Attribution Models

Attribution modeling is hard. Actually impossible to get even 80%+ accurate. It’s hard and anybody that tells you different is a charlatan.

Attribution modeling in simple terms is how you divvy up and attribute revenue by channel. It’s imperative to get this as correct as possible so that you can double down on what’s working and stop wasting time and money on what’s not working as well.

Attribution modeling is hard. Actually impossible to get even 80%+ accurate. It’s hard and anybody that tells you different is a charlatan.

Let’s think about all of the different conversion paths that could happen and how each one of your media touch points affect each and every one of those buyers. Now think about a brand that is running 10+ media channel and has millions of users coming to their site each month.

This is compounded by the fact that you’ll be using a multitude of measurement tools that have access to different data and use different attribution models.

I don’t have to imagine the above. I’ve spent countless hours doing media attribution for a site like this and gotten nowhere *accurate* fast.

Lucky for you, your website is likely much smaller. Attribution modeling gets infinitely more difficult as media and users grow.

Below you’ll see 3 conversion paths. We’ll go through each of the conversion paths, the most common attribution methods, and how each of the attribution methods give value to different channels in the conversion paths.

There’s more than the below listed, but this should get you started on how impossible it is to truly capture the *cause* of conversion once your site grows past a certain size.

**Note: This whole post assumes that you don’t have or aren’t utilizing your CRM (Customer Relationship Management) or CDP (Customer Data Platform). This is likely the case until an e-commerce business gets into the high 10’s of millions of revenue per year. CRM’s and CDP’s aren’t cheap!

Attribution Window

The attribution time window is how long after a cookie is set will the attribution model still give credit to the channel. Cookies can last indefinitely. There should be a cutoff period when enough is enough.

Fun fact: Amazon Associates’ attribution window is the shortest that I know of. After an affiliate’s cookie is set, the affiliate only gets credit for a sale that happens within 24 hours. For more, check out Affiliate Marketing Part 1, Part 2, Part 3a, and Part 3b.

Conversion Paths

All conversion paths really means is how did the customer get to the point of buying your product.

Conversion Path #1

Customer #1 hears a radio/tv ad. They then Google your brand and click through on your paid text ad. They don’t purchase that day. 1 week later, they come direct to your website and purchase.

Radio -> SEM -> Direct

The actual cause of this purchase was the radio or tv ad. There’s arguments to be made that your onsite copy and offsite reviews played a part, but to keep it simple, we’re going to stick with the basics of your media presence.

There’s also an argument to be made that if the text ad wasn’t there, they would have clicked on your organic listing. The flip side of that argument is that another brand might swoop some of your customers away by bidding on your “brand terms” space.

Conversion Path #2

Customer #2 gets shown a display ad. They don’t *consciously* notice it since so many people have trained themselves to ignore them. Same thing happens on FB the next day. They then are in a conversation a week later which reminds them that they should buy X *type* of product. (Not the specific brand. The type of product.)

They then Google “best X product”. They land on an affiliate page and click through to your website but don’t buy. One week later, they Google your brand name, click on an *Organic* listing and buy.

Display -> FB -> Affiliate -> Organic

The actual cause of this conversion is much harder to decide. Did the display/FB ads “prime” the buyer and the copy written on the affiliate site push the user over the edge? Or did the the display/FB ads have nothing to do with it at all?

Conversion Path #3

Some anonymous Ox recommends a product on a Substack. Since he’s jacked and has raved about it, you head to Google. You type in the supplement brand name and click onto a paid text ad.

Because you’re trying to save some money, right before purchase, you Google “X brand coupons”. You click on retailmenot.com or some other coupon site. You find a coupon but can’t see it until you click their link. That link opens another tab that you likely don’t notice and “sets” a cookie.

Ox -> SEM -> Affiliate

The actual cause of this sale is Ox. Google Analytics is going to say affiliate and Google Ads is going to take partial credit. These are the most difficult sales to track accurately since we don’t have data on random people’s recommendations that we didn’t pay for. This is where social listening tools come into play.

Last Non-Direct Click

Last non-direct click is the default attribution method for Google Analytics. If you look at the above examples, nothing should be attributed to the “Direct” channel. This should only happen in a few cases that is seemingly getting bigger by the day due to Apple’s privacy standards among other things. You’ll see revenue attributed to the direct channel when;

  • A user clears their cookies from their browser

  • A user goes straight to your website. This is normally an issue due to an ad being seen on one device and then searched on another device. This can be bypassed with cross-device tracking (above the level of this post). Can also be the case that your brand is well known and there’s no reason to go through any channel other than direct.

  • No-follow no-opener links

  • In-app browsers like Reddit, Instagram, etc.

Conversion Path #1

Radio -> SEM -> Direct

In this conversion path, SEM gets 100% of the credit while Radio gets 0%.

Conversion Path #2

Display -> FB -> Affiliate -> Organic

Organic gets 100% of the credit while your other channels get 0%.

Conversion Path #3

Ox -> SEM -> Affiliate

Unfortunately, affiliate (the coupon site) gets credit for this sale while Ox and SEM get 0% of the revenue attributed. Most all affiliate programs pay out based off of last non-direct click. So when driving traffic to your site utilizing affiliate, you need to be very very careful that it’s not cannibalizing sales like in this example.

Coupon sites and browser extensions like Honey are notorious for this. You’ll be paying 10%-20% of your revenue for a sale that would have likely happened anyways. Don’t listen to affiliates and publisher platforms that say “their” coupon caused the conversion since the customer wouldn’t have purchased without a coupon. They’re very biased and either lying to you or themselves.

There’s a reason why Amazon’s affiliate commission and attribution timeframe has steadily decreased as they’ve grown.

First Touch

A first touch attribution model is just like it sounds. The channel that introduced the customer to the website (within the attribution window) gets the credit.

Conversion Path #1

Radio -> SEM -> Direct

Unfortunately, there’s almost no way to track that radio introduced the customer to the brand. SEM gets 100% of the revenue attributed to the channel while radio gets 0%.

Conversion Path #2

Display -> FB -> Affiliate -> Organic

First touch attributes 100% of the revenue to the affiliate site and 0% to everything else. Since the user never interacted with (clicked on) the display or FB ad, they were never “touched”.

Conversion Path #3

Ox -> SEM -> Affiliate

Word of mouth here is completely un-trackable. Also, Ox may have caused the first touchpoint, but he didn’t refer the traffic. SEM gets 100% of the revenue attributed to it’s channel.

Multi-Touch

Multi-Touch attribution is designed for these types of conversion paths where there’s multiple touch points. There’s a variety of different ways you can weight this attribution model. First touch gets most of the credit and every touch after that point gets less and less credit. Or the exact opposite where the more recent the touchpoint, the higher the percentage of revenue attributed.

For the examples below, we’ll use the model that gives equal credit to each touch point.

Conversion Path #1

Radio -> SEM -> Direct

SEM gets 100% of the attributed revenue. Since you can’t track the radio touchpoint, it gets 0%. Direct gets 0% due to not being a real channel that you drive to with media.

Conversion Path #2

Display -> FB -> Affiliate -> Organic

Affiliate and organic split the revenue with 50% each. Since there was no touch (or click) on display or FB, they get 0% of the revenue attributed.

Conversion Path #3

Ox -> SEM -> Affiliate

SEM and affiliate split the revenue with 50% each. Even though Ox “caused” the sale, there’s no touchpoint and no way to track.

View Through

View through attribution is usually tracked via the media partner that you’re running media through so there’s an enormous bias by your media partner. The higher revenue they can attribute to their channel, the higher your “attributed” ROAS, the more some inexperienced media person will spend. This is compounded by the fact that most large companies utilize agencies that are paid a percentage of media spend. They encourage this behavior since it’s good for their bottom line.

This attribution method will be relatively accurate if you don’t have a large well known brand. Once your brand grows to a large size, the ads are being shown and ignored by people that would have already purchased from you without the ad.

I may or may not have a vendetta out for people that claim that view through is anything but a scam at scale.

Essentially your display partner, Google (SEM, display, Youtube), social networks (FB, Instagram, Reddit, etc) is serving the ad and also have a pixel on your website.

This allows them to connect the user from the view of the ad to landing on your site.

Note: There will likely be an upcoming post on ROAS vs attributed ROAS.

Conversion Path #1

Radio -> SEM -> Direct

No channel gets the credit here. SEM would have gotten the credit if the ad was shown within one day of purchase. Google’s default view through attribution period for text ads is one day. For more on Google Ads attribution models, go here. This can be changed to whatever you like after you start a campaign.

Conversion Path #2

Display -> FB -> Affiliate -> Organic

Display gets 100% of the revenue attributed to that ad. FB also gets 100% of revenue attributed to its ad. See the problem here? Both media partners utilize view through attribution and both partners are measuring it with their own tools. So not only can the tools be wildly incorrect, they’re double counting revenue.

One time I looked at the attributed revenue of all of our channels and found that over 300% of all of our revenue was being attributed to media…

Conversion Path #3

Ox -> SEM -> Affiliate

SEM gets the credit from view through attribution since the purchase happened within 24 hours of seeing the ad.

Testing

While this isn’t an “attribution model”, I had to include this. This can be as simple or convoluted as you’d like the setup to be. It takes a lot of time, money, and resources to constantly test your media, but it’s the right way to do it. This will provide you the most accurate impact of your different media efforts on revenue. A few ways to test your media are;

  • Pulse your media. Start and stop your media for 30 days. Run a pre-post analysis and see what happens to sales. Try to keep absolutely everything else constant when doing this. This means don’t increase/decrease other spend, run different promotions, etc.

  • Hold out or heavy up test. Block half of the country’s media spend in a channel and track orders back to geography. Then run a pre-post analysis on the hold out geography and compare it to the “Business as Usual” geography.

  • A/B test. This is going to be a little bit more difficult to do for small websites. At scale, companies can show a display ad to a percentage of users and a different ad (unrelated to the actual company) to similar users and see what the difference in buying habits are. Another example might be to utilize Google Optimize or Optimizely to block a percent of coupon affiliate users from using a coupon. Hence measuring how many of them would have purchased without the affiliate coupon.

Conversion Path #1

Radio -> SEM -> Direct

When running Radio or TV, try running a hold out group that represents the rest of the market. This way, you can measure the difference.

Conversion Path #2

Display -> FB -> Affiliate -> Organic

If you have the scale, A/B test your display and social media marketing. Or you can use the affiliate coupon example above.

Conversion Path #3

Ox -> SEM -> Affiliate

I once ran a large hold out test where we blacked out 25% of the country to our SEM ads. Another 25% of the country, we doubled spend. The other 50% was the control group that we compared each test group to.

Regression Analysis (Data Driven Model)

I won’t spend a ton of time diving into what a regression analysis is. Those interested can Google it and take a deep dive down the statistical rabbit hole. High level, it’s a statistical method measuring the correlation between two (or more) variables. Think about trend lines in excel after you’ve scatter plotted data.

If you have enough data (you likely don’t), you can then run regressions of your media spend vs your total revenue and attributed revenue. This will help you better find correlations (and maybe causations) of what’s working and what’s not.

Note: I’ve found strong correlations between media spend and attributed revenue. That does not mean that it’s a causation. Try running the regression on that media channel’s spend and total revenue - attributed revenue. If you see a negative correlation, that channel is likely over attributing and “stealing sales”

Conversion Path #1

Radio -> SEM -> Direct

With enough data (times this is happening), this conversion path would show a strong correlation and incorrectly attribute the cause of the sale to SEM. This is a well known problem, especially with “brand” spend.

Conversion Path #2

Display -> FB -> Affiliate -> Organic

A regression analysis would find a very strong correlation with this type of conversion path and attribute the sales to the affiliate channel.

Conversion Path #3

Ox -> SEM -> Affiliate

SEM will have a very strong correlation with enough data taking similar paths. Because affiliate marketing is the most performance based subsection of performance marketing, you’re always going to find correlation with this channel.

*Generally* you’re paying affiliates as a percentage of revenue. If you’re running a regression of spend vs attributed revenue, then your correlation coefficient is going to be 1. If you’re running the regression to total sales, it’s a toss up.

Different Platform’s Attribution Models

To complicate this even further, every tool has a different attribution method. Unless you’re only using GA and not the other platforms’ measurement tools, you’ll be forced to measure using different tools and different attribution models.

We’ll go through the basics of how a few platforms measure. You should realize that every platform’s attribution model tries to attribute as much revenue to their channel as humanly possible.

Google Analytics

If you look at every report in GA, you’re seeing last non-direct click attribution for your revenue. Google also provides another tool that lets you compare different models (and look back windows) and how that changes the revenue attribution.

Unfortunately (and fortunately for privacy concerns), Google isn’t measuring view through for all of your channels. This is strictly click based. Location below.

GA -> Conversions -> Multi-Channel Funnels

Google Ads

Google Ads has quite a few attribution models built in that you can change after an ad is built. Most can be some variation of click combined with view through with a changeable view through period. They also have a “data driven model” which is essentially their version of regression modeling on *only* their ads. This will generally be close to the last click method but weighted slightly in their favor. (Higher attributed ROAS leads to companies spending more)

It also has different attribution methods for the different type of ads. Product Listing Ads and the like will by default have longer view through attribution windows.

FB/Instagram/Reddit - Social Media in General

Social media channels generally have a 28 day click and 1 day view through model for their internal reporting. Note that advertising agencies like to turn that view through time period up as high as they can. The get a percentage of media spend after all. Higher attributed revenue -> higher attributed ROAS -> more spend -> more money in their pockets -> higher bonus for the account rep.

Display

Since less and less people click on display ads as time has gone on, companies have increased the view through time period. Since there’s a number of different companies that serve display ads, there’s no one default attribution model that’s used.

There is however a general rule to go by. Display ads will have view through attribution that are longer than other forms of media.

Affiliate Platforms

You control the attribution method here. You define how long the cookie lasts and whether it’s first click or last click attribution.

Solution & Conclusion

I have to be honest here. There’s no easy solution. There’s countless firms that offer Marketing Mix Modeling with 6 figure price tags using multivariate regression models and a few hundred hours of work. Even they are only partially correct. It’s similar to predicting where the stock market is going to go next. If you could crack the code at a large enough scale, you will become a billionaire.

The best thing that you can do with a sub 9 figure brand (without the money to pay for a Marketing Mix Model) is the following.

If you liked this post, please like it, share on social media, or email it to your friends/colleagues. Definitely bookmark it as you’ll want to come back again through your e-commerce journey.


If you want to get smarter about all things digital, upgrade today from free to paid for only $10 a month. The paid posts are where I dive even further into the weeds to help you truly understand and capitalize on the digital landscape.