better" is not.
Demand state to reach. Are you trying to capture people already searching for your category, reach people who have never heard of the product, re-engage people who abandoned a cart, or convert competitor consideration? Each is a different job, and some are far better served by one platform than the other.
Product and SKU scope. Whole catalogue, one hero product, one margin band, or one new launch? Channel fit changes by SKU more often than teams expect.
New versus returning customer treatment. Decide now whether the metric that governs this decision counts all orders or new customers only. Deciding after you see the numbers is how motivated reasoning enters.
Conversion and value definition. Which event, fired where, carrying what value — order value, net revenue, or contribution? Both platforms will optimise toward whatever you send them.
Attribution and reporting window. Which platform windows you will read, and which independent reporting period you will judge against.
Contribution boundary and CAC ceiling. The maximum acquisition cost the economics tolerate, derived from finance-owned unit economics rather than a target pulled from a benchmark post. The method for deriving it is covered in our e-commerce contribution margin framework.
Owner and review date. One named owner, one date the contract is revisited.
Pass, constrain, and stop rules. Written before the test starts. What result would make you fund this channel further, what result would restrict it to a subset, and what result would end it.
Teams that skip this step end up relitigating the comparison every month, because there was never an agreed definition of winning.
The comparison matrix, with the overlaps left in
"Google captures demand, Meta creates it" is a useful first sentence and a bad operating model. Google's Demand Gen and Performance Max inventory reaches people who are not searching. Meta's catalogue and retargeting surfaces frequently harvest demand that already exists. The honest comparison is about inputs, control, and failure modes.
Criterion Google Ads Meta Ads --- --- --- Primary demand state Existing, expressed demand — plus non-search inventory that reaches latent demand Latent and prompted demand — plus catalogue surfaces that harvest existing intent Discovery mechanism Query, category, feed match, audience and contextual signals across Search, Shopping, YouTube and partner inventory Interest, behaviour and lookalike modelling against a creative in a social feed Core inputs Product feed quality, query and category fit, landing page relevance, conversion values Creative volume and variety, catalogue quality, reliable purchase and value events Creative burden Moderate for Search; substantial for Performance Max and video inventory, which consume asset groups across formats High and continuous — creative is the primary lever and it fatigues Catalogue and feed burden High — Merchant Center product data has required and recommended attributes that govern eligibility and matching (Google Merchant Center Help, Product data specification) Moderate to high for catalogue-driven campaigns; lower for pure creative-led prospecting Signal and event requirements Conversion actions with values, which can be static or passed dynamically from the site (Google Ads Help, Set or change conversion values) Pixel plus server-side events; the Conversions API is the documented path for sending events from your server (Meta Business SDK, Conversions API reference) Controllability and diagnostic surface High on Search (query-level); reduced on automated campaign types where placement and query reporting is aggregated Moderate; automated sales campaigns consolidate targeting and placement decisions inside the system Conversion lag Often short for high-intent search; longer on upper-funnel inventory Typically longer and more variable, especially for considered purchases Attribution and incrementality risk Branded search capture — credit for demand that would have converted anyway View-through credit and modelled conversions inflating apparent contribution Common false positive Strong ROAS driven mostly by branded and returning-customer traffic Strong ROAS driven mostly by retargeting a warm audience Disqualifying condition No meaningful category or competitor query volume, or a feed that cannot be made compliant No creative throughput, or a product whose value is not demonstrable in a feed unit
Two overlaps deserve emphasis. First, automation has narrowed the structural difference: Performance Max spans Google's inventory as a single campaign type and decides placement internally (Google Ads Help, About Performance Max campaigns), while Meta's Advantage+ sales campaigns consolidate targeting and budget decisions similarly (Meta Blueprint, Advantage+ sales campaigns). On both platforms, the operator input has shifted from targeting toward inputs: feeds, creative, events, and values. Second, both platforms will happily report profitable-looking performance on demand you already owned. The false positives differ in mechanism, not in consequence.
These are descriptions of documented product behaviour. Neither vendor's documentation is evidence that its platform performs better commercially for your business, and vendor performance claims are not independent evidence.
Seven readiness gates before either platform gets budget
Most "Google versus Meta" failures are not channel-selection failures. They are readiness failures that a channel exposed. Run these seven gates first. Each has evidence you can point at, a failure mode, and a smallest responsible fix.
Gate 1 — Verified product economics
Evidence: a finance-owned contribution calculation per order and per SKU, with cost lines labelled verified, estimated, or lagged, and a derived CAC ceiling. Failure mode: a channel is judged against a ROAS target nobody can trace to a cost structure. Smallest fix: build the contribution waterfall for your top three SKUs before spending, not for the whole catalogue.
Gate 2 — A clean purchase and value event
Evidence: purchase events reconcile to order IDs in your commerce platform within a known tolerance, with values matching a defined revenue column. Failure mode: duplicate, missing, or inflated events train bidding toward the wrong customers on either platform. Smallest fix: reconcile one week of platform-reported purchases against store orders and resolve the delta before adding server-side complexity. Implementation is covered in our server-side tracking guide for Meta CAPI and Google enhanced conversions.
Gate 3 — Product and inventory data integrity
Evidence: feed passes required attribute checks, prices and availability match the storefront, and identifiers are stable. Failure mode: disapprovals and mismatches suppress Shopping and catalogue delivery, which then reads as "Google doesn't work for us". Smallest fix: fix required attributes and price/availability sync for the SKUs in scope, not the entire catalogue.
Gate 4 — Landing page and storefront readiness
Evidence: the destination matches the promise in the ad, loads quickly on mobile, and has a checkout path that does not lose qualified traffic. Failure mode: both channels get blamed for a conversion-rate problem neither owns. Smallest fix: one dedicated destination per tested offer. Our paid-traffic landing page CRO guide covers the diagnostic in detail.
Gate 5 — Creative throughput and message fit
Evidence: a realistic monthly count of distinct concepts — not variations — your team can produce, plus evidence about which message registers. Failure mode: a Meta test is declared a failure after four assets, which measures your production capacity rather than the channel. Smallest fix: commit to a defined number of distinct concepts for the test window, using the approach in our Meta Ads creative testing framework.
Gate 6 — Enough eligible demand or audience opportunity
Evidence: for Google, meaningful non-branded category and competitor query activity in your markets; for Meta, an addressable audience large enough to exit the learning phase at a sane budget. Failure mode: a "test" that never accumulates enough conversions to say anything. Smallest fix: size the opportunity before committing budget, and narrow the geography or SKU scope until the test is statistically meaningful at the budget you have.
Gate 7 — Measurement and experiment plan
Evidence: a written plan naming the independent metric, the period, the comparison, and the decision rules. Failure mode: the test ends and the debate is about which dashboard to believe. Smallest fix: agree the source of truth before launch, using the hierarchy in our MER versus ROAS budget framework.
A channel that fails a gate has not been disproven. It has been shown to be untestable today. Those are very different conclusions, and conflating them is how brands abandon viable channels for two years.
Role, Readiness, Evidence: a decision model
Three questions, answered in order. Answering them out of order is what produces circular arguments.
Role — what commercial job should this channel do? Capture existing category demand at a defensible CAC. Introduce a new product to people who do not know it exists. Defend competitor consideration. Recover abandoned carts. A channel without a named role cannot fail or succeed; it can only produce numbers.
Readiness — are the channel-specific inputs fit for that role? Google's binding input is usually feed and query fit; Meta's is usually creative supply and event quality. A role can be right while readiness is absent — that is a Fix, not a Stop.
Evidence — do the results reconcile outside the ad platform? Does the store show more new customers, more net revenue, and better contribution during the period, in a way that survives scrutiny of seasonality, promotions, and other channels? If the only evidence is the platform's own report, you have a claim rather than evidence.
Score each as adequate or inadequate. Role inadequate means stop and rewrite the contract. Readiness inadequate means fix the input before judging the channel. Evidence inadequate means design a better measurement, not a bigger budget.
Four diagnoses you will actually face
"Meta appears to work but Google does not"
Symptom: Meta reports strong ROAS; Google spends inefficiently or barely spends at all. Likely owners, in order: feed quality and disapprovals suppressing Shopping delivery; negligible non-branded query volume in a genuinely new category; conversion values mismatched between platforms so Google optimises toward the wrong orders; a Meta figure inflated by view-through credit and retargeting. Next proving test: audit feed compliance and query coverage for the SKUs in scope, then run a Meta holdout to establish how much of that strong ROAS is incremental.
"Google appears to work but Meta does not"
Symptom: Google reports efficient acquisition; Meta cannot find a profitable audience. Likely owners: Google efficiency concentrated in branded search and returning customers; Meta starved of distinct creative concepts; broken or duplicated purchase events preventing exit from learning; a product that is genuinely hard to explain in a feed unit. Next proving test: segment Google performance into branded versus non-branded and new versus returning before crediting the channel — our brand bidding incrementality guide sets out the method. Separately, verify Meta event integrity before adding creative volume.
"Both report strong ROAS but MER and contribution do not move"
Symptom: the sum of platform-reported revenue approaches or exceeds actual store revenue. Likely owners: duplicate credit across platforms for the same orders; heavy retargeting on both sides; discount-driven orders with thin contribution; returns arriving after the reporting window; a new-customer mix that has quietly collapsed. Next proving test: rebuild the period from order-level commerce data with new-customer flags and returns, compare blended spend against net revenue, and treat both platform reports as directional inputs only.
"Neither works"
Symptom: sustained spend on both platforms, no defensible contribution. Likely owners: an offer or price problem no channel can solve; a conversion-rate problem on the storefront; economics that leave no room for paid acquisition at any realistic CAC; or a measurement system so unreliable that success would be invisible. Next proving test: validate the CAC ceiling against verified unit economics first. If the ceiling is below a plausible market cost for your category, the problem is the business model, not the channel.
None of these diagnoses generalise. The value is in separating the symptom from the owner and from the test that would prove it.
A staged allocation workflow, without universal percentages
There is no correct split. Anyone quoting one has not seen your economics. What is reusable is the sequence.
Stage 1 — Protect the source of truth and the economics. Fix the denominator first: net revenue definition, new-customer flag, returns handling, blended spend, CAC ceiling. Skipping this stage means every later result is unfalsifiable.
Stage 2 — Run one interpretable channel-role test. One platform, one named role, one scope, a budget large enough to reach a decision, and a duration that accommodates your conversion lag. Two simultaneous first tests produce an attribution argument rather than a finding.
Stage 3 — Validate downstream contribution and cohort quality. Do the acquired orders behave? Check returns, discount dependence, new-customer share, and early repeat behaviour before declaring the role satisfied.
Stage 4 — Add the second channel only when its incremental job is explicit. "Diversification" is not a job. "Reach people who never search this category, at a CAC below the ceiling, measured against a holdout" is.
Stage 5 — Use holdout or lift methods when spend or risk justify them. Geo holdouts and structured experiments cost attention and some short-term revenue, which is only worth paying above a certain spend level. Platform-native experiment tooling exists for this on the Google side (Google Ads Help, About Performance Max experiments), and method selection is covered in our incrementality testing and geo holdout guide.
A worked example (illustrative only)
The numbers below are invented for illustration. They are not benchmarks, client results, or Sharply Labs results, and they should not be used as targets.
A fictional DTC brand sells three products and has £30,000 for one quarter of testing. Finance has confirmed a pre-acquisition contribution of £38 per order on a £90 average order value.
Two campaigns run for eight weeks:
Google Ads Meta Ads --- --- --- Spend £15,000 £15,000 Platform-reported revenue £94,500 £61,500 Platform-reported ROAS 6.3 4.1 Platform-reported orders 1,050 683
On the platform numbers, Google wins comfortably. Now apply the business denominator.
Order-level store data shows that 620 of Google's 1,050 reported orders came from customers who had purchased before, and a further 210 arrived through branded queries. A geo holdout suggests roughly 60% of the branded orders would have arrived regardless. Meta's 683 reported orders include 240 credited on view-through only, of which store data can match 95 to actual orders in the window; 510 of the matched orders are from new customers. Returns settle at 9% on Google-attributed orders and 14% on Meta-attributed orders, because the Meta cohort skews toward one higher-return product.
Rebuilt on new customers and net of estimated non-incremental branded orders and returns:
Google Ads Meta Ads --- --- --- Credible new-customer orders 304 439 Contribution before acquisition cost (at £38) £11,552 £16,682 Acquisition cost £15,000 £15,000 Post-acquisition contribution −£3,448 £1,682 Effective new-customer CAC £49.34 £34.17
The ranking inverts. Not because Meta is better, but because the platform-reported figures measured different populations: Google's number was substantially a re-measurement of demand the brand already had, and Meta's was suppressed by view-through credit that store data could not match and by a product mix with higher returns.
The correct conclusion from this fictional case is narrow. Google's role should be rewritten — it may be a retention and defence channel here rather than an acquisition channel — and its non-branded performance should be isolated and retested. Meta's role is supported but needs a second period to confirm the return rate and the repeat behaviour of that cohort. This is a decision about two roles, not a verdict about two platforms.
Keep, Fix, Constrain, Stop as channel-role outcomes
These are decisions about a role in your account at a point in time. They are not verdicts about a platform.
Keep — the role is right and mature contribution evidence reconciles outside the platform. Fund it, and watch marginal rather than average contribution as you scale.
Fix — the role is valid but an input is faulty: feed errors, creative starvation, a broken or duplicated event, a mismatched landing page, or an attribution definition nobody agreed. Repair the input and retest. Judging a channel through a broken input is the single most expensive mistake in this category.
Constrain — the channel works inside a boundary: certain SKUs, non-branded queries only, specific geographies, particular placements, or below a spend ceiling where marginal contribution turns negative. Write the boundary into the account structure, not the meeting notes. Meta-side structural boundaries are covered in our Meta Ads account structure guide.
Stop — evidence is mature, inputs are corrected, and the channel still cannot satisfy the commercial decision. Stop with the evidence documented, including what would have to change for it to be reconsidered.
Comparing two platforms without forcing one attribution model
You will never reconcile Google's and Meta's attribution to each other, and trying is wasted effort. Reconcile both to a third thing you own.
Order IDs and commerce data. The store's order list, not either platform's conversion count, is the population you are measuring.
Finance-owned net revenue. Use one named revenue definition from your commerce reporting — gross sales, discounts, returns, net sales, shipping and taxes are separately reported columns, so the definition must name the column (Shopify Help Center, Sales reports).
New-customer flags. Applied at order level in your own data, never taken from a platform's segmentation.
Returns lag. Hold judgement until the return window has largely closed, or apply a labelled estimate and say it is an estimate.
Blended spend. All paid spend in the period, including fees, against the period's net revenue.
Contribution margin. The final filter, applied consistently to both channels.
Then read platform reports as diagnostics, not as accounting. Platform data is excellent for telling you which creative, query, or asset group is performing relative to its own peers, and poor for telling you what the business earned. Value-based bidding depends on the accuracy of the values you supply (Google Ads Help, Value-based bidding FAQ), which makes the reconciliation loop an input to performance, not just to reporting. Break-even logic — comparing costs against revenue to decide whether an activity is worth funding — is the underlying discipline (U.S. Small Business Administration, Plan your business).
Limitations and tradeoffs you should state out loud
Branded search capture inflates Google efficiency where brand demand exists independently.
View-through attribution credits impressions without clicks and is not comparable across platforms.
Modelled conversions fill gaps left by consent and tracking loss; they are estimates, not observations.
Cross-device behaviour breaks single-session attribution on both platforms.
Consent and privacy loss reduces observable conversions unevenly by market and device.
Repeat purchase makes first-order economics misleading for subscription-like categories.
Catalogue and feed mismatches silently suppress delivery and read as poor channel performance. Server-side event sending is the documented mitigation for signal loss on the Meta side (Meta for Business, Get started with the Conversions API).
Promotion periods distort any comparison spanning them.
Inventory constraints cap a channel's real ceiling regardless of demand.
Creative fatigue means Meta results decay without continuous supply; a flat test period understates the channel's requirements.
Query opacity and automation limit diagnosis on consolidated campaign types.
Conversion lag makes short tests systematically favour the faster-converting channel.
Incrementality is the limitation that contains the others: correlation with spend is not proof of causation.
Questions buyers actually ask
Is Google Ads or Meta Ads better for e-commerce?
Neither, in general. The better channel is the one whose core input — demand and feed quality for Google, creative supply and event quality for Meta — you can actually satisfy, aimed at a role your economics support.
Which one is cheaper?
Cost per click is not a comparison. What matters is cost per credible new customer against your CAC ceiling. A cheaper click that acquires a returning customer you already owned is more expensive than an expensive click that acquires a new one.
Should a new DTC brand start with Google or Meta?
If people already search for your category and you can produce a compliant feed, Google's non-branded search and Shopping is usually the more interpretable first test. If the product needs to be shown to be understood and you can produce distinct creative concepts consistently, Meta is usually the more interpretable first test. In both cases, start with one.
Can an e-commerce brand use both?
Yes, and most eventually do. The condition is that each channel has a distinct named role and that you have a reconciliation method outside the platforms. Running both without that produces double-counted credit and an argument every month.
How much budget should go to each?
There is no defensible universal split. Allocate to roles, cap by CAC ceiling, and let contribution evidence move the allocation. Treat any percentage recommendation made without sight of your economics as marketing content.
How long should a fair test run?
Long enough to accumulate a decision-grade number of conversions and to cover your conversion lag and return window — typically longer for Meta and for considered purchases. Set the duration before launch, in the contract, and resist ending it early on an encouraging week.
Why do the platforms report different ROAS?
Different attribution windows, different click and view-through rules, different modelling, different de-duplication, and different underlying customer populations. Each number is internally coherent and not comparable to the other. This is the reason for a shared external denominator.
What should an agency audit before reallocating spend?
Conversion and value integrity, feed health, new-customer share, returns, branded versus non-branded split, creative supply, landing-page performance, and whether the CAC ceiling was derived from verified economics. An agency that proposes a reallocation before seeing those things is guessing with your budget.
Minimum viable evidence pack
Before a Google-versus-Meta decision is defensible, you need:
Order-level commerce export for the period, with order IDs, discounts, returns, and new-customer flags.
A finance-owned net revenue definition naming the report and column used.
Contribution per order and per in-scope SKU, with cost lines labelled verified, estimated, or lagged.
A derived CAC ceiling and break-even boundary.
Full paid spend, including fees, for the same period.
Platform exports from both channels covering the same period, with stated attribution settings.
Branded versus non-branded segmentation on the Google side.
Creative concept count and cadence on the Meta side.
Feed health status for the in-scope SKUs.
Any experiment or holdout result, with its design documented.
Decision checklist
Write the channel job contract and get finance to sign the CAC ceiling.
Run the seven readiness gates and record which are inadequate.
Fix the smallest blocking input before launching anything.
Choose one channel and one role for the first interpretable test.
Set duration and decision rules in advance, sized to conversion lag and return window.
Launch, and change nothing that breaks interpretability mid-test.
Rebuild results from order-level data, not platform reports.
Apply contribution margin and new-customer filters.
Assign Keep, Fix, Constrain, or Stop to the role, and record the evidence.
Add the second channel only with a distinct role and a measurement plan.
Re-derive the CAC ceiling whenever costs, prices, or return rates move materially.
Where this fits in a growth program
This is channel-selection governance, not media execution. It sits above the account-level work for e-commerce and DTC brands and feeds the testing and budget decisions inside a growth and paid acquisition program.
Talk it through with us
This conversation suits DTC and e-commerce teams choosing where the next channel test belongs, or reallocating between Google and Meta while platform reports, new-customer mix, returns, and contribution economics refuse to reconcile. It is not a channel audit and not a pitch.
In a scoped diagnostic discussion we examine four things: the channel job contract behind the decision, the readiness of the channel-specific inputs, how the two platforms' measurement differs in your setup, and the smallest reliable test that could settle the question.
You leave with a scoped diagnostic and a list of the evidence required — what is trustworthy today, what is not, and what would need to be true before a reallocation is defensible. We do not promise lower CAC, higher ROAS, profit, or any future outcome. Those depend on your economics, your data, and decisions you own.