In the article we look at:

  • Conversion paths
  • Why attributing a source and channel to a conversion is so difficult
  • Why attribution matters
  • How GA4 attributes source and channel to conversions
  • Activating Google Signals
  • Options on attribution models
  • What attribution model to use in your business


Conversions paths and attributing a source and channel to conversions on your site

Google Analytics (or other tools such as a customer data platform) can track information about each visit/session on your site and assign each visit to a channel.

What is more complicated, however, is attributing a conversion on your site to a channel.

The user who converted may have visited your site multiple times in the preceding days, weeks and months before converting – and used multiple channels across those visits.

Because Google’s global site tag or Google Analytics tag assigns a unique clientID to each visitor (read more about tracking) then it is possible to see the unique conversion paths (sequence of visits by channel).

Here are some example conversion paths:

Example conversion paths in GA

Say a user initially came to the site via Paid social, but on the visit that led to the purchase the user came through Organic search, then which channel do you attribute that purchase to?

The situation is more complex still

There are other considerations that make attribution even harder:

  • Multiple devices/browsers: The conversion path tracks users on the same device and browser (as the clientID is based on the combination of browser and device). In real life people may use multiple devices and these conversion paths may not be linked (so GA sees two separate users rather than one user on two devices/browsers).
  • View-throughs: A user may see an ad, not click on it but go to your website directly or through another channel (this is called a view-through conversion).
  • Data Privacy restrictions: A user on most iOS Apps will not have their click tracked (read more about impact of Apple’s App Tracking Transparency Framework). And some users may not agree to allowing cookies to record information about their visit.


Why is attribution important?

You want to know what return you are getting from your marketing investment.

Imagine you run a paid campaign that costs £100 and on a last (non-direct) click attribution model it delivers 5 orders (a CPO of £20). You stop the campaign as you need the CPO to be less than £15.

However – does the last click attribution fairly capture the true incremental orders from running that campaign? You’d need a parallel universe as an experiment to see the perfect overall increase in orders from running vs not-running that campaign.

Let’s say that GA4’s data-driven attribution model (read more below) attributes 10 orders to the campaign by looking at the vaue of touchpoints across the conversion journey – suddenly the CPO is £10 and the campaign is a success.

Both these are models and not perfect – but getting closer to the truth can help you make better decisions on where to spend more marketing and where to reduce spend.

This issue of understanding incremental orders from running a campaign is a particular issue for prospecting campaigns where view-throughs may be significant for conversions but hard to track the impact as there is no click or visit to your site. These people might then click on a remarketing ad but it was the prospecting ad that did a lot of the work – how do you quantify the impact of the campaigns at the top of the funnel (the best you can do is hold-out campaigns – comparing the behaviour of an audience that saw your ad versus one that didn’t).


How Google Analytics Attributes a channel to a conversion

GA4 (the lastest Google Analytics product) uses a ‘Cross-channel data-driven model’ attribution model by default.

This is a change from GA (Universal Analytics) – the previous product that will sunset in July 2023 – which used last non-direct click as the default attribution model.

You can change to a different attribution model in GA4 by going to Admin > Attribution Settings

The options are (as defined by Google):

  • Data-driven: Data-driven attribution distributes credit for the conversion based on data for each conversion event. It’s different from the other models because it uses your account’s data to calculate the actual contribution of each click interaction.
  • Cross-channel last click: Ignores direct traffic and attributes 100% of the conversion value to the last channel that the customer clicked through (or engaged view through for YouTube) before converting.
  • Cross-channel first click: Gives all credit for the conversion to the first channel that a customer clicked (or engaged view through for YouTube) before converting.
  • Cross-channel linear: Distributes the credit for the conversion equally across all the channels a customer clicked (or engaged view through for YouTube) before converting.
  • Cross-channel position-based: Attributes 40% credit to the first and last interaction, and the remaining 20% credit is distributed evenly to the middle interactions.
  • Cross-channel time decay: Gives more credit to the touchpoints that happened closer in time to the conversion. Credit is distributed using a 7-day half-life. In other words, a click 8 days before a conversion gets half as much credit as a click 1 day before a conversion.
  • Ads-preferred last click: Attributes 100% of the conversion value to the last Google Ads channel that the customer clicked through before converting. If there is no Google Ads click in the path, as in Example 6, the attribution model falls back to Cross-channel last click.


Cross-channel data-driven model

The data driven model does two smart things to try and improve attribution:

  1. uses machine learning on both conversion and non-conversion paths to understand better how touchpoints and timings affect the probability of converting.
  2. Compares the conversion probability of users who were exposed to the ad, to the conversion probability of similar users in a holdback group (this can help model the view-through impact of ads on Google-owned platforms).

Note that the data-driven model was launched on November 1st, 2021. If you select a date range that includes a timeframe before November 1st 2021 for a model, you will see partial data.


Google Signals

Google can also link paths for users that are signed into their Google accounts on multiple browsers and have ad personalization turned on.

This allows Google to join conversion paths for a user as well as use this data to infer cross device/browser behaviour of people not signed in to Google accounts.

[Note you need to activate Google Signals in GA4 to have this data incorporated into the model. Go to Admin > Data Settings > Data Collection]

Google doesn’t have visibility of people who, say, saw an ad on Facebook and then later went to your site directly.


Look back windows

A look back window dertermines how far back a touchpoint can be to be considered for credit towards a conversion.

The default in GA4 is 30 days for new visitors and 90 days for other conversions.

You can change the look back window in Admin > Attribution Settings


Comparing attribution models in Google Analytics

You can compare attribution models in GA4 by going to Advertising > model comparison:

Image showing attribution model comparison in GA4 (data-driven vs last click)

This is particularly useful for looking at individual campaigns.

Prospecting campaigns will typically be undervalued by a last click model and remarketing campaigns over-valued. This gives a good sense of the balance.

You can aslo view conversion paths:

Screenshot showing conversion paths in GA4


What are the options for managing attribution?

It’s useful to be able to attribute a source/channel to an individual conversion – it provides greater flexibility in segmenting and aggregating data about conversions (revenue, CM2, marketing spend, CM3 etc.)

last click

This is the most simple model – it simply attributes the source/channel of the conversion session. This is the default in e-commerce platforms such as Shopify. It has the benefit of each conversion mapped to a specific source/channel.

Last (non-direct) click:

This model attributes the last click unless the last click was direct in which case it attributes the most recent session that wasn’t direct. The thinking being that there must have been a nudge prior to a direct visit to make a user aware of your site.

If you change GA4’s attribution to last (non-direct) click then you can access this data at the order level via a third party tool such as Supermetrics or via a table in Data Studio (though this cannot be scheduled).

If you have Google Signals activated then, for users logged into their Google accounts across multiple browsers/devices, GA4 should be able to combine user journeys – this makes it more likely to find a prior source/channel if the conversion session was direct or finding a more recent non-direct session on a different device to the conversion.

Last (non-direct) click via customer data platform:

A Customer Data Platform can track each session and event on your site – it can give you more data for understanding each individual user journey.

You can use this data as a source for attributing a last (non-direct) click to conversions. It should map to the Google Analytics data pretty well. One bonus is that if your site has a customer login option then you can join paths of users that are logged in across multiple devices. Just like in the case of Google Signals above, this increases the chance of a finding a prior non-direct visit for direct conversions.

The data can also easily feed into a data warehouse. Because you are capturing data on all sessions and events on the site you can at any stage easily adapt or change your attribution model as you have the raw data saved to apply the model to.

First Click

This model attributes all conversions to the first recorded session for that user on your site. You might choose this model if your customers have a long decision making process. You need to set this as your defualt attribution on GA4 or use a customer data platform to collect this first click data.

GA4 cross-channel data driven model (or other split attribution models)

As discussed above, these models give credit to potentially multiple steps within the conversion path.

If your GA4 account has cross-channel data-driven as the default attribution model then you can create a report in Data Studio with the dimensions and metrics:

  • Transaction ID
  • Date
  • Medium
  • Source
  • Campaign
  • Total Revenue (metric)
  • Transactions (metric)

and see how the data-driven model attributes credit. If a particular conversion has two or more touchpoints that contributed to the conversion then GA4 splits the revenue pro-rata to how much each touchpoint drove the conversion.

This model is probably the more accurate than last-click or first click (or any of the more simplistic models) as it leverages powerful conversion path mapping and comparisons with hold-out groups. The issue here is that a single conversion can have two or more channels/sources/campaigns attributed to it.

It’s a little more complicated but you can use a tool like mapflo to cleverly join this attribution data to your conversion revenue and margin data.

Imagine you have an order with £100 revenue and £50 margin. The data-driven model attributes this order 50% to Paid search and 50% to Organic. We can use mapflo to create two rows for this order, each row worth 0.5 conversions and £50 revenue and £25 margin. One row has a source of Paid search and the other Organic.

Add an uplift

The final option is to apply an uplift to one of the attribution models above if you think particular campaign or channel is under-represented.

We recommend that you compare conversions from your core attribution model to conversions claimed by the advertising platforms (this is something that mapflo can help with). Below is a high level channel comparison but you can also compare at the campaign or even ad set level:

Chart showing comparison of GA4 data-driven orders versus ad platform orders

This shows a comparison of Paid search, Paid social and Shopping orders between GA4 data-driven model and the ad platforms’ own conversion data.

In this example, Paid search looks like a particularly good channel as it has a higher % New; high match between total orders on a data-driven model and what the ad platform is counting plus a low cost per order on both a GA4 basis and ad platform basis.

This allows you to identify channels or campaigns that might be delivering more conversions than the data-driven attribution suggests. If you think data-driven model is undercounts say Facebook conversions by 25% then you can add an uplift of 25% to your data-driven conversions and use this uplifted figure to measure your actual CAC versus your Target CAC.

mapflo is a great tool for joining and comparing data from different sources – give it a try for setting up valuable reports like this.



>>> Read our step-by-step guide to optimizing Google Ads


The interactive video below highlights some of the analyses we cover:


Please be really careful making changes to your Google Ads account

  • Google doesn’t always respond how you (or we) think it will. The way we think about Google Ads may not be the best set-up for your account.
  • Only change one thing at a time.
  • If possible, always use an experiment to test a change – particularly for significant changes such as moving bidding strategy to Maximize conversion value (Target ROAS).
  • Protect your financial downside by testing with limited spend in the experiment/change. Note that moving to a smart bidding strategy requires a learning phase where Google may not be efficient.
  • Be careful if adding/removing primary conversion actions – changing what Google is converting to can radically change what and who Google targets and how much it’s willing to spend.
  • Remember, all changes to your account are at your own risk. Mapflo shall not be liable for any damages; losses; lost revenue or lost profit.


Glossary of Terms

AOV = Average Order Value

CM1 = Contribution Margin 1 = revenue minus COGS (cost of goods sold) in an order.

CM2 = Contribution Margin 2 = margin on an order after all costs directly attributable to that order such as COGS, shipping, payment fees, customer service etc. (except for marketing).

CM3 = Contribution Margin 3 = CM2 less marketing spend. An ‘Estimated CM3’ value uses an assumed CM2 %.

CPA = Cost Per Action. In this report taken to mean cost per conversion or cost per order.

Keywords = words or phrases (assigned to an ad group) that match a user’s search term and trigger Google to bid to show an ad.

Lifetime CM3 = CM3 from all orders (or subscription payments) for a customer.

Profit = CM3 less all fixed overheads (such as salaries and office rent). Hence Optimising CM3 also optimises profit at the same cost base

ROAS = ‘Return On Ad Spend’ = conversion value divided by cost. A ROAS of 400% means you get four pounds of revenue back for every pound of ad spend.

Search term = the word or phrase that a user searches for on Google.