What do you do in this scenario?
Do you give all the credit to keyword 2? If you are using a search engine tracking script or Google Analytics to track your conversions and you are bidding based upon that conversion data, then the answer is yes.
The question is: Should you? Should you change your bid on keyword 1 that didn’t receive a conversion? If the user didn’t click on keyword 1, would they have ever clicked on keyword 2 that led to you receiving a conversion?
This is the confusing realm of assisted conversions and get quickly grow into a full attribution management conversation. Click assisted conversions are clicks that occurred that did not directly result in a conversion; but the user did convert after clicking on a different ad.
You can make this even more complex by throwing multiple channels and impression assisted conversions into the mix; but for our purposes today, we’re going to keep this simple and look at 4 common ways of handing these assisted conversions.
There are many times that doing nothing is just the lazy way out; however, there are times that doing nothing is the correct move.
For instance, in this account if you look at the ratio between assisted clicks and converted clicks – it’s the same for the majority of keywords. There are a few keywords where the assist values are higher or lower than the overall ratio; but for the most part the ratios of assisted clicks and converted clicks is pretty close:
Therefore, this account uses a two labels to identify outliers:
If a word has the high assist value label, the manager knows that the word can be bid a bit higher than last click attribution would allow and that low assist value words have no leeway in their bids.
While this account could build out a full attribution model; the time and energy to do so would far outpace the overall gains and they are better off not worrying about a complex model and spending the extra time testing or doing other account management duties.
This same approach is useful for when the data quickly gets so thin, that you’re working with non-statically relevant data and you’re as likely to find a random pattern as a pattern with actual meaning.
This account is the exact opposite of the previous account. Almost all their converted clicks come from banded terms and most of their assisted clicks are from non-branded terms.
If they were to set their bids by last click conversions; then their non-branded terms drop off page 1 and their branded terms have almost unlimited bids.
However, if they let their non-branded terms drop off of page 1, then their branded traffic drops dramatically and the company makes very little on paid search – they don’t have a strong brand. Essentially, they fit many of the classic molds of:
In this case, you have to give some value for the non-branded traffic. However, they don’t spend a huge amount on paid search, the account is managed by a part-time manager, and to do full attribution funneling isn’t worth their time; however, they can’t just ignore it.
So they use two overall rules for managing the account:
By giving some budget to the assisted click conversion words they are addressing the issue they have in that assisted clicks have a lot of value; but they aren’t going all the way in creating attribution funnels when the data quickly becomes so thin the models would start to break down.
We have another client who spends several million per month on AdWords and the average user clicks on 3.5 ads before converting. They need an attribution model; except a full model becomes an issue as they well over a million keywords that get more than 1,000 impressions/year; but 90% of all keywords get less than 100 conversions per year. Therefore, a full model quickly becomes difficult with so many long tail words. After fiddling with a few models, this is how they handle click assisted conversions:
While this is a simplistic model for a very large advertiser, it works for them – and that’s what matters.
Where we run into trouble is trying to create models for these scenarios:
Now, this could all happen in a day or a month. We could start to assign values to email as there was a conversion (although the email didn’t lead to a conversion in this case) which is a good intermediate step. We can start assigning values to each interaction. We can start to do multi-channel attribution; but the pathways vary a lot and when you get to the point of bidding; the data becomes exceptionally thin.
The first step for larger accounts is to start assigning values and budgets by channel. In some cases, the channel could be mapped to multiple AdWords campaigns (bundling brand, non-brand, etc as individual channels). However, you need to start at the channel level to understand resources and budgets before you get into highly specialized formulas for each keyword.
For some, attribution is very important and needs to be part of the bidding and budgeting strategy.
For others, attribution management is a total waste of time. It might sound good; but in reality, it is just an exercise in futility even if it sounds good to your boss or client.
So before you even start to think about attribution management models, first take a look at your assisted conversions and your top pathways. If your top conversion words are the same as your top assisted conversion words – then there’s no need to start making complicated models as your typical last click bidding is already accounting for the assists.
If the top words in terms of last click and assisted conversions are totally different, then you need to start thinking about attribution management and giving value to your assisted conversions. However, it might not be a full bid model. There may be much more simplistic – and equally effective ways – of managing your assisted clicks without relying on expensive software or complex bidding models to handle your assisted conversion values.