Cross-Device Tracking: Accuracy Versus Reach

  • .
It’s been more than a year since Awin launched the affiliate industry’s first cross-device tracking solution.

Share this

affiliate marketing multi device technology

It’s been more than a year since Awin launched the affiliate industry’s first cross-device tracking solution and became an active part of the cross-device phenomenon. I use that word specifically, because over the past year I have been stunned at how widespread the discussions have become right across the online industry about the virtues and pitfalls of cross-device tracking.

Awin have written before about how we use a deterministic model for cross-device tracking. We chose this type of solution specifically, because we felt it best fitted the needs of affiliate marketing. The industry’s model is based around rewarding publishers for converted sales. This means we need to be confident an affiliate-referred, multi-device sale has taken place before tracking it. Deterministic matching gives us the greatest levels of accuracy because first party data is being used to generate a match between devices.  So the likelihood of mismatching devices is much smaller than with a probabilistic method, where a device match is inferred based on a combination of observed data, like IP addresses, device type or time stamp and data algorithms.

Of course, the deterministic method has a downside. And that’s reach. Less device matches are made with a deterministic model, because it uses normally just one data-point to match a user to their devices. Probabilistic matching can make many more matches, because it uses much more observed, third-party data. To put this in practical terms our cross-device solution has become better over time at tracking sales, because as more advertisers have adopted it we’ve matched more users to more devices, and therefore been able to pick up more multi-device sales. At the time of writing we have collected more than 39 million cross-device user matches, working with 280 advertisers, so scale is definitely on our side. Not being restricted by the walled-garden effect suffered by the likes of Facebook is an added advantage for us. On the other hand, a probabilistic method would have made more device matches and therefore tracked more sales for us on day one. But the matches would have been less accurate with the end result being sales potentially incorrectly credited to affiliates when they had no involvement with the customer’s journey.

At Awin we remain steadfast that a solely deterministic matching solution is the most appropriate way to run cross-device tracking for our industry. Cross-device tracking methodologies are a trade-off between accuracy and reach. In the affiliate industry, we are striving for as much accuracy as possible. If we don’t track a legitimate cross-device conversion because we haven’t matched a user to their devices is better than making an inaccurate assumption that might mean we infer a device match and track a sale when we shouldn’t.

Is the truth out there?

Despite our experiences, I’ve seen a number of commentators extoll the virtues of using probabilistic data, either independently or in tandem with first-party data, to improve the so-called “truth” of cross-device matching. While not wanting to get overly spiritual here, the definition of “truth” when in comes to online tracking needs a bit more context. Every tracking solution is essentially providing its own version of the “truth”. Just like for humans, “truth” is very much determined by your point-of-view! No two tracking solutions – whether single or cross-device – are ever the same, and no solutions can claim to be accurate 100% of the time. They don’t all tell the same “truth”, but that doesn’t mean they are all telling lies either! Differences in methodology, use of cookies, cookie-duration, what is an acceptable interaction, how conversions are measured are just some of the factors that influence how online tracking technologies differ. Cross-device solutions are no different.

The chances of inaccurate matches in deterministic solutions like the one we use, or those used by Facebook and Google, are often overplayed. As I said above, the concept of total “truth” in online tracking just doesn’t exist. But the likelihood of incorrect device matches being made off deterministic, first-party data like a user-entered email address is slim. Somebody using your email address to log into your profile with a particular retailer on a device that doesn’t belong to you just isn’t going to happen often enough to cause device mismatches. We can vouch for this because in more than a year of deterministic matching across some of the UK’s biggest retailers we’ve had very few issues with incorrect device matches.

Accuracy v Reach

For some online marketing disciplines, reach is always going to be the more desirable side of the trade-off when it comes to cross-device matching. Prospecting and pre-targeting in display is a good example. This type of marketing needs as many relevant eyeballs as possible. Building user matches slowly and accurately isn’t as desirable as making as many matches as possible, and accepting the level of inaccuracy that might be introduced by using probabilistic data and assumptive device matches. The result of an incorrect match might be an ad shown to an irrelevant user. Much less damaging than tracking an incorrect conversion for a cashback site, and awarding cash to the wrong end user.

It’s not a case of who does cross-device tracking the best or most accurately. The best methodology for us at Awin isn’t necessarily going to suit other online marketing channels. Despite deterministic data having its flaws, it remains the most definitive way of us recording cross-device conversions. I’m generally quite surprised anybody working in affiliate considers probabilistic matching to be appropriate given the issues it might cause with our business model, yet I can understand why its important for other marketing disciplines.

Is there a need for standardisation in cross-device tracking?

Unfortunately, the reality outlined above has a downside, and that is the proliferation of fundamentally different cross-device methodologies now on the market. There are virtually no standard ways of cross-device matching. Which means when an advertiser invariably compares data from one cross-device solution to another, which they are now doing, the data isn’t going to match. Those familiar with online tracking in the single-device world will know that this isn’t a new problem. But if the tracking solutions are now based on different fundamentals of user matching the data isn’t even going to be close. Unless there is better understanding of the different approaches to cross-device tracking and why they are used, this issue is going to create a level of mistrust that might undermine its use. It’s beyond doubt that online tracking solutions now need to be cross-device capable to be relevant. Standardisation in cross-device tracking might be the future, but for now, understanding the balance between accuracy and reach, and the needs of different solutions, is an important first step to appreciating the “truth”.