How to Choose the Right Attribution Model

Attribution seems like the deep end of the pool in which no digital marketers want to swim. Mostly because it’s hard to know where to start, and we’ve built our collective lives around analyzing last click data. Luckily, we’ve put together the Digital Marketer’s Guide to Attribution that can help make sense of it all.

If you can muster the courage to choose an attribution model in Google Analytics (besides the default last click), it can reward your business in many ways, including:

  • Better budget allocation for paid media initiatives
  • Better prioritization of content projects
  • Better utilization of house lists, email and social alike

It would help if you began somewhere. Usually, that means moving away from the last click. Literally, any other attribution model tells a more complete story about your marketing efforts than last click attribution, because it takes away credit from the “Direct” channel.

Direct is notorious for swooping in when a customer has already been sold on your product or service by other marketing work and stealing credit for the conversion at the last stage of their journey.

So where do you go from the last click? Depends on what you sell and how long your sales cycle is.

Short on time? Skip ahead to the best models for:

Choosing an Attribution Model in Google Analytics

Instead of going through the models themselves, we’re going to start with your business type and back out to a model.

NOTE: This post assumes knowledge of Google Analytics’ Model Comparison Tool and a basic understanding of digital attribution models. Brush up on those concepts here before diving into our recommended models for each business type.

Subscription Service

If you’re running a subscription service business that depends on recurring revenue and ongoing engagement with your product, you might be one of the only companies fit for a straight-up Linear attribution model.

Think about it: A model that treats all touches equally rewards the first click for driving awareness, middle clicks for driving repeat visits and continued engagement, and last clicks for driving sign-ups after a free trial.

We surveyed the data on subscription service clients and found that, on average, Linear model applications boosted all channels except for Direct and Email (probably because subscription email lists are primarily people who have already subscribed):

Linear models work best in these verticals: Audio and Video Streaming Services, Gaming, Freemium Apps.

Short Sales Cycle Lead Gen

For businesses reliant on fills of a lead form and calls that lead to revenue in less than a month, a Time Decay model with a 30-day look back and 15-day half-life does the trick.

For all lead gen, ramp models are useful because they give increasing credit to educational touches that lead up to a conversion. However, with a tighter look back and half-life, you’ll ensure your model rewards folks that are serious about buying soon and not just poking around for consideration in the future.

We looked at a handful of clients that meet the short sales cycle description and, on average, Time Decay models boosted nearly all channels except Direct and Social.

Time Decay models with short lookback windows work best in these verticals: Catering, HVAC, Home & Garden, Insurance.

Long Sales Cycle Lead Gen

For lead generation businesses where a purchase decision may not happen for several months, and multiple stakeholders are involved in the buying process, a Time Decay model with a 90-day look back and a 30-day half-life is what you need.

Downloading whitepapers, fact sheets, and dodging retargeting ads is hard work for a potential customer, but if they keep coming back to the site and discovering more about your service on each trip, you should handily reward the touches as they move closer to getting that all-important lead form fill.

Similarly, we looked across a list of clients in the tech B2B space, with pretty high-end SaaS products and we found the holy grail. A model that only robs credit from Direct. Every other legitimate marketing effort gets a boost.

Time Decay models with long lookback windows work best in these verticals: SaaS Software, Law Offices, Advertising Agencies.

Impulse Buy E-Commerce

For retail businesses with most products under $100, a Position-Based model with a look back window under 15 days is ideal. It limits the impact of flighty comparison shopping clicks that occur in the middle of the funnel when visitors are double-checking for the best price elsewhere.

After reviewing client data from this sector under the U-curve model, we found small-ticket items benefit from interruption-based marketing like Display ads and email campaigns, especially in situations where your business model thrives on repeat purchases.

Again, only a few losers under this setup, Direct and “Other,” which in this case represents the dying world of Affiliate marketing. Bringing people back to the site at the last second with an Affiliate link no longer pays like it used to and cannibalizes profits for brands that do a great job promoting themselves via other means.

Position-Based models with short look back windows work best in these verticals: Health & Beauty, Apparel, CPG.

Big-Ticket E-Commerce

For retailers with products well over $100, that require more than just a passing thought to buy, and potentially consultation with your significant other, Position-Based models with a 45-day look back window are the way to go.

This model gives even customers who only get paid bi-weekly up to three pay periods to decide if your product is right for them. That’s also enough time to plan a decent vacation.

When we ran this model on several clients with higher end purchases on their site, we found Direct lost big along with Display while organic search and partner channel strategies received a boost.

Position-Based models with long look back windows work best in these verticals: Auto, Hospitality, Travel, Electronics.

You’re already doing attribution

If you’re worried at the data you see when you start to apply an attribution model and are tempted to go back to the last click, or if you encounter naysayers who talk down your model for not including offline marketing, remember this:

You’re already doing attribution if you’re using last click. Just because it’s the default model in Google Analytics and most other platforms doesn’t make it the best for how you should measure your performance.

Pick a new model, or even compare a few models, use them to make some marketing decisions and see if they drive incremental revenue for your business. Show your bosses the money, and you’ll build the trust to invest in more sophisticated tools and strategies.

Start call to action

See how Portent helps companies of all sizes sort out their analytics and report on their data.

End call to action


  1. Great stuff as always – thanks, Michael!

    Question: for the model “Long Sales Cycle Lead Gen”, did you leave turned-off “Adjust credit based on user engagement” and “Apply custom credit rules”?

Comments are closed.

Close search overlay