Which Attribution Model Should You Use?

What Are the Different Types of Attribution Models?

An attribution model is the rule that splits credit for a sale or lead among the touchpoints that came before it, and the six usually compared are last click, first click, linear, time decay, position-based and data-driven. The first five are fixed rules anyone can explain, while data-driven uses the account's own conversion paths to estimate how much each interaction contributed.

The six attribution models usually compared

Every one of these models divides credit among the touchpoints the tracking system recorded, so changing the model changes how the credit is split, never which touchpoints are counted. The Channel Selection step covers the source of truth that has to sit underneath whichever model you pick.

Which Attribution Model Should You Use in Google Ads?

In Google Ads the choice is now between data-driven and last click, since Google no longer supports first click, linear, time decay or position-based, and data-driven is the default for most conversion actions.¹ Google stopped letting advertisers select the rule-based models in June 2023 and switched the conversion actions still using them to data-driven from September 2023.²

Google Analytics 4 made the same cut, removing first click, linear, time decay and position-based as of November 2023, and now offers data-driven, paid and organic last click, and Google paid channels last click.³ So any comparison that still recommends one of those four is describing a setting neither Google product offers anymore.

Inside Google Ads, we'd use data-driven when the account has the volume for it, and Google recommends at least 200 conversions and 2,000 ad interactions in supported networks within a 30-day period, though the model still runs on less.⁴ Last click is the plainer choice for accounts with few conversions or a long gap between click and sale, because it's easy to explain and doesn't shift as the model retrains, and how to mitigate last-click attribution covers correcting for what it misses.

Two things matter more than which of the two you pick:

• Data-driven attribution in Google Ads looks at interactions on Google's own ad networks, which are Search (including Shopping), YouTube, Display and Demand Gen,⁴ so it decides which Google campaign gets the credit and has no view of LinkedIn, trade shows or referrals at all.
• What you count as a conversion shapes the bidding more than the model does. In our own campaigns and in client accounts, counting low-value actions taught the platform to find people who looked and didn't buy, while counting qualified leads instead of every form fill changed who it found.

How Do You Choose an Attribution Model for Your Business?

Choose the model by naming the decision it has to support, because a model that's right for daily bidding inside one platform is the wrong tool for deciding next quarter's budget across channels. Work through it in this order.

1. Write down the decision. Steering bids inside one platform, comparing campaigns within it, splitting budget across channels, and proving a channel causes sales are four different questions.
2. For bidding inside a platform, use its data-driven model if you have the volume and last click if you don't, and point it at qualified leads or sales rather than every form fill.
3. For comparing campaigns inside a platform, keep one model fixed, so this quarter's numbers compare like for like with last quarter's.
4. For splitting budget across channels, leave the platforms' models out of it. Record the source of every lead in your CRM from tagged links and hidden form fields, ask every new customer "How did you hear about us?", and compare channels on signed customers and the gross profit they bring.
5. For proving cause, test it. Pause or hold back a channel in some regions or periods and compare results, once spend is large enough to justify the test.

If you're a smaller business, steps 2 and 4 are most of the work, and a spreadsheet with a source column and the "How did you hear about us?" answer will do until a CRM is worth the cost.

Can Switching Attribution Models Fix Platform Over-Counting?

Switching models can't fix over-counting, because every model inside a platform divides credit among the interactions that platform recorded, so a channel it never saw gets nothing under any of them. Only a record kept outside the platforms can give that channel its share.

Take a commercial HVAC contractor with about forty million dollars in annual revenue, which sells maintenance agreements to property management firms through Google search, LinkedIn ads aimed at property and facilities managers, and a local building owners' association.

The correction came from asking the buyer and recording the answer in a system the contractor controls, which no change of model could have done. The full quarter, including how far the platforms' combined claims ran past the agreements signed, is on How Do You Measure Digital Marketing ROI?, and What Is a View-Through Conversion? covers the credit platforms give for ads people saw but never clicked.

When Is Multi-Touch or Data-Driven Attribution Worth It?

Multi-touch attribution earns its keep when most of the buying path happens on channels you can track, such as online sales with short cycles and plenty of conversions, and it helps least when deals close offline over months through referrals, events and phone calls. In the second case a third-party multi-touch tool will still produce confident-looking percentages built on the same recorded touchpoints, which what is a multi-touch attribution model explains along with what building one takes.

Even a good model shows the path buyers took, which isn't the same as showing what caused the sale, since some of the people who clicked your branded search ad would have found you anyway. The extra results a channel caused (its incrementality) only show up when you compare people who saw your marketing with similar people who didn't, and how to measure marketing effectiveness covers the tests that make that comparison.

Self-reported answers have limits too, because buyers don't always remember accurately what influenced them, so they belong beside tracking data rather than in place of it.