Let’s pretend that you’re a marketing manager, in charge of allocating millions of dollars of online display advertising spend. In other words, you’ve got a lot of money to give out, but don’t know which vendor (or combination of vendors) should get the media budget.
Determined to be objective and fair, you reject all martini lunch invitations and inform the vendors that their fate will be decided through brute logic – a third-party attribution system will assign credit where it’s due. Then, you can simply tally up the spends at the end and figure out the true CPA (Cost Per Acquisition) for each one.
Sounds easy, yes?
Actually, yeah. It’s not too difficult – provided all the impression and conversion tagging is set up properly before the whole show begins (we won’t get into that here). As long as you just pick a model and stick to it, you will get to an answer. There will be a ranking of vendors at the end.
But it may not be the right ranking. In fact, it could represent the exact opposite of the actual value that each of your vendors is delivering. How could this be?
Every common attribution model can be easily gamed.
Let’s go through them one by one.
1.) Last Touch Attribution Model
How credit is assigned: the last touch (e.g. last click, impression, or interaction) gets all the credit for the conversion.
How to game it: a whole lot of retargeting. While it’s possible that almost all the visitors to a particular advertiser’s website convert on the first visit (perhaps a time limited form fill offer), most of the time the conversion event gets trigged on subsequent site visits. This is particularly common for e-commerce campaigns: to beat your competition, simply retarget every cookie that has visited the advertiser’s page – keep serving ads to them, because these are the cookies that are more likely to convert.
How to dominate the media plan: this is such a common scenario that most companies have already figured it out. As long as you have any kind of visibility into the advertiser’s site visitors (i.e. you have placed tracking codes on any subset of the website pages), you’re able to play the retargeting game.
So how do you ensure your retargeting is leaner and meaner than the others?
- Increase your bids substantially for the cookies in the retargeting pool
- Increase your bids even more for cookies that have made it farther through the conversion funnel (e.g. have added items to cart, but haven’t converted yet)
- Increase the bids on inventory that is more likely to be visited immediately prior to a conversion (e.g. if the advertiser is a clothing retailer, coupon code aggregators are great places to find that “last touch”
- If the attribution model is not distinguishing between clicks and impressions, don’t even worry about clicks
- Intensify the bidding with each passing day on any retargeted cookie that still hasn’t converted
Ironically, a non-trivial amount of R&D efforts in the advertising industry are being assigned to game this particular attribution model. Much like flared jeans and MySpace, this method of assigning credit belongs in the 2000’s.
Do the other models pose any more of a challenge? Let’s find out.
2.) First Touch Attribution Model
How credit is assigned: first click, impression, or interaction takes it all! If you didn’t serve the first ad, you get nothing.
How to game it: serve one ad to as many people as possible. In other words, stop serving ads to a cookie after just one impression – within the attribution window, of course (so if the window is 30 days, set your system to serve at most 1 ad to each cookie every 30 days).
This one requires a bit more effort than the Last Touch model, seeing as how you need access to a large cookie pool to play this game (if you’re on any major ad exchange, you’re good to go). Don’t worry about bidding too low, as an ad doesn’t even need to be seen by a user for it to count towards the conversion credit – low quality, cheap below-the-fold inventory is your friend here.
How to dominate: if by some chance there’s a competitor on the media plan who has also figured out the above, you’ll need a way to win more overall credit.
First, you could ask the advertiser to lower the agreed-upon CPM rate, thereby giving you more impressions to work with for the same amount of media spend (quality of impressions doesn’t matter too much here). And if your competitors have figured that out too, well, this is where the actual value creation actually begins. You’ll have to do some analytical work to determine which cookies are more likely to eventually convert – and prioritize bids on those.
3.) Linear Attribution Model
How credit is assigned: every touch point (e.g. click or impression) gets an equal share of the conversion credit.
How to game it: while Socialism sounds good on paper, we know by now that there are some serious challenges in practice – even with this model, some impressions are more equal than others. You can win this game one of two ways: bid on cookies that others are undervaluing (i.e. find diamonds in the rough that will convert in the future), or… just retarget like crazy, knowing that most cookies are unlikely to convert on the first visit anyway. Retargeting for the win, once again.
How to dominate: OK, so it doesn’t take a genius to figure out that this model is really flawed. Everyone will be retargeting to get as many impressions in prior to conversion. In other words, it’s a race to the bottom.
To complete our Cold War analogy, the trick here is disappointingly simple: outspend and outbid the competition, regardless of your profit margin. It’s an arms race. Lose money if you have to, just so you can prove that you’re delivering more value in a given month (the hope is that the advertiser will then choose you as the sole ad partner for this channel, at which point you can go back to bidding very little).
4.) Time Decay Attribution Model
How credit is assigned: according to some pre-set parameters, a certain portion of the credit is reserved for touches (e.g. clicks or impressions) that happened just prior to the conversion. A little less credit goes to the touches prior to those, and so on. The parameters can sometimes be toggled – there’s usually a window to set (e.g. 14 days) and a rate of decay (a chance to use “half-life” in the world of advertising!)
How to game it: while this model seems more complex (and is arguably more sensible than the ones above), it’s not difficult to game. In fact, you can safely apply the same techniques as recommended in the Last Touch Model section.
If you’re limited to a certain number of retargeting impressions (i.e. you have to also prospect new cookies), spend as little as possible on the prospecting portion so you have more budget to outbid others on the retargeting. Forget the time-decayed long tail – accrue conversion credit on the touches nearest to the conversion.
5.) Weighted Attribution Model (a.k.a. the “Position Based Attribution Model”)
How credit is assigned: according to (adjustable) parameters, a significant portion of the credit is assigned to the first and last touches. The remainder of the credit is split among the touches in between. So if there were 5 impressions total (as in the diagram above), the middle 3 touches will have to split whatever credit is left over after the first and last are rewarded.
How to game it: this one is easy – it just requires you to pay attention. At the onset, you can try both tactics (from the First Touch Model and Last Touch Model sections) simultaneously. That is, have 50% of your budget dedicated to gaming the First Touch, and the 50% to winning Last Touch credit. Forget trying to collect credit from the middle touches (they’re not worth it).
How to dominate: the first to get a detailed conversion credit report wins. In other words, request reporting from the advertiser as soon as possible – and adjust your strategy accordingly. If you’re spending 50% of your money on the Last Touch strategy but only getting 20% of the total plan credit, reassign more budget to the First Touch technique. Often, vendors will just be going for Last Touch, anyway. Don’t engage in races to the bottom, and you should be fine.
OK, so all that is low hanging fruit. What about the custom attribution models?
Let’s suppose that you come across a completely different attribution model in the wild. This “custom” model could be based on extensive research and testing, or it could simply be based on someone’s creative pivoting in Microsoft Excel.
In any case, it’s highly likely that you’ll be able to game it after just a few rounds of trial and error. Even if you don’t get regular (or any) conversion credit reporting to gauge how well your tactics are working, you may be able to reverse-engineer the attribution logic. Or you may discover that, once again, that the model strongly favours retargeting.
All this cynicism is simply a reflection of the online advertising industry’s state today. Here’s hoping that more conversations happen regarding the flawed nature of common conversion attribution models. And that we ditch the Last Touch Model, at the very least.
Finally, we believe that assigning conversion credit within the same channel (e.g. among display advertising vendors) is premature unless you have first taken the time to carefully and systematically allocate credit among channels. Worry about whether you should even be running this much search advertising. Or display. Or social. That’s an important topic that we’ll save for another post.