+2 votes
The answer makes complete sense, was wondering how people typically figure out multi touch point attribution (or portfolio mix) in the case where you are running multiple concurrent offline campaigns (radio + TV + billboards + mailers) at the same time.
by (400 points)

2 Answers

+1 vote
Best answer

I think the only valid answer to this question is that you need a Media Mix Model that somehow imputes value to each of those channel sets with some reliable level of (measured) accuracy. But these kinds of models are probabilistic and the assumption should always be that they carry some level of error.

In general, I think marketing teams beat themselves up unnecessarily over trying to attribute revenue / conversions to specific channels in complex marketing systems. If it is necessity for your team that you be able to attribute revenue down to the cent to the source channel, and it's also a necessity that you run advertising campaigns in multiple media formats, some of which are offline / non-digital, then you're just setting yourself up for failure. You can't do that -- it's impossible. Even if you isolate each of those channels and run them exclusively and capture performance that way, when you mix them all up into an ensemble, there's no way you can predict the overlap or the amplification effect. What happens when TV makes your digital more effective? What happens when your paid campaigns start cannibalizing organic?

I think a better approach than wasting lots of money on trying to establish individual baselines for each advertising format you eventually want to run in a broader mix is to test budget levels within the mix and track how changes in budget impact overall effectiveness. But no one jumps into a situation like that: even when companies have raised huge rounds of funding and are under pressure to grow quickly, it's impractical for them to want to 10x or 100x marketing spend immediately with brand new channels without ramping them up.

Questions like this sometimes assume that the marketing team has no prior knowledge of the overall mix: are you really spinning up radio + TV + mailers all at once? More than likely, you are layering those channels in one at a time and measuring the incremental impact as spend grows. And when that's the case, it's the change (in performance) that matters, not the absolute value of spend at any given point in time.

by (4.2k points)
selected by
Thank you this is exactly the framework I was looking for!
Btw Eric, how do I add more tags to the question after I've posted it? I realized now the original question doesn't have any tags.
+1 vote

Hey LiveInk, 

Offline attribution is tough. So much so that some of the best performance marketers I know won't touch offline channels - even when there are clear eamples of DR teams (Credit Karma, Smile Direct, Peloton) making them work. 

There are two common approaches I've observed or heard repeated over and over again:

1) Spot Measurement - This method plots on-air time (works better for TV, radio) against conversions in a time series and measures the lift. Ie, what did conversions look like 5-10 minutes after the spot ran vs the 5-10 minutes before and what was the subsequent lift. It's not perfect, and gets tougher for multi-touch and/or high consideration funnels, but it's one of a few helpful data points. 

2) HDYHAU - This method surveys customers after they purchased with a quick "How did you hear about us?" form. It then compares the % breakdown of the survey (which includes offline channels as an option) to the direct attribution (click-based) and uses the diff to assign credit. 

Your question may have been rooted specifically in the scenario where multiple offline channels are running at once, which certainly adds to the complexity. Hopefully the above serves as a helpful look at how other teams may be looking at the problem.

Offline can be daunting for DR teams, but that certainly provides an advantage to the teams that can make it work and an argument for taking a well-guided shot.



by (190 points)
This was great as well, gives amazing context for HDYHAU