Last Thursday was the first Search Firestarters event, hosted by Google and Only Dead Fish – and we were lucky enough to get an invitation. Here are our top takeaways!
Martin McNulty, from Forward3D
- Don’t try to attribute your way out of a rubbish marketing campaign. If your marketing is rubbish no attribution model will fix that.
- Modelling is not measuring – it’s about statistics and odds.
- Concentrate on what you can control – it’s all very well finding out the volume of search depends on cloud cover, but you can’t control the weather so you can’t do anything about it.
- Some companies will get most conversions from just one click – there’s not much point doing attribution modelling if it’s only for a couple of conversions.
Helen Southgate, from BSkyB
- Attribution has flaws:
- People buy and research on many devices, and you can’t track them between these – this is only going to get worse in the future
- Tracking is based on cookies, and assumes one person only has one cookie.
- There are several objectives:
- Want to understand advertising and justify spend.
- Want to prove worth of the brand.
- Want to understand customer journey and message differently depending on what stage the customer is in.
- At Sky they’re using a staged approach:
- Getting all the data sorted (so everything can be meaningfully compared)
- Working out the objectives of the stakeholders
- Get top line information – what the channels are doing
- Look at different media partners, keywords – deeper level stuff
- Test different combinations of different channels, and different ads at different stages
David Richards, from ZenithOptimedia
- Attribution does provide a wealth of insight.
- Generics can influence people and push them to the brand
- Look at different behaviour of people from different sources.
- But it isn’t a silver bullet.
- You should only do it if it makes sense for your business.
- There are lots of things you can fix before it’s worthwhile doing attribution modelling (such as getting the best website possible)
- We’re missing so many things, such as TV and radio advertising, or the effect of competitors’ advertising pushing generic searches.
Tom Cull, from Carat Manchester
- There are two levels to attribution modelling
- The macro scale (with econometrics) – for looking at the channel mix
- The micro scale (with clicks and conversions) – for optimising at a finer level
- There’s no single standard model. Tom starts with a standard model and customises for each client as they get more data.
- Beware the “so what?” trap: some clients ask for attribution data but then don’t do anything with them. Take action based on your data.
- Eliminate non-converters
- Change the channel mix
- Build an engagement model, based on on-site activity, and feed that into the attribution model
- Process has to be collaborative – you have to involve all the right client-side data holders.
- Don’t rely solely on attribution modelling – the tools are getting better but the data are getting worse. You could optimise yourself out of important keywords your model can’t assign the correct value to. Accept the limitations of the data.
From the Q&A
How do you choose the right model? Martin starts with big chunks of data and working out the relationships between them, and doesn’t get more complicated. He suggests looking at Google Analytics with different attribution models and seeing if there’s a difference between them.
How do you line up everyone in a large organisation? Helen said that it was a challenge, but she got buy-in by having an unbiased and holistic view of attribution, and sold it as it could drive a lot more high quality sales.
Doesn’t last click work best for many real world situations? “Yes”, said Martin. Helen said that it was still worth looking at last click attribution, but using the more complicated models could help understand and optimise.
To round things up, the panel were asked for a single recommendation:
- Martin – Keep it simple.
- Helen – Know your objectives.
- David – Work with the things you can change today.
- Tom – Don’t fall into the “so what?” trap. Don’t do attribution unless you’re prepared to do things on the back of it.