Advertising platforms and consultants are always pushing ecommerce merchants to “move beyond Last Click attribution” and use fancy new “data driven” attribution models. For related reasons, they are also pushing us to use longer conversion windows. But they’re rarely explaining the unintended consequences of the changes they’re trying to sell us. If you’re considering changing your attribution model, go into it with both eyes open, and an understanding of how it will distort the figures you’ve been using to make business decisions.
With Last Click attribution, every time a conversion is measured from someone who clicked on an ad, it is fully attributed to that ad click. This is done over a lookback window that has historically defaulted to 30 days. This means that a customer might have clicked on an ad a week ago, but came back to the site today to make a purchase. In this situation, Google Ads would count the conversion in last week’s ad data.
This is the key reason you should never evaluate advertising results based on recent Conversion Value data in advertising platforms like Google Ads. If you look at data from yesterday, you’ll see accurate cost data, but the Conversion Value and Conversions metrics will be far below actual sales since some of those were moved further into the past. Since this can be done over the length of that 30-day conversion window, Conversion Value reports don’t fully stabilize until 30 days later. So, if you pull reports of something like last month’s ROAS in the first week of every month, you’ll always think your performance is going down. But if you pull the same report a month later, it’ll look better.
The two key changes that ad platforms keep pushing on all of us are to change that conversion window to a longer value like 90 days, and to change the attribution model from Last Click to a multi-touch attribution model like Data Driven attribution or a positional attribution model. Let’s look at what each of these changes will do to your reports.
The Effects of Multi-Touch Attribution Models
If you change to a multi-touch attribution model, this has the effect of spreading your conversions out further into the past. If a user viewed or clicked 3 ads before converting, the conversion counts and conversion values will be split across all three of those ads now. Since, under a Last Click model, that would have all been attributed to the most recent of those ad interactions, and now it’s going to be spread across a couple more that are even further back in time, you can see how this is going to have the impact of magnifying the distortion caused by the conversion window.
If you move to multi-touch attribution, you must understand this. Your reports are always going to be skewed far beyond what you saw under Last Click attribution. Recent conversion data just isn’t reliable. Never base any business decisions on recent graphs, or you will make very inappropriate decisions. Don’t bother looking at your Conversion Value for the past few weeks. Making decisions based on longer-term trendlines up to about a month ago is probably relatively fine. But just ignore that last month.
Note this will make it much more difficult to know when you actually have a problem impacting recent sales. And so, what you need to do is keep an eye on data from other sources, such as your ecommerce platform and Google Analytics to understand what the actual sales are for more recent time periods that are coming in from your advertising.
Also remember that since sales data is effectively being spread out across a multi-week period into the past, based on actual ad interactions, this is going to further skew the differences between what your actual accounting system shows as sales for a given period and where those sales are tracked in your ad platform. No matter what time period you are looking at, the Conversion Value data will differ even more wildly from Google Analytics, your ecommerce site, and your accounting system.
Remember, Conversion Value is not an accounting system metric. It is a tool for driving marketing decisions. Don’t use it when what you really want to understand is actual sales for a time period.
The Effects of Longer Conversion Windows
The second change being pushed a lot is expanding your conversion window from 30 days to 90 days. The reason often given is so data driven attribution models and smart bidding algorithms will have more data available to them so they can make even better decisions. If you want to understand more about the biases of automated bidding algorithms and why more data won’t help them make better decisions, check out our article on All the Ways Smart Bidding is Dumb.
If you understand the issues above about how conversion values are spread out over time, understanding the impact of changing to a three-month conversion window should also be easy to get. In a nutshell, there will be more potential ads to spread the credit over, and more of those will be further in the past. So, all of the points above are still true. But the skew in your reporting will be magnified even more. For example, when you view a report of returns in the past month, now you will see much larger drops. And in fact, if you view your data over a longer time period, you’re likely to freak out and think that there’s been a steady trend of declining sales over the past quarter. And this is all just a relic of moving sales back in time over a longer period.
Don’t make bad decisions here. If you move to a 90-day conversion window, you really get no useful Conversion Value graphs in platforms like Google Ads any longer. Give up on that. Go check your recent sales data in other systems.
But moving to longer conversion windows also has another effect. If you change your conversion window, then your ad platform is also going to claim credit for more sales. Let’s say someone viewed your ad on Facebook 80 days ago. Today they came to your website and made a purchase. Facebook is now going to claim credit for that. Perhaps that user called and talked to a salesperson, clicked on a Google ad, and came back to the site directly after clicking some organic search links. None of that matters to Facebook, and they don’t even know about those events.
Changing to a longer conversion window inflates the sales that each ad platform claims credit for, and results in far more instances where multiple ad platforms claim credit for the same sale.
Because of this effect of including more conversions from events that happened further back in time, ad platforms and agencies that manage your ads all have an incentive to convince you to change your conversion windows to longer timeframes. A common tactic is for a new agency taking over an account to convince you that you need a 90-day window for better decision making. But in reality, they just want you to see a sudden bump in Conversions and Conversion Value right after they take over, tricking you into thinking they are dramatically outperforming the previous agency. This is completely unethical, and it is rampant in the ad industry.
The Bottom Line
You could be in an industry with longer sales cycles where a longer conversion window makes a lot of sense. But most ecommerce merchants are not in that situation. You could be working with someone who is a true believer that more data will make the automated systems better. But there are a lot of people who mislead merchants and want to inflate conversion windows to claim credit for more sales when they haven’t actually achieved them. You might be in a situation where a multi-touch attribution model can give you some insights that are useful for your business. But don’t assume that to be true if you can’t plainly articulate a reason why.
Above all, go into any decision about changing conversion windows and attribution models with an understanding of exactly how those changes will skew your metrics. After making these changes, the definition of what a conversion is fundamentally changes. If you continue using these metrics the way you did in the past, you will make poor business decisions. You have to understand what is being measured, how it is being measured, and how reports based on those measurements will change if you don’t want to be misled.