Do different ways of day parting make a difference?

Posted on January 18, 2016

I was recently fortunate enough to get in a conversation with Mike Roberts of SpyFu on AdWords day parting.

Mr Roberts points out that the database behind AdWords is almost certainly “eventually consistent” which means that a change accepted by the database may not be reflected in all nodes until some time has past. So the Google data centre in the UK may know about your changes but the one in Japan might not. NB: This isn’t just to do with geographical distance; it also can apply within a single data centre.

This would mean that making a change in AdWords will not be reflected in the auction instantly.

I pointed out that “eventually consistent” databases are usually consistent within seconds; an amount of time that is inconsequential for most advertisers.

Roberts is not convinced by my arguments and suggests that 30 minutes is a reasonable timescale for changes to peculate throughout Google’s system - but this is more to do with batching calculations together rather than delays in the underlying database technology.

This reasoning seems plausible to me but I would guess the delay is much lower.

Either way, we are in the situation where Google batches some calculations together and that these are only done at intervals. What does this mean for day parting methods?

Right now, I use a mixture of scripts and the API for running day parting schemes. I’m not doing anything really complicated here and I’d guess that 90% of what I’m doing could be managed in the simple day parting interface in the campaign settings.

When my scripts make bid/status changes, Google doesn’t know what these are going to be until the script runs. Then we need to wait for the next batch processing job to run before the changes are reflected in the auction.

Compare this with using the simple day parting interface; in this case Google knows exactly what changes are going to happen and when the changes need to be in the auction. The batch processing can then be scheduled accordingly.

I hypothesise that using the campaign settings for day parting rather than an equivalent script will give slightly better results (“better” in this case means “closer to what you intended to happen”) with the magnitude of the improvement depending on how often the batch job is run; once per hour = big difference; once per second = not much difference at all.

Scripts and the API allow far more flexible tactics than the simple day parting settings. This must be considered when deciding which option to use.