What we have seen in ecommerce reporting is this: several channels can claim the same order, and a platform can attribute a conversion that would have happened without the ad. Attribution describes which touchpoint received credit under a rule. Incrementality asks a harder commercial question: what happened because the marketing ran?
This guide explains how UK Shopify brands can begin testing causal lift without pretending every business has the budget or data for a perfect experiment. If your dashboards are precise but budget decisions still feel uncertain, Contact StoreBuilt.
Table of contents
- Keyword decision and research inputs
- Attribution and incrementality are different
- The main ecommerce test designs
- Choose the right outcome
- A practical testing workflow
- Common failure modes
- An anonymous StoreBuilt example
- StoreBuilt point of view
Keyword decision and research inputs
| Decision | Direction |
|---|---|
| Primary keyword | ecommerce incrementality testing |
| Secondary keywords | Shopify incrementality, conversion lift, incremental ROAS, ecommerce attribution UK |
| Search intent | Understand and run tests that estimate the causal value of ecommerce marketing |
| Funnel stage | Middle to bottom |
| Page type | Measurement and experimentation guide |
| Why StoreBuilt can win | StoreBuilt can connect advertising tests to Shopify orders, margins, customer cohorts, CRO, and implementation quality |
Research included current SERP intent, official Google Ads Conversion Lift guidance, current Shopify acquisition and analytics material, UK agency measurement themes including Charle’s growth content, public related-query signals, and a duplicate-risk review against StoreBuilt’s GA4, attribution, paid-landing-page, KPI, and experimentation articles. The content gap is a decision-ready incrementality workflow for ecommerce operators.
Attribution and incrementality are different
Suppose a loyal customer searches for your brand, clicks a paid search ad, and buys. The advertising platform may attribute the sale to the ad. The order is real, and the attribution rule may be working exactly as configured. But would the customer have bought anyway?
Incrementality estimates the difference between an exposed treatment group and a comparable group that did not receive the intervention. Official Google Ads guidance describes Conversion Lift as measuring causal, incremental conversions by comparing treatment and control groups. It also distinguishes incremental conversions from standard attributed conversions.
That distinction matters because ecommerce budgets are allocated on marginal returns. The question is not only which channel appears in the journey. It is whether the next pound of spend creates additional contribution.
| View | Answers | Does not prove |
|---|---|---|
| Last-click attribution | Which eligible touchpoint got final credit? | The sale would not have happened otherwise |
| Multi-touch attribution | How credit is distributed across touchpoints | Causal lift |
| Platform ROAS | Revenue attributed under the platform’s rules | Incremental profit |
| Incrementality test | Difference caused by treatment under test conditions | Permanent results in every season or budget level |
The main ecommerce test designs
User-level holdout
Eligible users are randomly assigned to treatment and control groups. The treatment can see the ads; the control is withheld. This is conceptually strong, but platform eligibility, audience size, privacy thresholds, and campaign types affect availability.
Geo experiment
Comparable regions receive different media treatment. This can work for brands with enough geographic spread and stable regional patterns. UK geography is compact, so spillover, national promotions, PR, and uneven store coverage need careful handling.
Time-based test
A campaign or channel is changed for a period and performance is compared with a baseline. This is easier but weaker because seasonality, payday, weather, competitors, promotions, and stock can explain the difference. Use matched periods and several controls, not a simple week-on-week claim.
Audience or CRM holdout
A randomly selected part of an eligible owned audience does not receive a campaign. This is useful for email, SMS, loyalty, or remarketing tests where the brand controls assignment. Protect consent rules and avoid contamination through overlapping campaigns.
Market or product holdout
Promote selected products or markets while leaving comparable ones untreated. Product substitutability, stock, and differing demand make matching important.
Choose the right outcome
Do not stop at orders or revenue if the commercial decision is about profit.
| Outcome | Formula or definition | Use |
|---|---|---|
| Incremental conversions | Treatment conversions minus estimated control conversions | Causal order/action lift |
| Relative lift | Incremental conversions divided by control conversions | Scale of change versus baseline |
| Incremental CPA | Test spend divided by incremental conversions | Cost of a net-new conversion |
| Incremental revenue | Revenue difference attributable to treatment | Top-line lift |
| Incremental contribution | Incremental revenue less relevant variable costs | Better budget decision |
| New-customer lift | Incremental first-time buyers | Acquisition quality |
Include refunds, cancellations, and returns when the category needs time to mature. A campaign that drives low-quality orders can look strong at day seven and weak at day 45. For repeat-purchase categories, connect the test to ecommerce LTV:CAC cohort analysis without waiting years to make every decision.
A practical testing workflow
- Write the decision. Example: should we increase non-brand paid social spend for UK new-customer acquisition?
- State the hypothesis. Define the expected causal outcome and why.
- Choose one primary metric. Use guardrails for margin, returns, branded search, and existing-customer share.
- Check feasibility. Estimate baseline conversions, detectable lift, test duration, and platform eligibility. A test without enough signal can create expensive ambiguity.
- Define treatment and control. Prevent avoidable audience or geographic contamination.
- Freeze disruptive changes. Record promotions, stockouts, price changes, site releases, PR, and other media.
- Validate Shopify data. Check order source, customer status, discounts, cancellations, tax, market, and refunds.
- Run for the planned period. Do not stop because an early graph looks favourable.
- Read uncertainty. Confidence intervals and practical significance matter; a point estimate is not certainty.
- Choose an action. Scale, maintain, redesign, or retest—and record what would invalidate the result later.
StoreBuilt’s CRO and UX optimisation service can help separate media quality from landing-page and storefront friction.
Common failure modes
Testing too many changes
If creative, audience, offer, landing page, and budget all change, the combined programme may show lift but the team will not know which element earned it. That can be acceptable for a package decision, but name it honestly.
Ignoring brand demand
Branded search and remarketing often harvest existing intent. They can still be valuable, but high attributed ROAS is not proof of equal incremental lift.
Underpowered tests
Small brands may not have enough conversions for platform lift studies. Use larger interventions, longer windows where appropriate, controlled CRM tests, or directional geo/time evidence with explicit limitations.
Using revenue instead of contribution
Discounts, returns, product mix, and fulfilment can reverse a revenue win. Connect the result to the economics the business actually keeps.
Treating one result as permanent
Incrementality changes with budget, creative, audience saturation, season, competition, and brand awareness. Build a testing calendar rather than a one-time certificate.
An anonymous StoreBuilt example
In one measurement review, a brand saw strong platform ROAS from a campaign that concentrated on people already close to purchase. The campaign may have improved conversion timing, but the report could not show how many orders were net new.
The recommended next step was not to switch the campaign off. It was to define a controlled holdout, separate new and existing customers, and compare contribution after refunds. This turned an argument about dashboard ownership into a testable budget question.
If your team needs a cleaner measurement plan before changing spend, Contact StoreBuilt.
StoreBuilt point of view
Attribution is useful for navigation; incrementality is better for investment decisions. StoreBuilt’s view is that UK ecommerce teams should keep both, but never let a platform’s attributed revenue answer a causal question it was not designed to prove.
Start with one meaningful decision, a credible control, a commercial outcome, and a pre-agreed action. The purpose of a test is not to make the dashboard look scientific. It is to reduce the chance of spending the next pound on demand that already existed.
For a storefront, analytics, and growth-readiness review, request a free Shopify audit.