PPC Day Parting: Some Considerations

Day parting is one of those things that I’m fairly sceptical about. Yes, it does make sense to skew ad budget towards well-converting periods of time, but then isn’t this too much of a simplistic view? I feel that our good old last click model, the usual suspect, can be misleading us into lowering bids during some important initial stages of the path to conversion. So I took some data and had a little play to see if this is true. Read on, if that’s your cup of tea.

Some Sample Data
The graph below can be a little confusing at first, but it simply plots the disparity between first click and last click conversion for different days of the week. If, for example, a campaign converts at 10% on first click, but only 5% on last, then it’d score very highly on the graph and hit the 200% mark. The higher the score, the more “first click heavy” a campaign is. Below the 100% mark, the campaign is last click heavy.

First click bias is happening to Paid Search on Thursdays in particular, and we might be tempted to cut bids. This may or may not be a good idea, but you need the visibility to make a decision. The opposite can also be true, as is the case with affiliates, represented by the blue line. They tend to convert disproportionately well at last click, but are more rarely the initial touchpoint.

Moral Of The Story
Day parting is often based on last click data and my ramblings above highlight its potential inaccuracy. Sometimes customers will research a product mid week, but actually buy it when they have spare time towards the weekend. Similarly with time parting, there is often a strong research phase during lunchtime but conversions occur in the evenings. For most of us, we’d need to have the ads running in both instances of the customer path to ensure a sale.

Ideally, the exercise above should be done on a regular basis and at campaign, not channel, level. Splitting Brand from Generics in PPC makes sense, as I imagine that Generics will experience much higher disparities. As always, the more you dig into the data, the more valuable and actionable it becomes.

When It Comes To Attribution, More Is More

First of all, apologies for the lack of posts over the last month on this blog. I blame Twitter!

Now, we all know that attribution windows have been a big topic over the past year and will be even more so in 2010. And for a reason. Customer paths are complex, lengthy and those users that are more involved with your brand can have touch points across all of your marketing channels. Even if you decide to simplify things and investigate a specialist case like Search it can still be quite messy. Google estimate that a user carries out just under 6 searches on average before they convert. With such a dynamic customer path what attribution models should we use?

Let’s look at some data first. I can’t miss the opportunity to draw a pretty graph, and below is one based on real data for a recent and adequately lengthy time frame. It shows the number of unique keywords that have been credited for a conversion by different attribution models. The four windows are Same Session, 30-day backwards-looking First Click, Last Click, and Average (or Multi-) Click. If we assume that the Multi-Click method has captured 100% of activity then we can work out how many unique keywords the other methods are missing out.

As you can see, the Same Session has captured 73% of all keywords that were used in the conversion path. That means that almost 30% of your keywords won’t receive the credit they’re due and are likely to have their bids reduced. Last Click hasn’t faired much better at 74%, while First Click is at 80%. There is obviously quite a number of keywords in between First and Last Click as Average Click adds the last 20% which is quite a substantial amount. (An important note to make is that the data set used includes all keywords – that’s both brand and generics. There is likely to be some difference in the way these two different sets of keywords work and that is something I will write about soon.)

Back to my initial question – which attribution window should you use to run successful paid search campaigns? I think the answer is all of them. It is important to have visibility of all search touchpoints, but then it’s down to the judgement of your PPC ninja on how to interpret the data and adjust bids. If a certain generic keyword tends to initiate the sales process but never converts as a last click then it’s more than worth investing in.

Not surprisingly Google are aware of the problem of attribution. Their Analytics product works on 30 Day Last Click as standard – although SEOmoz discuss how you can bypass that. Google themselves, however, are rumoured to introduce new metrics such as Assisted Sales and even Touch Point Analysis right into the AdWords interface. Where Google might fall short though is their limited channel reporting since they can only track Paid Search and possibly Natural Search. If you want the full picture – how users move between search, display, email, affiliates, etc – a platform like Tagman can help. This is part of the reason why the clever guys at Boden adopted it recently (read more on Tagman’s blog) although the correct reporting tool is really only the start of the solution to our attribution tribulations.