Attribution models in Google Analytics: moving past last click

Most people reading the Acquisition reports do not realize they are looking at an attribution model. The default for everything outside Multi-Channel Funnels is last non-direct click: a conversion is credited to the last channel before the conversion that was not direct, looking back up to six months. A customer who clicked a Facebook ad in March, returned by typing your URL in April and bought in April is credited to Facebook. That is more forgiving than pure last click, and it is why direct looks smaller in your conversion reports than in your sessions reports.

It is still one view of a multi-touch reality. The Path Length report inside Multi-Channel Funnels will show you how multi-touch your own business is. For many ecommerce and lead generation sites, 30 to 50 percent of conversions involve more than one session, and for considered purchases the share is higher. If that number is small for you, attribution modeling will not change much. If it is large, your channel reports are hiding the start of most of your sales.

Before changing models, read the four Multi-Channel Funnels reports under Conversions.

• Assisted Conversions lists each channel with the conversions it assisted, the conversions it closed as last click, and the ratio between them. A ratio well above one marks a channel that introduces customers and rarely closes them. Display, social and non-branded search often sit here. A ratio near zero marks a closer, typically branded search and direct.
• Top Conversion Paths shows the actual sequences. Look for the common patterns, such as paid search followed by email followed by direct, and whether the first touches are channels you are considering cutting.
• Time Lag shows how many days conversions take. If a third of your sales take more than a week, a seven-day lookback window in your ad platforms is undercounting them.
• Path Length shows how many interactions are involved, as above.

Set the lookback window to 90 days for these reports if your sales cycle justifies it; the default is 30. Make sure your goals or ecommerce tracking are set so that the conversions being analyzed are the ones that matter financially, not newsletter sign-ups mixed in with purchases.

The Model Comparison Tool, also under Conversions, lets you put up to three models side by side and see how conversions and value shift between channels.

There is also a last AdWords click model, which exists mainly to reconcile with AdWords reporting.

A practical routine is to compare last non-direct click, first interaction and position based. Export the table and add a column showing the percentage change in conversion value for each channel between the last-click and position-based views. Channels that gain more than 20 to 30 percent are being undervalued by your standard reports. Channels that lose that much are taking credit for work done earlier. Those are the lines in your budget to look at.

The built-in models are generic. The Model Comparison Tool lets you create a custom model from one of them with rules that reflect how your business actually works. Three adjustments cover most needs.

Adjust the lookback window to match your Time Lag report, up to 90 days.

Adjust credit based on engagement. You can reduce credit for touches where the visitor bounced or spent less than a set time on site, which stops a misclick on a display ad from receiving the same credit as a ten-minute visit.

Apply custom credit rules. For example, give branded paid search half credit, since the customer was already looking for you, or give direct visits that occur within a day of another touch reduced credit, since they are usually a return rather than a discovery.

Start from position based or time decay rather than linear, keep the rules few, and write down why each one exists. A custom model that nobody can explain in a budget meeting will be ignored in a budget meeting.

Every model is only as good as the touchpoints it can see. Three gaps distort most accounts.

Untagged campaigns are the biggest. Every email, every social post you control, every QR code and partner link needs campaign parameters, consistently named. Without them those touches arrive as direct or referral and the credit flows to whatever happened to be tagged. Agree a naming convention and build links with the URL builder, not by hand.

Cross-device journeys are the second. Someone who researches on a phone and buys on a laptop appears as two people in two paths. If you have logins, implement the User ID feature so that identified visitors are stitched together; it will not cover everyone, but it will show you how much of the problem you have.

Offline conversions are the third. Phone calls, store visits and sales closed by a rep are invisible unless you bring them in. Call tracking services can post calls as goals. For leads that close later, capture the Analytics client ID in your form and send the eventual sale back through the Measurement Protocol with its value, so the path that produced revenue, not just a form fill, gets the credit.

Attribution reports are a decision tool, not a dashboard. Use them in three specific ways.

First, protect channels that assist. If display or social shows a high assisted-to-last-click ratio and gains value under first interaction, cutting it on last-click numbers will show up later as fewer branded searches, by which time the cause is hard to prove.

Second, set bidding and budget targets on a blended view. Many teams now set cost-per-acquisition targets on a position-based model for prospecting campaigns and on last click for branded and retargeting, which reflects the job each campaign is doing.

Third, test the conclusions. Attribution shows correlation. If a model says a channel is undervalued, increase its budget in one region or period, hold the rest steady, and watch total conversions, not just the model's redistribution. That is the closest most businesses can get to proof without the Data-Driven Attribution model, which is available only in Google Analytics Premium and requires substantial conversion volume to work.

We start with discovery: an audit of goal and ecommerce setup, campaign tagging coverage, and the Multi-Channel Funnels reports as they stand. From that we plan the tagging standard, the lookback and model choices and any offline or cross-device data that needs to be connected. Build covers the tracking fixes, call and CRM integrations, custom models and a reporting view that shows last click and a blended model side by side. Then we improve each quarter, testing the model's conclusions with controlled budget changes. The work connects growth, brand, web and technology: the media plan, the brand searches it generates, the site that converts and the integrations that make the full path visible.

A representative engagement is the performance marketing work we did with a business that had cut its prospecting display and social budgets after a last-click review showed them closing almost nothing. The Assisted Conversions report told a different story, with those channels starting a large share of paths that ended in branded search. We fixed tagging across email and social, connected call tracking and closed sales back into Analytics, built a position-based custom model with reduced credit for branded search, and set channel targets on it. Then we tested the conclusion by restoring prospecting spend in one region. Total conversions there rose while branded search costs fell, which gave the leadership team the evidence that a model alone could not.