Options for fixing broken e-commerce attribution compared (2026): what each one can fix.
- “what are the best options for fixing broken e-commerce attribution”
- “how to fix ecommerce attribution options”
- “attribution tool vs MMM vs incrementality testing”
- “do I need an attribution tool or better tracking”
- “will changing the GA4 attribution model fix attribution”
- “does server-side tracking fix attribution”
- “post-purchase survey vs attribution software”
- “multi-touch attribution software vs fixing tracking”
Baris Asa, tracking consultant, London · Last checked: 14 November 2026
If paid sales show in GA4 as Direct, Unassigned or PayPal, the best option for fixing broken e-commerce attribution is the measurement itself: only that puts a lost campaign label back on the order. If the labels arrive but GA4, Google Ads and Meta disagree, the useful options are a post-purchase survey, a lift test or, with years of weekly data, a mix model. GA4 settings and platform reports only reshare credit.The attribution tracking page lists the fixes by symptom, and the order to work in gives a test for each. This note compares the alternatives people buy or switch on instead, and what each one can fix.
I don’t name or rank attribution tools, and I have no partnership, affiliate or commission arrangement with any of them. The only measurement products named are Google’s and Meta’s own, including the two open-source mix models they publish.
Is your attribution broken, or just disputed?
Two problems share the name, and the options split along the same line.
- The label is lost. A paid sale lands under Direct, Unassigned, your own checkout domain or PayPal, because the campaign tag or click ID never reached GA4. That is a fault with a cause you can test.
- The credit is disputed. The label arrived, but GA4, Google Ads and Meta each give the sale to a different channel, or all three claim it. That is how each one is built.
A two-minute check:
- In GA4, open Reports, Acquisition, Traffic acquisition (under Life cycle in some properties), last 28 days. Leave the primary channel group as the dimension, then switch to Session source / medium.
- Direct or Unassigned growing while paid spend grows, a step up on the date of a site or checkout change, or your own domain, PayPal, Klarna or Stripe in the list? That is problem 1.
- If not, add up one closed week of purchases in GA4, Google Ads and Meta next to the store’s web orders. Platforms claiming more than the store sold, by a steady amount, is problem 2. How to read those totals: three different purchase totals. For problem 2, skip to the survey, lift test and mix model sections.
Problem 1 is the part I fix. A screenshot of that report to [email protected] is enough to start.
Which option answers which question?
Rows run in the order to consider them.
| Option | Answers | Can’t fix | Needs |
|---|---|---|---|
| Fix the measurement | Did the label reach the order? | Disputed credit | Developer time, test orders |
| Server-side tracking | Did the sale reach Meta, Google? | A tag stripped before landing | A server, someone to run it |
| GA4 settings | How is credit shared? | A lost label | Marketer role |
| Platform reports | What each platform claims | Overlap | Nothing new |
| Attribution software (MTA) | One cross-channel view | A lost label | A subscription, a weekly reader |
| Post-purchase survey | What buyers remember | What they don’t | One checkout question |
| Lift test | What one channel added | Where one order came from | Sales volume, a held-back group |
| Mix model | What each channel added | Campaign detail | Years of weekly data |
Does fixing the measurement fix attribution?
Yes, for problem 1. Click IDs are the codes an ad adds to its link (gclid for Google, fbclid for Meta); campaign tags are the utm_ parameters.
What it does. It closes the gaps between the ad and the order. Symptom first, fix in brackets:
- Ad links pass through a redirect that drops the tag (keep everything after the
?in the address on every redirect; direct traffic jumped after a redesign) F-204 - The checkout or booking step is on another domain (cross-domain measurement, so one visit stays one visit; cross-domain tracking not working) F-204
- PayPal, Klarna or Stripe gets the sale (GA4’s unwanted referrals list; payment gateway showing as referral) F-205
- Facebook ads show as Organic Social, campaigns as Unassigned (one tagging convention that matches GA4’s channel rules)
- Paid sales drop when visitors decline cookies (Consent Mode v2 wired to the banner, with URL passthrough, which keeps Google’s click ID on links within your site). This recovers part of the gap, not all of it: GA4 only models missing data on properties that meet Google’s modelling thresholds. F-201
Best for: every shop with paid sales filed under the wrong channel. Only this puts the label back on the order; every option below reads what arrived.
The catch: each fix works from the day it goes live. Sessions already recorded keep their old source. Each one is filed with a test order that proves it held.
Does server-side tracking fix attribution?
What it does. Your server, or a tagging server on your own domain, sends purchases to Meta through the Conversions API and to Google, instead of relying only on the browser tag. Meta says the Conversions API is less affected by ad blockers than the Pixel. Sending Meta’s click ID with each purchase (the fbc value, kept from the ad click) helps Meta tie more of those purchases to an ad F-401 .
Best for: a gap to Meta or Google that is still there after the labels are fixed. Whether there is one worth a server is a test, set out in you may not need server-side tracking.
Not for: a tag stripped by a redirect or a visit split across domains. The server only forwards what reached the site, and it should still follow the visitor’s consent choice.
The catch: two senders for one purchase need deduplication, which tells the platform the two reports are the same sale. Meta needs the same event_name and event_id from both; Google drops a repeated transaction_id. Someone also has to run the server. More on the setup: server-side tracking.
Will changing GA4’s attribution settings fix it?
No. A model shares credit between visits GA4 recorded. It can’t give a sale back to a campaign whose tag never arrived.
The model only changes reports like Model comparison and Key event paths. Traffic acquisition and User acquisition always use paid and organic last click (Google: scopes). What else to know, from Google’s attribution settings page:
- Three models are left (Google: attribution models): data-driven (the default), paid and organic last click, and Google paid channels last click (the last Google Ads click, falling back to paid and organic last click). Linear, first click, time decay and position-based were removed in November 2023, so guides that still recommend them are out of date.
- Lookback window (how far back GA4 looks for a touch): first visits 30 days by default, 7 optional; other key events 90 days by default, 30 or 60 optional.
- A model change applies to historical and future data, so last month’s report moves when someone switches the model. Lookback changes apply only from then on. Write down the date. Changing settings needs an editing role on the property (Marketer or above).
What about the ad platforms’ own attribution?
Each platform credits sales to its own ads. In Google Ads the model is set per conversion action: data-driven is the default for most, last click the alternative, and the Conversions column includes modelled conversions alongside observed ones. Meta credits purchases by the attribution setting on each ad set, which can count clicks, views and engage-through, and it models conversions it can’t observe.
Use these numbers inside each platform, for bidding and comparing campaigns; bidding runs on them, not on GA4’s. Don’t use them to decide between platforms. A buyer who clicked a Google ad and saw a Meta ad is a sale in both, so together they can claim more sales than the store made. Each platform counts its own ads first. That is by design; it just isn’t a neutral referee.
Do you need attribution software?
What it does. Multi-touch attribution (MTA) platforms and Shopify attribution apps collect the same landing URLs, click IDs and orders GA4 does, usually through their own first-party pixel or a store connection, and share credit by their own model across every paid channel in one report.
Best for: shops spending across several paid channels that want one cross-channel view, with someone who will read it every week.
Not for: broken inputs. If a redirect strips the tag, the tool loses it too.
The catch: its model is one more opinion on credit. It shows which touches came before a sale, not whether the ad caused it.
Five checks before you buy one:
- Does it read the same landing URL as GA4? Then a stripped tag is lost to both.
- Does it also send purchases to Meta or Google? Then it is a second sender, and the first must be switched off or deduplicated.
- Which model does it use, and can you see or change it?
- Does its pixel follow your consent banner?
- Can you follow one tagged test order from the ad click to the order inside it?
What can a post-purchase survey tell you?
What it does. One question after checkout: where did you first hear of us?
Best for: channels no click ID records: word of mouth, podcasts, creators, an ad seen on one device and bought on another.
Not for: campaign-level detail. Answers rely on memory, so read them as a pointer, not a record.
The catch: a survey whose biggest answer is “Other” needs better options.
Should you run an incrementality test?
What it does. It holds the ads back from a held-back group (part of the audience or part of the country) and measures the difference in sales. Google Ads’ Conversion Lift splits by user or by geography, and the geography version can use offline sales. Meta’s lift studies randomise accounts into a group that sees the ads and one that doesn’t. Neither is open to every account: Google says Conversion Lift isn’t available for all accounts, and Meta’s runs through a Meta representative. A shop can also run its own regional holdout by switching spend off in chosen regions.
Best for: one high-stakes question. Is this channel adding sales, or claiming sales that would have happened anyway?
Not for: low sales volume. Google notes that what you can read depends on conversion volume and on how the groups are split.
The catch: the held-back group costs the sales the ads would have made there, and the answer covers one channel and one period.
When does marketing mix modelling make sense?
What it does. A marketing mix model (MMM) estimates each channel’s effect on sales from weekly totals: spend, sales, seasonality, promotions. Fed with sales from the store’s back end rather than from analytics, it isn’t touched by cookie consent or ad blockers. Google publishes an open-source MMM, Meridian, and Meta publishes Robyn. Both can take lift test results to calibrate the model.
Best for: splitting a budget across many channels, including ones with no clicks (TV, radio, posters), at a brand with a long, steady history.
Not for: a young or small shop. Meta’s Robyn guide recommends at least two years of weekly data and about ten observations for every variable in the model. It isn’t for day-to-day campaign decisions either.
The catch: it needs an analyst, and the guide itself says to validate the outputs before acting on them.
Which option should you start with?
If the two-minute check found problem 1, fix the measurement first. Then:
- Read GA4 with the right settings, and keep Traffic acquisition as the steady view, because its model doesn’t move.
- Add server-side tracking if a tested gap to Meta or Google remains.
- Use each ad platform’s numbers inside that platform, for bidding.
- Add a one-question survey if word of mouth, creators or podcasts matter to you.
- Run a lift test when one channel takes enough of the budget that you need to know if it adds sales or just claims them.
- Consider a mix model or an attribution tool when you spend across many channels and have the history and someone to read it.
When you need a person rather than a tool
When paid sales land in the wrong channel and nobody can say why, or several tags and tools all send purchases and nobody knows who installed them. That is the measurement part, and it’s mine: find where the label is lost, fix it at source, prove it with a tagged test order. How that work runs is on the attribution tracking page. Or email [email protected] with a screenshot of your Traffic acquisition report. The model, the survey and the lift test stay with you and your marketers.
Common questions
What are the best options for fixing broken e-commerce attribution?
Pick by the question. Fixing the measurement answers whether the campaign label reached the order, and only it puts a lost label back. Server-side tracking answers whether the sale reached Meta and Google. GA4 settings and platform reports reshare credit. Attribution software gives one cross-channel view, a post-purchase survey catches word of mouth, a lift test measures what one channel adds, and a mix model splits budget across channels.
Will changing the attribution model in GA4 fix my attribution?
No. A model shares credit between visits GA4 recorded; it can’t give a sale back to a campaign whose tag never arrived. Traffic acquisition and User acquisition use paid and organic last click whatever you choose. Linear, first click, time decay and position-based no longer exist in GA4: Google removed them in November 2023, so guides that recommend them are out of date.
Does server-side tracking fix attribution?
Only part of it. Sending purchases from your server helps them reach Meta and Google when an ad blocker stops the browser tag. It can’t restore a campaign tag stripped before the landing page, it should still follow consent, and it needs deduplication so one sale isn’t counted twice: the same event_name and event_id from both senders for Meta, the same transaction_id for Google. Fix the labels first, then decide if the remaining gap is worth a server.
Do I need an attribution tool if GA4 is set up properly?
Not to fix broken attribution. An attribution tool reads the same landing URLs, click IDs and orders GA4 does and shares credit by its own model. It earns its place when you spend across several paid channels and someone will read it every week. If the inputs are broken, it reports the same losses in a different layout.
Is an incrementality test or a marketing mix model better than attribution?
They answer a different question. Attribution says which touch a sale is credited to. A lift test says how many sales one channel added, by holding some people or places back. A marketing mix model estimates each channel’s effect from weekly totals. Neither repairs tracking, and neither says where one order came from.
Which option suits a small online shop?
Measurement fixes first, then GA4’s own reports read with the right settings, and a one-question post-purchase survey if word of mouth matters to you. A lift test needs enough sales in the held-back group to show a difference. A mix model needs spend across several channels and a long history; Meta’s Robyn guide recommends at least two years of weekly data. Most small shops have neither yet.
Sources
All read on 14 November 2026.
- Google Analytics Help: attribution models (data-driven, paid and organic last click, Google paid channels last click), attribution settings (lookback windows and defaults, older models removed November 2023, change applies to historical and future data, Marketer role), scopes of traffic-source dimensions (session and user scoped dimensions use paid and organic last click; event scoped follow the selected model, data-driven by default), transaction IDs and duplicate purchases, default channel group, behavioural modelling (eligibility thresholds)
- Google Ads Help: about attribution models (data-driven default for most conversion actions, last click the alternative), about modelled online conversions (modelled and observed together in Conversions), about Conversion Lift (user and geography studies, offline data in geography studies, not available for all accounts, metrics depend on conversion volume and group split)
- Meta Business Help Centre: about attribution models and attribution settings (click, view and engage-through), about Meta’s modelled conversions, Conversions API (less affected by ad blockers than the Pixel), deduplication for Pixel and Conversions API events (matching event name and event ID)
- Meta for Developers: fbp and fbc parameters (recommended with every event; helps attribute more conversions), lift studies (randomised test and control, access currently limited, through a Meta representative)
- Google: Meridian (open-source MMM, experiment results as priors), Meridian on GitHub (geo-level data encouraged, national-level possible)
- Meta: Robyn analyst’s guide to MMM (minimum two years of weekly data, about 1 variable to 10 observations, calibrate with experiments, validate outputs)