Attribution Analysis on How to Measure Successfully

Polish e-commerce has become mainstream: 78% of internet users in Poland shop online (Gemius/Przelewy24). More stores and more advertising mean every złoty of budget has to be justified, onsite (UX) and offsite (campaigns). That makes measuring what really drives sales one of the key skills in running an online store.

Which metrics mislead marketers?

Some metrics have always been treated with devotion: agencies are judged on them, case studies are built on them, conferences celebrate them. Many are easy to inflate:

Metric	What it measures	How it misleads
CTR (click-through rate)	Clicks ÷ impressions	Aggressive or clickbait formats raise CTR, but visitors leave immediately
Likes and followers	Reactions to posts	Generic posts collect likes without building any relationship with the brand
Last-session conversion rate	Conversions ÷ sessions, credited to the last visit	Ignores everything that brought the customer to that last visit

Conversion rate matters: it translates directly into money. The problem is how it’s usually analysed, based on the last session. Few people buy on their first visit. Depending on the product, the customer journey takes from a day to several weeks: discovering options, checking specifications, narrowing choices, comparing prices, returning several times from different sources.

Why do ads that “sell” not tell the whole story?

If you kept only the campaigns that show conversions in last-click reports, you’d soon stop selling. Sales work like a funnel: the more new people see your ads, the more can buy later. Cut prospecting because it “doesn’t sell”, focus on remarketing because it converts people who “have already visited”, and the stream of new customers dries up. Remarketing then has nobody left to remarket to.

That’s what attribution is for: assigning credit for a conversion to the visits and sources that took part in it, so you can see which channels and campaigns contributed, and how much.

Which attribution models can you use in 2026?

Model	How it assigns credit	Where available
Data-driven (DDA)	Machine learning compares paths that convert and paths that don’t; credit by each touchpoint’s estimated contribution	GA4 (default), Google Ads (Google)
Last click (paid and organic)	100% to the last non-direct click	GA4
Google paid channels last click	100% to the last Google Ads click; other channels get none	GA4
First click, linear, time decay, position-based	Rule-based splits	Removed from GA4 and Google Ads in 2023 (SEL); can be recreated on raw data
Custom models (Markov, Shapley, own rules)	Your own logic on user-level paths	BigQuery export of GA4 data (GA4 + BigQuery, custom models)

In GA4, compare models in Advertising → Attribution → Model comparison, and see the sequence of sources in Conversion paths. Look at both conversions and conversion value. Note that Google Ads conversions based on GA4 key events use last click in GA4 reports (Google). For path-level analysis beyond the interface, use the BigQuery export (user paths).

Compare stages, not just totals: which campaigns open journeys (organic for generic keywords, prospecting campaigns, YouTube or Demand Gen), which support them, and which close them (remarketing to cart abandoners, abandoned-cart emails, branded search). In 2026, also watch the AI assistant channel in GA4 (SEJ).

How should you tag campaigns for attribution?

Analysis at channel level (Meta Ads, Google Ads, organic, affiliate) is too general. You need at least campaign or ad group level. That’s only possible if every campaign is tagged consistently, so its name carries the information you’ll want to analyse:

Element in the name	Example values
Funnel stage	prospecting, remarketing
Campaign goal	branding, sale, new collection
Optimisation	CPA, CPC, ROAS, reach
Audience	women 25–34, cart abandoners, conversion window 1d / 7d
Device	mobile, desktop
Product group	footwear, TVs, bikes

Example: pros_sale_troas_w25-34_mob_footwear. Use the same convention in utm_campaign, ad platforms and cost files. Then you can compare, say, two remarketing campaigns for the same audience and find that a 1-day conversion window contributes more than a 3-day one, or that prospecting optimised for ROAS beats target CPA.

Why do costs matter as much as conversions?

Effectiveness is how many sales a campaign brings; efficiency is what each sale costs. Two campaigns can each contribute to 100 transactions. If one costs PLN 10,000 and the other PLN 50,000, they’re equally effective, but the first is five times more efficient.

GA4 pulls cost data from linked Google Ads accounts automatically. For other sources (Meta, TikTok, LinkedIn, affiliates), use campaign data import, which joins costs to GA4 data by source, medium and campaign ID (Google). Then compare ROAS (revenue ÷ cost) and CPA within the same attribution model, and decide which campaigns to scale, change or stop.

Also check the balance across the journey: the number of sources, their cost and the revenue credited at the start, middle and end of paths. Large imbalances lead, over time, to a shrinking stream of new customers or to too few touchpoints that close the sale.

What can’t attribution measure?

• Missing data: under consent mode v2 (required in the EEA since March 2024), users who decline cookies are only partly measured and some conversions are modelled (SEJ, cookieless measurement).
• Views without clicks: video, display and social ads that influence people who never click.
• Offline and other channels: TV, outdoor, word of mouth, AI assistants that recommend you without a trackable click.
• Causality: attribution shows who was present on the path, not whether a campaign caused the sale.

That’s why larger advertisers combine attribution with marketing mix modelling (MMM), such as Google’s open-source Meridian, available since January 2025 (Google), and with incrementality tests (geo or holdout tests) that measure what happens when a campaign is switched off.

What are the steps of an attribution analysis?

1. Tag every campaign with a consistent naming convention.
2. Collect data in GA4 (and, ideally, the BigQuery export), with consent mode set up correctly.
3. Assign costs to every campaign: Google Ads automatically, other sources via data import.
4. Compare models (data-driven vs last click) and conversion paths.
5. Evaluate campaigns with ROAS and CPA within the chosen model.
6. Check the balance of sources, costs and revenue across funnel stages.
7. Validate big decisions with MMM or incrementality tests.

Attribution shows not only who buys and why, but how they buy and how much each purchase costs. That’s what lets you invest in channels that actually work, including the ones that lead customers to the cart without closing the sale. Visualise the results so decisions are easy (data visualization).

FAQ

Which attribution model is best?

For most stores using GA4, data-driven, compared regularly with last click. For full control, build your own models on BigQuery data.

Why were first click and linear models removed?

Google removed four rule-based models in 2023, citing low adoption; fewer than 3% of Google Ads web conversions used them. You can still recreate them on raw data.

How do I add Meta Ads costs to GA4?

Use GA4’s campaign (cost) data import, with matching utm_source, utm_medium and utm_id values in your ad links and cost file.

What’s the difference between attribution and MMM?

Attribution assigns credit along user paths it can track; MMM uses aggregated data (spend, sales, seasonality) to estimate each channel’s contribution, including offline and view-based effects.

Sources

• Analytics Help: Advertising and attribution
• Google Ads Help: Rules-based attribution models are going away
• Search Engine Land: Sunset details for 4 attribution models
• Analytics Help: Import cost data
• Google: Meridian is now available to everyone
• Search Engine Journal: Consent mode v2
• Przelewy24/Gemius: E-commerce in Poland 2025