With the upcoming merger of the Microsoft and Yahoo search networks in the UK, we’ve been looking at optimisation techniques on the adCenter network. Wayyyy back in the dawn of time (well, a year or two after AdWords started) we were doing work on optimising ROI, when we spotted an intrinsic economic problem – a conflict of need between what the search engines want and the best interests of the advertiser. The issue is easy to explain.
Advertising networks want to maximise their revenue. The way they do that is to pick the highest CTR, highest paying adverts. This turns out to be quite simple. Forget about pay per click, and instead calculate adverts as if they were CPM. That’s easy to do. If you have an advert that costs £1.00 per click, and has a 10% CTR, then it is showing at a CPM of £100.00 – because a 10% CTR means that you have 10 clicks per hundred impressions or 100 clicks per thousand impressions, and 100 clicks would cost you £100.00. Now rank your advertisers based on that calculation of the value of an advertisement and you have the first simple AdWords auction – it got more complex, but that’s still the basic premise of how it works, but with a “Quality Score” factor instead of just CTR.
However, while that maximises revenue for the advertising system, it can’t guarantee that it has maximised the revenue of the advertiser. The practical example was when we were working on Thomas Cook. If we offered adverts with “Book now…” then we got an increase of 40% more bookings and they mostly happened a week earlier, than if we used the copy “Browse…” or “Search…” instead of “Book now…”. But the CTR was slightly lower on the “Book now…” adverts. Which did Google choose to optimise? And what was the impact?
Yup, Google picked the higher CTR adverts, which de-optimised revenue for our client. So we stopped allowing Google to choose which adverts to optimise. It’s the only safe defence.
So, very early on, we made a policy decision that we don’t allow the advertising networks to choose which adverts they want to show, unless the client purpose didn’t involve a strict ROI target (e.g. a brand awareness advertising exercise). And we now inspect adverts to see which ones have a lower ROI, and lose them – not based on the CTR, but based on sales.
adCenter doesn’t have the option to show adverts equally, on rotation. That means that adCenter will optimise its’ own revenue, at the expense of an advertiser, by removing the advertisers ability to collect a statistically useful sample of results, unless the advertiser goes to daft lengths (e.g. sequential tests rather than A/B or multiple parallel).
If you are interested in optimising for the new adCenter, see if you can register your interest in having Microsoft permit you to have equal rotation, so you can assess which adverts work and which fail. Note that there are other requirements if you do this – if you have multiple keywords per AdGroup, for example, the performance of each keyword needs to be roughly the same, or you may be selecting adverts to display for high volume low conversion rate keywords, when the right driver to select the adverts is a low impression volume high conversion rate keyword. Making web advertising effective is a bit more complex than just equal rotation – the optimising method has to include how many times each advert is showing for each keyword.
This lack of control by the advertiser has been noticed by other adCenter users. I was going to add this comment (or one rather like it) to the adCenter discussion group, until I found out that to do so, I’d have to stop using this computer, and use a Windows machine. I’d rather not, thanks. I have picked this operating system for a lot of reasons – and Microsoft’s tedious obstacles don’t make me more likely to use Windows, it makes me less likely to do so.
I should point out that this observation about optimisation is not unique. In fact there’s quite a lot of other people making this or a very similar point. Here’s some:
- Use Split Testing to Create Killer Ad Copy
- Ad Copy Testing: Systematically Improving CTR – Microsoft’s own blog for advertisers recommends split copy testing
- Is it possible to make ads rotate evenly? – Microsoft adCenter discussion forum with some strong points made by advertisers
This is an issue that Microsoft needs to take seriously, It can have a substantial impact on performance, and the value of the tried and true technique of split testing is significantly reduced when Microsoft chooses which advert to show. It may help Microsoft’s revenue, but not that of advertisers. And offering “just create a new AdGroup because they start without optimisation” – is just not do-able. We tried. Merging the data from multiple attempts so that you capture the period before Microsoft has optimised is a data horrorfest, especially on a large account. And testing like this is yet another management overhead on a smaller fraction of advertising than Google – Microsoft need to reduce the administration overhead, not increase it, relative to AdWords’s cost of operations.
